Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Communicating organizational knowledge in a sociomaterial network: the influences of communication load, legitimacy, and credibility on health care best-practice communication
(USC Thesis Other)
Communicating organizational knowledge in a sociomaterial network: the influences of communication load, legitimacy, and credibility on health care best-practice communication
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
COMMUNICATING ORGANIZATIONAL KNOWLEDGE
IN A SOCIOMATERIAL NETWORK:
THE INFLUENCES OF COMMUNICATION LOAD,
LEGITIMACY, AND CREDIBILITY ON
HEALTH CARE BEST-PRACTICE COMMUNICATION
by
Amanda M. Beacom
________________________________________________________________________
A dissertation presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In partial fulfillment of the requirements for the degree
DOCTOR OF PHILOSOPHY
(COMMUNICATION)
May 2016
Copyright 2016 Amanda M. Beacom
ii
ACKNOWLEDGMENTS
Numerous people contributed to this research project and to my broader
endeavors as a doctoral student. For their generous contributions to data collection and
analysis, I am thankful to the anonymous survey participants at Alpha Medical Group, as
well as Les Schwab, Karen DaSilva, Nancy Kampf, and Steven Lampert. Ryan Brooks of
the USC University Park IRB and Nelson Walker of Qualtrics were also very helpful in
the research design and data collection phases of the project.
I want to thank Patti Riley, Peter Monge, and Tom Valente for their guidance and
support throughout the doctoral program and as members of my dissertation committee. I
also want to thank the Annenberg School for Communication for the opportunity to earn
a doctorate in and contribute to a field of scientific inquiry, something my younger self
would have never imagined. At Annenberg, my thanks in particular to Anne Marie
Campion, Tom Goodnight, and Imre Meszaros.
I am indebted to many people for their friendship and encouragement over the
past eight years of doctoral student life, including Daniela Baroffio, Lauren Frank, Mike
Kerkman, Janessa Mount, Sandee Newman, Evelyn Novello, Kevin Shiramizu, and Julie
Watkins. I am particularly thankful for my dear friends and colleagues Nancy Nien-Tsu
Chen, Young Ji Kim, Zhan Li, and Katya Ognyanova, each of whom has provided me
with inspiration simply by virtue of their own example, and whose friendship collectively
has been a Band-Aid for all of those slings and arrows of academic research and
professional life.
Most of all, I offer my gratitude to my family, always on my team, in good times
and bad: Bianca, my goodwill ambassador; Rufus, best RA ever; and Joe Kimura, MVP.
iii
TABLE OF CONTENTS
ACKNOWLEDGMENTS .................................................................................................. ii
LIST OF TABLES ............................................................................................................. vi
LIST OF FIGURES .......................................................................................................... vii
ABSTRACT ..................................................................................................................... viii
INTRODUCTION .............................................................................................................. 1
CHAPTER 1: SOCIOMATERIALITY AS AN ONTOLOGY FOR SOCIAL
NETWORK RESEARCH ................................................................................................... 4
Traditional Definitions of Social in Network Research .......................................... 5
Foundations and Concepts in Sociomateriality....................................................... 9
Materiality ................................................................................................. 11
Material agency ......................................................................................... 12
Human agency .......................................................................................... 17
Sociomaterial agency ................................................................................ 19
Network..................................................................................................... 24
Applications of Sociomateriality in Organization Science and Communication . 27
Sociomateriality and organization science ................................................ 27
Sociomateriality and organizational communication ................................ 30
Applying Sociomateriality to Social Networks .................................................... 32
CHAPTER 2: EMPIRICAL CONTEXT: COMMUNICATING HEALTH CARE
BEST-PRACTICE KNOWLEDGE IN ORGANIZATIONAL NETWORKS ................ 37
Communicating Knowledge in Organizations ...................................................... 38
Knowledge typologies and epistemologies ............................................... 39
Knowledge behavior ................................................................................. 47
Communicating Knowledge in Organizational Networks .................................... 49
Social knowledge networks ...................................................................... 49
Sociomaterial knowledge networks .......................................................... 59
Communicating Knowledge in Health Contexts .................................................. 64
CHAPTER 3: ORGANIZATIONAL LEARNING THEORY AND THE
SOCIOMATERIALITY OF COMMUNICATION LOAD ............................................. 70
Organizational Learning Theory and Communication Load ................................ 74
Communication Load as a Form of Sociomaterial Agency .................................. 79
The Role of Communication Load in Sociomaterial Knowledge Networks ........ 82
iv
CHAPTER 4: INSTITUTIONAL THEORY AND THE SOCIOMATERIALITY OF
LEGITIMACY AND CREDIBILITY .............................................................................. 98
Institutional Theory and Legitimacy ..................................................................... 98
Credibility ........................................................................................................... 109
Legitimacy and Credibility as Forms of Sociomaterial Agency ......................... 114
The Roles of Legitimacy and Credibility in Sociomaterial Knowledge
Networks ............................................................................................................. 119
CHAPTER 5: RESEARCH DESIGN ............................................................................. 130
Study Population ................................................................................................. 130
Best-Practice Knowledge .................................................................................... 133
Data Collection ................................................................................................... 138
Measures ............................................................................................................. 140
Network structure.................................................................................... 140
Legitimacy .............................................................................................. 143
Credibility ............................................................................................... 148
Communication load ............................................................................... 153
Qualitative data ....................................................................................... 155
Mass media status ................................................................................... 156
Demographic characteristics ................................................................... 156
Treatment of Missing Data ................................................................................. 160
Preliminary Analyses .......................................................................................... 162
Analysis of non-response bias ................................................................ 162
Scale reliability and confirmatory factor analysis .................................. 162
Network Analysis................................................................................................ 164
The exponential random graph model .................................................... 165
Multilevel exponential random graph modeling ..................................... 168
General approach to model specification ................................................ 172
Parameters for network structure ............................................................ 175
Parameters for demographic characteristics ........................................... 178
Parameters for research hypotheses ........................................................ 187
CHAPTER 6: RESULTS ................................................................................................ 194
Descriptive Statistics for Study Population ........................................................ 194
Preliminary Analyses .......................................................................................... 196
Analysis of non-response bias ................................................................ 196
v
Scale reliability and confirmatory factor analysis .................................. 197
Descriptive Statistics for Actor Attributes .......................................................... 203
Network Analysis................................................................................................ 207
Unipartite physician-physician network ................................................. 207
Bipartite physician-artifact network ....................................................... 220
Multilevel network .................................................................................. 225
Research Question 1 and Hypothesis 9 ................................................... 231
Qualitative Results .............................................................................................. 237
CHAPTER 7: DISCUSSION AND CONCLUSION ..................................................... 238
Interpretation of Results ...................................................................................... 238
Network structure.................................................................................... 238
Demographic characteristics ................................................................... 241
Research hypotheses ............................................................................... 242
Limitations .......................................................................................................... 249
Contributions and Directions for Future Research ............................................. 252
Conclusion .......................................................................................................... 257
REFERENCES ............................................................................................................... 259
APPENDIX A: SURVEY INSTRUMENT .................................................................... 293
APPENDIX B: CONFIRMATORY FACTOR ANALYSIS OF
COMMUNICATION LOAD.......................................................................................... 304
APPENDIX C: CONFIRMATORY FACTOR ANALYSIS OF LEGITIMACY ......... 306
APPENDIX D: CONFIRMATORY FACTOR ANALYSIS OF CREDIBILITY ......... 308
APPENDIX E: KNOWLEDGE SOURCE ARTIFACTS USED BY PHYSICIANS.... 310
APPENDIX F: POORLY-FITTING GOODNESS OF FIT STATISTICS FOR
NETWORK MODELS ................................................................................................... 312
vi
LIST OF TABLES
Table 1. Research Hypotheses and Corresponding Model Parameters .......................... 126
Table 2. Model Parameters for Demographic Characteristics ........................................ 180
Table 3. Descriptive Statistics for Demographic Characteristics ................................... 195
Table 4. Descriptive Statistics for Actor Attributes ........................................................ 205
Table 5. Descriptive Statistics for Network Structure .................................................... 208
Table 6. ERGM Results for Unipartite Physician-Physician Network ........................... 212
Table 7. ERGM Results for Bipartite Physician-Artifact Network ................................ 221
Table 8. ERGM Results for Multilevel Network ............................................................ 226
Table 9. Comparison of Results for Version 1 Models Without Physician Credibility
Parameter ........................................................................................................................ 235
Table 10. Comparison of Results for Version 2 Models With Physician Credibility
Parameter ........................................................................................................................ 236
vii
LIST OF FIGURES
Figure 1. Generic multilevel network model in MPNet. ................................................ 171
Figure 2. Multilevel model of the sociomaterial knowledge network. ........................... 172
Figure 3. Visualization of the unipartite physician-physician network. ......................... 209
Figure 4. Visualization of the bipartite physician-artifact network. ............................... 219
Figure 5. Visualization of the multilevel network. ......................................................... 224
viii
ABSTRACT
This study investigates how best-practice knowledge is communicated in an
organizational network. Using sociomateriality as an ontological foundation, I
conceptualized knowledge communication as occurring in an organizational network in
which the actors are both people and artifacts. I drew on propositions from organizational
learning theory and institutional theory to investigate how three forms of sociomaterial
agency—the communication load of knowledge consumers, and the legitimacy and
credibility of knowledge sources—influenced the structure of the knowledge network. I
tested these hypothesized influences using data on the communication of new cholesterol
treatment guidelines within an organizational network of primary care physicians and
knowledge artifacts. Results from the estimation of recently-developed exponential
random graph models for multilevel networks indicated that the legitimacy of the
organizational authors of best-practice knowledge exerted a significant influence on
network structure, whereas communication load had no effect. The results also offered
qualified support for the premise that in many empirical contexts, multilevel
sociomaterial networks hold the potential to explain more variance in the structure of
knowledge communication than unipartite or bipartite networks alone.
1
INTRODUCTION
This dissertation examines how knowledge is communicated in organizations.
More specifically, it investigates how best-practice knowledge is communicated in an
organizational network. Using sociomateriality as an ontological foundation, I
conceptualized knowledge communication as occurring in an organizational network in
which the actors are both people and things: human seekers and sources of knowledge,
and nonhuman sources of knowledge such as Web sites, reports, and databases. I drew
upon propositions from two theoretical perspectives, organizational learning theory and
institutional theory, to investigate how three forms of sociomaterial agency—the
communication load of knowledge consumers, and the legitimacy and credibility of
knowledge sources—affect the structure of communication in an organizational
knowledge network.
Theoretically, the motivation for this research was to investigate the structure of
communication ties in an organizational knowledge network with both human and
nonhuman actors. There have been hundreds of insightful studies of organizational
knowledge networks, but few that consider multiple types of actors, human or otherwise,
and/or multiple types of ties. As Butts (2009) notes, a network can be conceptualized in
many ways, and the conceptual choices we make in defining network boundaries, nodes,
ties, and levels of analysis determine the results we observe. By applying different and
novel conceptualizations of networks and the actors within them, we may gain a better
understanding of our phenomenon of interest.
Empirically, this research was motivated by the observation that in health care, as
in many other fields, there is great emphasis on and supply of best-practice knowledge,
2
and at the same time, much concern over the challenges of communicating such
knowledge. Government agencies, professional associations, and health care provider
organizations produce hundreds of best-practice guidelines, recommendations, and
metrics about health promotion and disease prevention and management—childhood
vaccination schedules, screenings for cancer, management of chronic diseases such as
diabetes and hypertension, appropriate multidrug regimens for psychiatric disorders and
HIV infection. Members of health care provider organizations may access this best-
practice knowledge from numerous, diverse sources: colleagues, experts at conferences,
journals and reports, Web sites, traditional media, social media, electronic databases, and
so on. Such best-practice knowledge has been associated with a number of positive
outcomes, including reduction of medical errors and health care costs; reduction of
variations in clinical practice and in geographic, racial, and other health disparities; and
improvement in quality of health care (Field & Lohr, 1990; Institute of Medicine, 2011;
Shaneyfelt, Mayo-Smith, & Rothwangl, 1999). These positive outcomes are only
reached, however, when health care professionals actually access, communicate about,
learn, and apply a particular “best” practice to their own everyday work practices.
Evidence suggests that such communication, learning, and application occurs slowly and
sometimes not at all. For example, scholars estimate that diffusion of health care best-
practice knowledge requires an average of 17 years from the time of initial research
publication to widespread clinical practice (Balas & Boren, 2000), and that across best
practices for a wide range of medical conditions, adoption and adherence is achieved for,
on average, only about half of physicians in the target audience (Timmermans & Mauck,
2005).
3
These two motivations informed the overarching research question investigated in
this study: the question of how the characteristics of both people and knowledge artifacts
affect the structure of best-practice knowledge communication in a health care
organizational network.
This dissertation is organized as follows. Chapter 1 proposes sociomateriality as
an ontology for social network research, and for this research project. Chapter 2 discusses
aspects of the empirical context of the research: the communication of health care best-
practice knowledge in an organizational network. Chapters 3 and 4 examine how two
theories, organizational learning theory and institutional theory, may help us understand
knowledge communication in organizational networks of people and material artifacts.
Chapter 5 explains the research design and Chapter 6 presents the results. Chapter 7
offers discussion and conclusions.
4
CHAPTER 1: SOCIOMATERIALITY AS AN ONTOLOGY FOR SOCIAL
NETWORK RESEARCH
This study used sociomateriality as an ontology for conceptualizing networks of
heterogeneous actors and the relationships between them,
1
following the proposals and
work of Contractor, Monge, and Leonardi (Contractor, 2009; Contractor, Monge, &
Leonardi, 2011). Sociomateriality is an ontological perspective that proposes that both
human actors and nonhuman actors, such as material artifacts and technologies, mutually
contribute to social structures and social phenomena. The social is in part material, and
the material is in part social. This “entanglement” (Orlikowski, 2007) of the social and
the material implies that when we examine a social network, we should also be able to
observe the presence of the material.
The application of sociomateriality to social network research provides an
ontological foundation for conceptualizing networks composed of both human and
nonhuman actors and the relationships between them. In this study, I used this foundation
to conceptualize a network of people and material artifacts in an organization, and to
investigate how knowledge is communicated within this network. With its recognition of
nonhuman, or material, agency, as well as human agency, sociomateriality offers the
ontological basis to consider the transfer, or “communication,” of knowledge that occurs
between a material artifact and a human actor. In the remainder of this chapter, I review
how human and nonhuman actors traditionally have been conceptualized in social
networks, and outline concepts and foundations in sociomateriality relevant to its
application to social network research. I then discuss how sociomateriality previously has
1
The title of this chapter is adapted from the title of an article with a similar objective (Banks & Riley,
1993), albeit regarding a different ontological perspective and field of research.
5
been used in two fields, organizational communication and organization science, and
explain how it was employed in this dissertation.
Traditional Definitions of Social in Network Research
In network research, the basic units of analysis are the relationships (also referred
to as ties, links, or edges) that exist between two or more actors (or nodes or vertices).
These relationships between actors form a network with a particular structure, which in
turn may produce a particular social system or phenomenon. Networks may be classified
in many ways, according to various characteristics of the actors and relations of which
they are composed. One way of classifying a network is according to whether the nodes
are ontologically homogeneous (for example, human nodes only), or heterogeneous (for
example, human and nonhuman nodes).
In the social sciences, the networks of interest are usually ontologically
homogeneous: the nodes are all people or a collective of people, such as a group or
organization, and the ties are social relationships. Typically, these networks have not
included material artifacts or other nonhuman nodes. The 2012 volume of the journal
Social Networks, for example, included 64 articles, all of which featured analyses of
networks containing human nodes only. Likewise, most studies of knowledge networks
in and among organizations define nodes as either members of organizations (as in
intraorganizational networks), or as collectives of these members, that is, the entire
organization is one node (as in interorganizational networks). The ties in knowledge
networks reflect knowledge-related social interactions between these organizational
members or organizations, such as advice-sharing, information-seeking, and consulting.
6
Outside of the social sciences, the networks of interest are also usually
ontologically homogeneous, containing nonhuman nodes only, and with relations that are
not considered social. (The animal social networks studied by biologists and ecologists
feature nonhuman actors engaged in social relations and are an exception; see Faust,
2011.) Network analysis has been employed in physics, mathematics, computer science,
biology, and other fields outside of the social sciences, in order to study networks with
many different sorts of nonhuman nodes (Brandes, Robins, McCranie, & Wasserman,
2013; Butts, 2009; Newman, 2010). Some well-studied nonhuman networks include
hyperlink networks, in which the nodes are web pages and the ties are the hyperlinks
connecting these pages; transportation networks, in which the nodes are geographic
locations and the ties are routes between the locations; and citation networks, in which
the nodes are academic journal articles and the ties are the citations these articles make to
one another (Newman, 2010). Some scholars use the term “network science” to broadly
refer to the study of networks consisting of any types of nodes that share interdependent,
but not necessarily social, relationships, in order to distinguish these lines of network
research from the more narrow scope of “social network analysis” (Brandes, et al., 2013).
An accounting of the categories in network research that stops here, with the
traditional disciplinary definitions and boundaries—networks of humans and social
relations in the social sciences, and networks of nonhumans and non-social relations
outside of the social sciences—ignores an ontological complexity that has also been
present in network scholarship. Some scholars have recognized that traditional
distinctions between social and non-social are fuzzier than often acknowledged, and that
there are multiple ways to conceptualize and define nodes, ties, and networks. Such
7
questions are highlighted when the ontology of a particular set of nodes is debatable. For
example, some might consider the citation networks mentioned above as homogeneously
nonhuman, and therefore not “social” networks, because the nodes are academic journal
articles and the ties are the citations these articles make to one another. Others, however,
might argue that citation networks are homogeneously human and social, because the
journal articles are written by people, and the citations in the articles are made by people.
As Newman (2010) notes, in discussing the distinctions between “information” networks
and “social” networks:
There are some networks which could be considered information networks but
which also have social aspects to them. Examples include networks of email
communications, networks on social-networking websites such as Facebook or
LinkedIn, and networks of weblogs and online journals. . . . The classification of
networks as social networks, information networks, and so forth is a fuzzy one,
and there are plenty of examples that, like these, straddle the boundaries. (p. 63)
The question of appropriate network definition also comes to the fore when one
considers conceptualizing networks with actors of heterogeneous ontologies, in which the
actors are both human and nonhuman. Returning to the previously-mentioned 2012
volume of Social Networks, one can imagine how the authors of several articles in this
volume could have conceptualized their networks of interest as having heterogeneous
rather than homogeneous ontologies, had they considered the nodes and ties in the
networks differently. For example, in a study of adolescents’ conversation networks,
Friemel (2012) examines whether television use acts as an independent variable
influencing the selection of conversation partners (i.e., social selection), or as a dependent
8
variable being influenced by the network structure of social conversations (social
influence). To address this question, Friemel conceptualized longitudinal networks in
which adolescents represented the nodes and conversation exchange represented the ties
between the adolescents. He found support for the social selection mechanism (intensity
and type of TV use influences the structure of the conversation network) but not for
social influence (the structure of the conversation network influences TV use). In
Friemel’s study, TV use was conceptualized as an attribute of the adolescent actors in the
network, but one could also imagine the network being conceptualized differently: with
two types of nodes, adolescents and the TV programs they watch, and two types of ties,
conversation exchange ties between adolescent nodes and viewership ties between an
adolescent node and a TV program node.
Similar network reconceptualizations could be considered in several other articles
published in the 2012 volume of Social Networks. These include a study of social and
geographic influences on adolescent substance use (Mennis & Mason, 2012), in which
specific neighborhood locations (e.g., parks) could have been brought into networks of
adolescents’ friends and family, as a second type of node; and a study of the impact of
social capital on health information seeking (Song & Chang, 2012), in which nonhuman
sources of health information could have been brought into networks of subjects’
discussion partners.
Such a network reconceptualization is also possible in the case of the knowledge
networks that exist within and among organizations. As noted above, organizational
knowledge networks are typically conceptualized with the nodes as organizations or their
(human) members, and the ties as some form of knowledge exchange between people.
9
However, an organizational knowledge network with human nodes only could be
reconsidered such that nonhuman sources of knowledge could be brought into the
network as a second type of node, and sociomateriality could be used as the ontology to
support such a move (Contractor, 2009; Contractor, et al., 2011).
Foundations and Concepts in Sociomateriality
A thorough review of the breadth and varied applications of sociomateriality is
beyond the scope of this inquiry, and thus the focus of this chapter is narrowed to a
handful of concepts relevant to using sociomateriality as an ontology for network
research. These include materiality, material agency, human agency, sociomaterial
agency, and the sociomaterial conceptualization of “network.” The following review of
these concepts draws particularly on actor-network theory, which is one of a handful of
theoretical perspectives commonly employed by scholars investigating sociomateriality,
socio-technical systems, and materiality and society (Contractor, et al., 2011; Orlikowski
& Scott, 2009). Other perspectives include agential realism (Barad, 2003, 2007;
Leonardi, 2013b), activity theory (Groleau & Demers, 2012; Kaptelinin & Nardi, 2006;
Vygotsky, 1978), and practice theory or practice studies (Feldman & Orlikowski, 2011;
Gherardi, 2006; Schatzki, Knorr Cetina, & von Savigny, 2001). It is important to note
here that throughout this dissertation, and particularly in this chapter, I refer to
“sociomaterialists” and “sociomateriality” as if the latter is an established scientific
subfield about which general consensus exists and to which a sizeable group of scientists
identify themselves as belonging. This is not yet the case. Sociomateriality is a subfield in
its infancy and its definition and even the term “sociomateriality” itself are still subjects
of intense debate.
10
Actor-network theory was developed by sociologists working in the field of
science, technology, and society (Latour, 2005; Law, 1992).
2
It is particularly relevant for
this project for two reasons. First, it offered a basis for the conceptualizations of material,
human, and sociomaterial agencies, described in more detail below, that provided the
critical ontological scaffolding for this dissertation. Second, actor-network theory
conceptualizes the relationships between humans and material artifacts as constituting a
network. One of its primary claims is that society is produced by heterogeneous networks
of human and nonhuman actors. Each human and nonhuman actor possesses agency, and
collectively, the relationships between the actors constitute networks, which in turn
produce particular social phenomena. “The task of sociology,” according to actor-
network theory, “is to characterize these networks in their heterogeneity, and explore how
it is that they come to be patterned to generate effects like organizations, inequality and
power” (Law, 1992, p. 381). To many social network scholars, this “task” may sound
quite similar to their own research on how the connections between actors in networks
produce particular social structures and phenomena, such as the diffusion of new ideas
and the consolidation of political power. Actor-network theory’s strong argument for the
role of material, or nonhuman, agency in the production of our social world—and the
common ground it shares with network science, in its insistence of the primacy of
relationships—made it salient to the present project of examining communication in an
organizational network of human and material knowledge sources.
2
For more comprehensive reviews of actor-network theory, see Latour (2005) and Law (2009).
11
Materiality
The “materiality” of sociomateriality refers to the world of human-made,
inorganic objects that surround us, either physically or virtually—a chair, book, mobile
phone, the application software on the mobile phone, the Internet, and so forth.
Sociomaterialists have analyzed how specific material properties or features of an artifact
contribute to or make possible various social phenomena (Leonardi, 2012), in the same
way that the specific skills, personality traits, and predilections of a person may
contribute to or make possible various social phenomena. The unique “affordances” and
“constraints” of these properties of artifacts—the particular actions or structures that they
enable or constrain (Faraj & Azad, 2012; Leonardi, 2011; Siles & Boczkowski, 2012)—
are often illustrated by comparing how two different artifacts contribute differently to
similar social phenomena. For example, the material properties of the Internet enable and
constrain knowledge retrieval in different ways than the material properties of a book
enable and constrain knowledge retrieval.
Some artifacts lack “matter,” at least according to our conventional understanding
of matter as having physicality, as being something we can touch. Many digital
technologies and artifacts (e.g., computer software, the Internet) lack the physical
presence or embodiment of a calculator or car, or are composed of both physical and
virtual components (e.g., video games with a physical console and online competitors),
complicating, as many scholars have pointed out, what we mean by “materiality.”
Further, an artifact’s features, physical or non-physical, may have variable salience and
presence to different users. For example, one user of Microsoft Word software may use
its “Track Changes” editing tool on a daily basis, whereas another user may never use it,
12
and may not even be aware of its existence. Synthesizing the ways in which a number of
sociomaterialists have used the term and struggled with the trickiness of its application,
Leonardi (2012) defines materiality as “the arrangement of an artifact’s physical and/or
digital materials into particular forms that endure across differences in place and time and
are important to users” (p. 31). This definition allows materiality to encompass both
physical and non-physical artifacts, and attempts to recognize both the stability of
materiality and the way in which the materiality of an artifact is contingent on how other
actors perceive and use it (Leonardi, 2012, 2013a).
In this dissertation, the materiality of interest consisted of the artifacts that people
use as sources of knowledge in organizational settings. Law (1992) explains the
materiality of knowledge in the following way: “I put ‘knowledge’ in inverted commas
because it always takes material forms. It comes as talk, or conference presentations. Or
it appears in papers, preprints or patents” (p. 381). These nonhuman, material sources of
knowledge, together with human sources of knowledge—colleagues, friends, experts
speaking on a panel—made up the sociomaterial knowledge networks that were the focus
of this project.
Material agency
Latour (2005) defines agency as the ability to “modify a state of affairs by making
a difference” (p. 71) in another actor’s actions. According to Latour (2005), to say that a
nonhuman actor has agency means that the actor has the ability to translate or
transform—rather than simply serving as an intermediary that “transports meaning or
force without transformation” (pp. 38-39). For example, to say that the Google search
engine has agency, one recognizes that instead of serving simply as a neutral retrieval
13
system for information and knowledge, the particular search algorithm employed by
Google acts as a gatekeeper, retrieving and ranking some knowledge at the expense of
other knowledge (Cheney-Lippold, 2011; Nahon, 2011), in a way that may be different
from the knowledge retrieval service offered by a different search engine, such as Bing or
Yahoo, or from the knowledge retrieval service offered by a reference librarian.
Callon (2004) defines nonhuman agency in this way: “Like humans, non-humans
and especially technologies participate in their own right in the definition and course of
action, and in the production of knowledge” (p. 4). As Leonardi (in Aakhus et al., 2011)
explains, we experience information and communication technologies (ICTs) like mobile
phones and online communities as actors that “do things” and that affect what we, in turn,
do. A number of sociomaterial scholars argue that material agency differs from human
agency, however, in that a material artifact lacks its own inherent intentionality and goals
(Leonardi, 2012). At the same time, sociomaterialists recognize that the human engineers
and designers who create material artifacts always, whether mindfully or not, “imprint”
or “inscribe” these artifacts with intentionality and goals (D'Adderio, 2011; Latour,
1992); with what the engineers and designers hope the artifacts will “do” or
“accomplish.” Here we observe that “entanglement” of the social with the material
(Orlikowski, 2007), the seemingly inseparable nature of material and human agencies.
So exactly what do material artifacts “do”? Sociomaterial scholars have addressed
this question in several related ways, both general and specific. First, and most broadly,
sociomaterialists propose that artifacts “actively participate in the production of the
social” (Callon, 2004, p. 6). By this, they mean that nonhumans participate with humans
in creating our social world: through their relationships with one another, nonhumans and
14
humans form a heterogeneous network, the effect of which is a particular social structure
or phenomenon. Leonardi (2011) and others (e.g., Taylor, 2001; Taylor & Van Every,
2000) call the interdependent participation of humans and nonhumans in creating social
structure “imbrication,” employing a term used in ancient Roman and Greek architecture
to describe how two differently-shaped tiles were laid in an interlocking pattern to build a
waterproof roof. Like the two differently-shaped tiles that together compose the roof,
human agency and material agency together contribute to a social structure or
phenomenon. Callon and Muniesa (2005) describe the relations between humans and
nonhumans as a “process of co-elaboration,” arguing that “instead of postulating a break
between human agents and the things that they conceive, produce, exchange and
consume, it is possible to highlight the growing importance of the processes of mutual
adjustment between things and human beings” (p. 1233). Other sociomaterialists use the
terms “mutually constitute” and “co-construct” to express a similar idea (Kallinikos,
Leonardi, & Nardi, 2012, p. 3).
Second, sociomaterialists propose that the various features of material artifacts
offer, according to people’s perceptions, either affordances for or constraints to human
agency and behavior (e.g., Faraj & Azad, 2012; Leonardi, 2011; Siles & Boczkowski,
2012). Scholars have proposed a number of hypotheses about how people interact with
these material affordances and constraints—that is, how human agency interacts with
material agency. Leonardi (2011), for example, proposed that people’s perceptions of
material constraints lead them to change the artifacts or technologies that they use,
whereas perceptions of material affordances lead people to change their own behaviors.
DeSanctis and Poole (1994) described an alternative reaction to material affordances and
15
constraints, called “unfaithful appropriation,” (p. 130) where people use artifacts in ways
that are novel or unintended by the artifacts’ original designers.
Within these general descriptions of what nonhumans “do”—that is, co-
constructing social phenomena and providing affordances for or constraints to human
agency—scholars have elaborated a number of more specific actions that material
artifacts may accomplish in various contexts. For example, they facilitate and mediate
interaction between people, including people otherwise unknown to one another, as when
a book mediates the communication of ideas between the book’s author and its reader
(Callon, 2004; Law, 1992). Without the material artifact, the book, there would be no
relationship between the two people. Artifacts help structure and define interaction
between people (Law, 1992), as when PowerPoint presentation software structures how
speakers in many professional fields present their messages to an audience. Artifacts also
signal and represent status, credibility, and legitimacy in various communities (Bechky,
2003), as when professionals’ mastery of the latest communication technology at work
may signal to colleagues their competence and currency, or corporations’ use of recycled
paper in their products may signal to consumers their progressiveness and dedication to
environmental responsibility.
Latour (2005, pp. 80-82) notes that nonhuman agency, and the relationships
between human and nonhuman actors, are particularly visible in and relevant to several
empirical contexts. One of these contexts involves situations of disorder, when nonhuman
actors do not behave as expected, such as in accidents and breakdowns. Another such
context is that of innovation and novelty, such as in product development, when we are
less likely to take artifacts and what they “do” for granted. Similarly, Leonardi and
16
Barley (2010) note that the implementation of new technology in an organization
represents a particularly opportune context for examining sociomaterial processes:
“implementation marks a time when an existing sociomaterial fabric is disturbed, offering
researchers an opportunity to ‘see’ more clearly how the social and material become
constitutively entangled” (p. 34).
In the present project, the empirical context was also one of innovation or
implementation: I examined the communication of new best-practice knowledge in an
organizational network. In this case, what was being “innovated” or “implemented” was
not a new technology, which is Leonardi and Barley’s (2010) focus; rather, it was new
knowledge, accessed from technologies but also from non-technological artifacts and
from people.
Sociomateriality’s conceptualizations of material agency have prompted strong
concerns and critiques among scholars. One main concern is anthropomorphism. That is,
if we claim that artifacts “do” things, do we run the risk of attributing human
characteristics and behaviors to artifacts? For example, in an article about the network of
relationships among fishermen, scallops, and marine biologists, Callon (1986) describes
the scallops as having agency, as “anchoring themselves” to a net and “accepting a
shelter” (p. 202). Callon (1986) addresses the concern about anthropomorphism, writing
about his descriptions of the scallops’ actions,
The reader should not impute anthropomorphism to these phrases! The reasons
for the conduct of the scallops—whether these lie in their genes, in divinely
ordained schemes or anything else—matter little! The only thing that counts is the
definition of their conduct by the various actors identified [the fisherman, marine
17
biologists, and the scallops themselves]. The scallops are deemed [by the marine
biologists] to attach themselves just as fishermen are deemed [by the biologists] to
follow their short term economic interests. They therefore act. (p. 220)
For Callon, the scallops have agency because the other actors in this sociomaterial
network—the fishermen who catch and sell the scallops, and the biologists who study the
scallops and sustainable fishing practices—perceive the scallops as having agency, as
“doing” things such as accepting shelter. Similarly, in the present study of a sociomaterial
knowledge network in a health care organization, I conceptualized that both material
artifacts and people “communicate” knowledge to physicians. One might raise the
concern that I anthropomorphized the artifacts by describing how they “communicated”
with physicians in the organization. However, in describing the relationship between a
physician and, say, a journal article, in the same way as I described the relationship
between that physician and her colleague, my intention was not to anthropomorphize the
journal article, but to echo what both the creators and readers of the journal article hoped
it would “do”: communicate knowledge.
Human agency
Human agency is, as Leonardi (2011) puts it, “the ability to form and realize one’s
goals” (pp. 147-148). For my purposes here, human agency is perhaps best explained by
examining another scholarly debate about sociomateriality’s conceptualizations of
material and human agency. In addition to anthropomorphism, a second concern
expressed about sociomateriality is technological determinism. According to
sociomateriality, and in particular, to the many researchers careful to distinguish a new
scholarly turn toward sociomateriality from old conceptions of technological
18
determinism, material agency does not in any way diminish human agency. That is, the
existence of material agency does not detract from the agency of the people who design,
create, and use the material artifact. For sociomaterialists, the distribution of agency
among humans and nonhumans is not, as Callon (2004) writes, a “zero sum game” (p. 7).
The recognition of nonhuman agency does not cancel out the agency of people or render
them passive; it distributes agency between humans and nonhumans. As Leonardi (2011)
argues, instead of viewing sociomateriality as ontologically radical in this regard, one
might alternatively see sociomateriality as staking out a moderate middle ground between
extreme interpretations of technological determinism (i.e., technology and materiality
determine social life), on the one hand, and extreme interpretations of social
constructivism or voluntarism (i.e., only people determine, or construct, social life), on
the other. (For further elaboration of this point of view, and of the broader conversation
about determinism, constructivism, and sociomateriality, see, for example, Callon &
Muniesa, 2005; Cooren, 2006; Kallinikos, et al., 2012; Latour, 2005; Leonardi, 2013a;
Leonardi & Barley, 2008; 2010.)
The American public policy debate over gun ownership, “control,” and violence
offers what has become a classic example in the sociomateriality literature regarding how
the distribution of agency can be viewed (or not) as a zero sum game (Cooren, 2004;
Latour, 1994). Gun control proponents argue that “guns kill people,” placing blame for
gun-related deaths solely on the presence of the gun, and on material agency. Gun rights
proponents counter that “guns don’t kill people, people do,” arguing that it is the person
who pulls the trigger of the gun, that human agency is solely responsible for gun
violence. Both of these arguments are zero-sum perspectives. A third group of advocates
19
in the debate about gun violence steps away from zero-sum arguments and supports
measures that address both human and material agency: background checks for
prospective gun owners and better mental health care for those at risk of committing gun
violence (human agency), as well as limits on the types of guns that can be purchased
(material agency). According to this third, “sociomaterial,” perspective, blame for gun
violence cannot be solely assigned to people only, or guns only, but instead is distributed
across both agencies. When considering the problem of gun violence, it is difficult to
separate material agency from human agency; they are entangled. As Cooren (2004)
explains, “both humans and nonhumans contribute to what is happening. The gun and the
gunman both make a difference and it is their association [emphasis added] that becomes
the origin to action” (p. 377). In other words, according to sociomateriality, it is the
relationship between, or imbrication of, material agency (guns) and human agency
(people) that produces a social phenomenon (gun violence).
In this dissertation, the primary human agency of interest was that of physicians
who sought best-practice knowledge through the communication ties they formed with
both material (e.g., a Web site) and human (e.g., another physician) sources of that
knowledge. This human agency of the physician knowledge seekers was entangled with
the material agency of the artifact sources, and with a second form of human agency
exercised by physicians who also acted as knowledge sources. Together, these actors and
their agencies formed a sociomaterial knowledge network.
Sociomaterial agency
Theoretically, then, there is material agency on the one hand and human agency
on the other. Mutch (2013) refers to this accounting of agencies as the “non-conflationary
20
approach, in which the social and the material are held apart for the purpose of exploring
their interplay” (p. 29). Empirically, however, trying to isolate material from human
agency sometimes poses practical challenges. Recall Latour’s (1994) classic example of
gun violence, in which it is difficult to attribute this social phenomenon solely to human
agency or solely to material agency, and much easier to argue that gun violence is a
sociomaterial phenomenon that should be attributed to the combination of the two
agencies.
Similarly, consider the task of categorizing the forms of agency involved in the
phenomenon of interest in the present study: the communication of best-practice
knowledge in an organizational network. There were three types of actors in the network:
physicians who sought or received best-practice knowledge, physicians who were sources
of best-practice knowledge, and material artifacts that were also sources of best-practice
knowledge. Following the definitions of agency outlined above, we might expect the
physicians to exercise human agency, and the artifacts to exercise material agency. When
a knowledge tie forms between two of these actors, however, it can become tricky to
delineate human from material agency.
Imagine that a physician is listening to the radio on her drive home from work and
happens to hear a news report about a new set of clinical guidelines for the treatment of
high cholesterol, and learns about the guidelines from this report. In listening to this radio
news report, the physician forms a knowledge “tie” with a source of best-practice
information. What agencies are involved in this knowledge communication between
physician and radio news report? Who or what is responsible for the physician learning
about the new cholesterol guidelines from the radio station? There is the human agency
21
of the physician, who turned on the radio and then listened to the report, rather than not
listening and daydreaming about another subject, or changing the station when the report
started because she did not feel like hearing about a work-related topic. Because the
physician happened to turn on the radio, and then listened to the report, one might argue
that the agency responsible for the knowledge communication is the physician’s human
agency. But this view ignores the material agency of the radio news report that contained
the knowledge. It also ignores the human agency of the radio station reporter and
producers, who authored and contributed to the production of the report; the material
agency of the original clinical guidelines document being reported on; the human agency
of the authors of the clinical guidelines document; and so on. The distinctions between
material and human agency, clear when one considers human actors and material actors
in isolation, not engaged in activity, become conceptually fuzzy when one considers
those actors interacting, with the result of their interaction being a phenomenon like the
communication of knowledge, or in Latour’s example, gun violence.
How, then, does one decide to which agencies to attribute such phenomena? How
does one determine whether these agencies are material or human? Pentland and Singh
(2012) describe this dilemma as an “agency stew,” writing:
“But in any practical situation, we always have an agency stew. It is hard to tell
where the agency is: some of it is in the carrots (people), some of it is in the
potatoes (things), and some is in the sauce (protocols, languages, etc.). You can
pick it apart, but it would not be the same dish” (p. 289).
22
Likewise, even Leonardi (2013b), who in his work on sociomateriality generally portrays
human agency and material agency as separate and distinguishable forces, makes a
similar observation:
In the context of human-created artifacts such as information technologies, the
view that materiality is not necessarily social is a bit of a quagmire. Of course, all
information technologies were created by people and are the result of social
processes. (p. 69)
For this reason, to avoid this potential “quagmire,” I chose in this project to
identify the agencies of interest, that is, those engaged in and responsible for knowledge
communication, as “hybrid” (Castor & Cooren, 2006), or sociomaterial, agencies, rather
than as either solely human or solely material. Doing so made it much simpler to think
about the characteristics of physicians, and of their human and nonhuman knowledge
sources, that may have interacted to influence the knowledge communication that
occurred between them. Specifically, in the present project, I was interested in examining
how three such characteristics—the communication load felt by physicians, and the
perceived legitimacy and credibility of the human and material knowledge sources—
affected the phenomenon of knowledge communication in an organizational network. In
the extant medical and public health literature, these three forces are repeatedly invoked
as holding critical influence over the diffusion of clinical best-practice guidelines, as
discussed in greater detail in Chapter 2.
In considering these three forces, one might first imagine that it would be easy to
associate them with a particular form of agency: communication load is felt by humans
(in this case, physicians), and thus affects, and could be considered a form of, human
23
agency. Legitimacy and credibility are characteristics of material knowledge source
artifacts, and thus affect, and could be considered forms of, material agency. And
credibility is also a characteristic of human knowledge sources, and thus affects, and
could be a considered a form of, human agency as well. Upon closer inspection, however,
using the above reasoning, one slowly becomes stuck in Leonardi’s quagmire.
Communication load, or overload, is a perception experienced by humans, yes, but
research suggests that for many of us, it is a perception frequently driven by the ongoing
influx of material knowledge artifacts in our daily work routines (Barley, Meyerson, &
Grodal, 2011). Legitimacy and credibility are characteristics of artifactual knowledge
sources, of course, but they are socially constructed characteristics, and in and of
themselves possess no material properties whatsoever. Credibility is also a characteristic
of human knowledge sources, but the credibility of a human source, say a work
colleague, may be contingent on the material knowledge sources that colleague herself
uses to learn about a new practice. The sociomateriality of communication load,
legitimacy, and credibility are elaborated in Chapters 3 and 4, but for now, my point is
this: it was difficult to support the argument that these hypothesized influences on the
communication of best-practice knowledge were solely that of human agency or solely
that of material agency. I therefore chose to conceptualize them as forms of sociomaterial
agency.
Note that this dilemma over the delineation of agencies—Leonardi’s quagmire
and Pentland and Singh’s agency stew—has been interrogated elsewhere in much more
philosophical depth than was possible in the present project or by the present author, and
there is an extensive literature that disagrees with the concept of sociomaterial or hybrid
24
agency that I adopted here. For more on this debate, see, for example, Faulkner and
Runde (2012), Kautz and Jensen (2013), Leonardi (2013b), Mutch (2013), and Scott and
Orlikowski (2013).
Network
A discussion of sociomateriality and network science would be remiss not to note
the similarity in the use of “network” as a construct in both fields. In network science,
“network” has a directly empirical meaning: it refers to “a set of actors or nodes along
with a set of ties of a specified type (such as friendship) that link them” (Borgatti &
Halgin, 2011, p. 1169). A primary task of network science is to associate particular tie
structures, and nodal positions within these structures, with particular phenomena or
outcomes at the actor, group, and network levels (Borgatti & Halgin, 2011). “Network” is
also used in a directly empirical way in sociomateriality, particularly by actor-network
theorists. In actor-network theory, a network is analyzed as an interconnected “string of
actions” (Latour, 2005, p. 128) and actors, with each actor in the network, human or
nonhuman, playing a role as a mediator of action. A primary task of actor-network theory
is to trace the network of actors and relationships associated with a set of actions, in order
to understand the broader social structure or phenomenon they collectively produce.
According to actor-network theory, actors and their actions must always be understood
through the prism of the larger network of related actors and actions to which they are
connected.
Other sociomaterialists often use “network,” and to similar effect, “web” and
“social fabric,” as descriptions or metaphors, rather than as empirical constructs. Used
this way, these terms are what Scott, in his introductory text on social network analysis,
25
calls “textile metaphors,” often accompanied by verbs such as “interweave” and
“interlock” (J. Scott, 2000, pp. 4-5). For example, Leonardi (2013a) describes material
objects like computers and desks as “artifacts that are embedded in a web of social
practice” (p. 143), and D’Adderio (2011) writes of information systems often being
“entangled into a thick web of organizational relationships” (p. 215).
In both actor-network theory and social network analysis, the primary, focal units
of analysis are the relations within the network, rather than the actors. Latour (2005)
explains, “So, an actor-network is what is made to act by a large star-shaped web of
mediators flowing in and out of it. It is made to exist by its many ties: attachments are
first, actors are second” (p. 217). As Law (2009) notes, actor network theory’s “single-
minded commitment to relationality makes it possible to explore strange and
heterogeneous links and follow surprising actors to equally surprising places” (p. 147).
Similarly, write Wasserman and Faust (1994), “from the view of social network analysis,
the social environment can be expressed as patterns or regularities in relationships among
interacting units” (p. 3)—that is, social structure and phenomena are expressed as
patterns of relations.
Network science and sociomateriality also both emphasize the dependent nature
of action within the network. In actor-network theory, Callon (1991) offers these
definitions of “actor,” “network,” and the relations between them: “The actor has a
variable geometry and is indissociable from the networks that define it and that it, along
with others, helps to define” (p. 154). In actor-network theory, then, an actor cannot be
understood without at the same time considering the actor’s sociomaterial relations with
other actors. The actor and the actor’s network are “indissociable,” as Callon (1991)
26
writes, and hence the punctuation of the term “actor-network theory,” with the hyphen
emphasizing the connection between the two constructs. Explains Latour (2005), “when
we speak of actor we should always add the large network of attachments making it act”
(pp. 217-218).
Similarly, in network analysis, the structure of the relationships within a network
influences the behavior of actors within the network and the behavior of the network as a
whole (Newman, 2010; Wasserman & Faust, 1994). The components of networks (and
the data that describe them) are interdependent; as Wasserman and Faust explain, “social
network analysis is explicitly interested in the interrelatedness of social units. . . .[it
provides] a theoretical alternative to the assumption of independent social actors” (pp.
16-17). Brandes and colleagues (2013) write that “at the heart of network science is
dependence, both between and within variables” (p. 6). By dependence, network scholars
are referring to the fact that network ties are dependent on one another; the presence of
one tie affects that of other ties, and thus the actors connected by these interdependent
relations are also interdependent. As Koskinen and Daraganova (2013) explain,
dependence means that “the likelihood of a tie may not only be a function of individual
characteristics of actors who share the tie, but also a function of presence or absence of
other network ties in the network” (p. 51).
The similarities between the use of “network” in sociomateriality and in network
science—with the fields’ shared emphasis on relationships and the dependent nature of
action—are thus both linguistic and substantive. In this dissertation, the concepts of
network in sociomateriality and network science are integrated to examine a
sociomaterial network, in which the actors were people and material artifacts, and the
27
relations represented the interdependent nature of knowledge communication between
these actors.
Applications of Sociomateriality in Organization Science and Communication
This dissertation drew particularly on two lines of research employing
sociomateriality: studies of technology in organization science and management, and
studies of the role of material artifacts in organizational communication, the latter
conducted by those identifying with the “communicative-constitution-of-organization”
(CCO) school. Both lines of research utilize many or all of the foundations and concepts
discussed above, but to slightly different effects. Scholars in organization science have
typically focused on the use of newer ICTs in organizations (Orlikowski & Scott, 2009),
and on the adoption and implementation of these technologies in organizations (Ashcraft,
Kuhn, & Cooren, 2009; Leonardi, 2012; Leonardi & Barley, 2010; Orlikowski, 2007).
Studies in the CCO school of organizational communication have more broadly
examined the use of both old and new material artifacts in organizations, in the context of
everyday practices (versus that of adoption and implementation only). The remainder of
this section briefly situates the present work in the context of these two ongoing lines of
research.
Sociomateriality and organization science
According to sociomaterialists, organizations are one example of the structures
and phenomena that are mutually constituted by the social and material worlds (Leonardi,
2011, 2013a). Within an organization, the relationships between people and material
artifacts are ubiquitously implicated in actions large and small, monumental and
mundane. As Orlikowski (2007) observes,
28
A considerable amount of materiality is entailed in every aspect of organizing,
from the visible forms—such as bodies, clothes, rooms, desks, chairs, tables,
buildings, vehicles, phones, computers, books, documents, pens, and utensils—to
the less visible flows—such as data and voice networks, water and sewage
infrastructures, electricity, and air systems. (p. 1436)
Many organizational researchers have tended to examine organizational sociomateriality
with a narrower agenda, however, primarily in the service of understanding one aspect of
organizations and their environments: use of new technology and the relationship
between technological change and organizational change (Leonardi, 2013a; Orlikowski,
2007; for exceptions, see Rafaeli & Pratt, 2006). Empirically, these studies often offer
detailed descriptions of how the material properties of newly-adopted or newly-
implemented technologies affect and are used (or not used) by people in organizations
(for reviews of this research, see Leonardi & Barley, 2010; Orlikowski & Scott, 2009).
For example, in a study of how groups of diverse professionals coordinated their work on
a large construction project, Whyte and Harty (2012) examined how the features of an
integrated software system stymied coordination, while the features of other material
artifacts provided more flexibility, facilitated collaboration, and thus became favored
tools among project members.
Theoretically, organizational scholars have proposed or deployed a number of
organizational theories as useful companions to a sociomaterial ontology. These include
work-life balance theory (Robey, Raymond, & Anderson, 2012), transactive memory
theory (Contractor, et al., 2011), organizational routine theory (D'Adderio, 2011; Gaskin,
Berente, Lyytinen, & Yoo, 2014; Leonardi, 2011; Robey, et al., 2012), resource
29
dependence theory (Leonardi, 2013a), population ecology theory (Leonardi, 2013a), and
institutional theory (Leonardi & Barley, 2010; Orlikowski & Barley, 2001). Robey and
colleagues (2012) call for the incorporation or elaboration of materiality in existing
theories of organization and information systems in order to “elevate materiality to a
more central theoretical role” (p. 219). Similarly, Leonardi (2013a) advocates for the
pairing of the sociomaterial perspective with established organizational theories to move
beyond basic description of sociomaterial entanglement and advance a deeper
understanding of “the processes of entanglement” (p. 163).
The present project drew upon the application of sociomateriality in organization
science research in at least two ways. First, following the efforts of organizational
researchers, I examined the pervasive sociomateriality of organizational life,
investigating how organizational knowledge networks are composed of knowledge
sources that are both material artifacts and other people. Second, following the examples
and calls about theory described above, I built hypotheses about the relationships within
these sociomaterial knowledge networks using two organizational theories,
organizational learning theory and institutional theory, integrating them with the
sociomaterial perspective.
One difference between the present project and some of the sociomaterial
research in organization science is also worth noting briefly here. The variable of primary
interest in a number of organizational studies is the artifact, positioned as either the
dependent variable or as the primary independent variable (where the dependent variable
may be, for example, the organizational routine, or the organizational form) (Orlikowski
& Scott, 2009). Here, the artifact was not particularly singled out; it was recognized as
30
one category of knowledge source, together with human knowledge sources. The focus
was not on the artifact per se but instead on knowledge sources, plural, both human and
nonhuman, and how they interacted in an organizational network. The implication of this
distinction is that the purpose of my study was not to understand artifacts better, per se,
but rather to understand how people and material artifacts together determine and enact a
particular social structure, in this case, a sociomaterial knowledge network in an
organization.
Sociomateriality and organizational communication
Communication is inherently sociomaterial, given the vast range of material
artifacts—microphones, Microsoft Word software, mobile phones, and so forth—used in
communication. Scholars working in the fields of mass communication, media effects,
and computer-mediated communication, with their focus on the effects of different media
channels and technologies on communication processes and outcomes, have considered
many of the same concerns that occupy sociomaterial scholars, without explicitly evoking
sociomateriality as an ontological lens. Additionally, communication researchers from a
variety of sub-disciplines have employed an explicit sociomaterial perspective to study
communication phenomena (e.g., Aakhus, et al., 2011; Contractor, et al., 2011; Leonardi
& Barley, 2011; Siles & Boczkowski, 2012). Organizational communication researchers
identifying with the CCO school, however, have particularly embraced sociomateriality.
The CCO school proposes that communication is constitutive of all aspects of
organizations; that to organize is to communicate (Ashcraft, et al., 2009). CCO
researchers have applied sociomateriality to their work by viewing organizations as
composed of “a plenum of agencies,” (Cooren, 2006) both material and human, with
31
communication as “the site of their interpenetration, the process through which agencies
collide to co-create realities” (Ashcraft, et al., 2009, p. 35). Leonardi (in Aakhus, et al.,
2011) elaborates:
If we buy this idea that organizations are, as the CCO perspective would say, ‘a
plenum of agencies’ (Cooren, 2006)—both material agencies and human
agencies—then the way those agencies are organized becomes of particular
interest. This is where I think the idea of communication does real work. I would
argue that communication is the process by which material and human agencies
are imbricated (Leonardi, 2011). (p. 566)
In other words, according to the CCO perspective, organizational communication
involves both human agency and material agency, and thus, just as both human actors and
nonhuman actors have agency, albeit in different forms, both human actors and
nonhuman actors “communicate” (Ashcraft, et al., 2009), albeit in different ways. As
Leonardi and Barley (2011) explain, to say that an artifact “communicates” with a person
does not mean that, for example, that “an email client uses artificial intelligence to ask its
user about her day,” but rather that “when people engage with the material makeup of the
technology directly, meaning is produced, sustained, and changed” (p. 103).
For CCO scholars, material artifacts of interest in organizations have included not
only newer ICTs but also older, low-tech artifacts. For example, in a study of a budget
crisis at a university, and the subsequent reduction and elimination of academic
programs, Castor and Cooren (2006) demonstrate how several documents—a faculty
code of conduct, a set of letters about the proposed program changes written by the dean,
and a resolution sponsored by members of the faculty senate—were viewed by
32
participants as actors playing key roles, along with the dean himself and various other
faculty members—in the controversy. Similarly, Cooren (2004) discusses how sticky
notes helped a manager organize and perform daily routine tasks; and how a property
management company used security personnel and a new sign, both posted in a building
lobby, to communicate new security procedures to building visitors.
From these applications of sociomateriality in the CCO school of organizational
communication, this dissertation took two points of departure. First, like the CCO
scholars, I examined a range of types of material artifacts, including knowledge source
artifacts that were both newer technologies as well as commonplace, low-tech tools.
Second, I followed the CCO sociomaterialists’ proposal that both human and nonhuman
actors communicate in organizations, conceptualizing all of the relations in an
organizational knowledge network as different forms of communication relations—those
that exist between two people and those that exist between a person and an artifact. For
example, a communication relationship that exists between two people in the knowledge
network could be conceptualized as forming when one member of the organization asks a
colleague for their opinion about the effectiveness of a best practice. A second type of
communication relationship, one that exists between a person and an artifact, could be
conceptualized as forming when an organizational member accesses the Web site of a
professional organization for information about the best practice.
Applying Sociomateriality to Social Networks
If one accepts the premise, posed by scholars who have applied sociomateriality
to organization science and organizational communication, that material artifacts play a
significant role in organizational life, then it follows that they also play a significant role
33
in the social networks that emerge within and among organizations. In order to
conceptualize organizational networks as including material artifacts, however,
researchers need an ontology that supports the existence of a social network with
heterogeneous actors, both human and nonhuman. As Contractor and colleagues
(Contractor, 2009; Contractor, et al., 2011) have discussed, the application of
sociomateriality to social network research provides an ontological foundation for
conceptualizing networks composed of both human and nonhuman actors and the
relationships between them. With its recognition of material agency and sociomaterial
agency, sociomateriality enables consideration of how the patterns of relations among
human and nonhuman actors contribute to network structures in organizations.
For network scholars, the idea of including human and nonhuman nodes together
in a “sociomaterial” network is not as radical as it may seem to some at first blush, since,
as Butts (2009), Newman (2010), and others have observed, the ontology of “social”
networks has always been more complicated and fuzzy that the traditional network
boundaries and distinctions between “social” and “nonsocial” have implied. For network
scholars, sociomateriality acknowledges the ontological fuzziness found in empirical
contexts.
In addition to expanding our ontological conceptualization of networks,
sociomateriality also holds potential to expand our understanding of the structure of
empirical networks. As Contractor and colleagues (2011) observe, analysis of
sociomaterial networks may lead to the identification of distinctive structural signatures
not found in social-only networks. Or, alternately, analysis of sociomaterial networks
may lead to the finding that in some contexts, networks structures commonly found in
34
social-only networks are also found in sociomaterial networks, thereby increasing the
generalizability of such structures. This argument—that the application of
sociomateriality to social networks may expand our understanding of empirical network
structure—is parallel to previous arguments made in support of incorporating nonhuman
actors and materiality into other fields of inquiry. It echoes the argument made by
Orlikowski (2007), about incorporating the material world into organizational studies.
She writes that insights about organizing have been limited by the neglect of materiality,
“that we can gain considerable analytical insight if we give up on treating the social and
the material as distinct and largely independent spheres of organizational life”
(Orlikowski, 2007, p. 1438). It echoes the argument for greater consideration of
materiality in the field of communication, made by Flanagin (in Aakhus, et al., 2011),
who comments that “engaging with materiality and communication may lead us both to
consider domains new to the field and to discover novel conceptions of communication”
(p. 563). It is also similar to the argument made by Faust (2011), who points out that
studies of the social networks of animals—a different type of nonhuman actor—provide
valuable insights not only on animal behavior but on the similarities and differences that
exist across social networks of different types of actors, including humans. Faust
observes that such studies particularly provide insights “about social organization that are
perhaps lost on researchers who restrict attention to a single species (e.g., on Homo
sapiens)” (2011, p. 161). The same might be said of studies of sociomaterial networks of
human and material actors and relations, which may provide insights about social
phenomena that may not be observable in studies restricted to human actors only.
35
This is not to suggest that network scholars should henceforth conceptualize all
networks as sociomaterial, rather than social, networks, or that sociomateriality is always,
as Butts would say, the best “network representation for the problem at hand” (2009, p.
414). Instead, these arguments simply suggest that in some network theoretical and
empirical contexts and for some network research questions, a perspective that extends
the social to the sociomaterial is of value.
In the case of organizational knowledge networks, the application of
sociomateriality holds potential to expand scholars’ understanding of how people
communicate and use knowledge in organizations, supplementing the rich body of
literature that already exists on organizational knowledge networks of human nodes only.
Studies of sociomaterial knowledge networks offer the promise of yielding insights on
the similarities and differences between knowledge networks with different types of
nodes, and on what is distinctive about social knowledge networks versus sociomaterial
knowledge networks, thus enhancing our understanding of both.
In this dissertation I used sociomateriality to conceptualize an organizational
knowledge network of human and nonhuman actors. The human actors were physicians
in a health care organization and represented both consumers and sources of best-practice
knowledge. The nonhuman actors were material artifacts such as journals, Web sites, and
databases, which also represented sources of knowledge. The relations in the network
represented the communication of knowledge between all of the actors.
To move beyond simply a descriptive account of the sociomaterial entanglement
of people and artifacts communicating knowledge, I use two theories, organizational
learning theory and institutional theory, to propose and test hypotheses about the
36
structure of this network. Specifically, I used organizational learning theory to understand
how people’s communication load acts as one form of sociomaterial agency in a
knowledge network, and I used institutional theory to understand how the legitimacy and
credibility with which knowledge sources are imbued acts as a second form of
sociomaterial agency. This sociomaterial conceptualization of an organizational
knowledge network was used to illustrate how human and nonhuman actors and their
agencies become entangled and affect the structure of the network and the resultant social
phenomenon, the communication of organizational knowledge.
The next chapter elaborates on this empirical context, examining, through the lens
of sociomateriality, our understanding of organizational knowledge, organizational
knowledge networks, and the communication of best practices in health care
organizations.
37
CHAPTER 2: EMPIRICAL CONTEXT: COMMUNICATING HEALTH CARE
BEST-PRACTICE KNOWLEDGE IN ORGANIZATIONAL NETWORKS
Giddens (1990) proposes that people increasingly live in a world mediated by
“expert systems,” by knowledge produced by government agencies, scientific and
professional groups, and other institutional, rather than interpersonal, sources of
expertise. Building on Giddens’ argument, Knorr Cetina (1997) notes that this
development has implications both for our material and social worlds:
The expanding role of expert systems does not only result in the massive presence
of the technological and informational products of knowledge processes. It
implies the presence of the processes themselves, and of knowledge-related forms
of embeddedness and structures. . . . It means that knowledge cultures have
spilled and woven their tissue into society, the whole set of processes, experiences
and relationships that wait on knowledge and unfold with its articulation. . . . In a
postsocial knowledge society, mutually exclusive definitions of knowledge
processes and social processes are theoretically no longer adequate; we need to
trace the ways in which knowledge has become constitutive of social relations. (p.
8)
Here Knorr Cetina articulates a sociomaterial perspective of knowledge. When
she writes of “the massive presence” of the products of knowledge processes, she refers
to the ubiquity of material knowledge artifacts in everyday life, to how Law (1992, p.
381) describes knowledge, as quoted previously, as something that “always takes
material forms.” Knorr Cetina suggests that knowledge artifacts and the “knowledge
processes” they embody are entangled with other “social processes” and “social
38
relations,” which “unfold” when knowledge is “articulated” by expert systems. And she
attributes both the ubiquity of knowledge artifacts, and their extensive entanglement in
social processes, to the expansion of the “expert systems” produced by the institutions
and organizations to which we confer expertise. One product of such expert systems is
the “best-practice” knowledge communicated within professional and organizational
communities. This is the type of knowledge of interest here. In the following sections, I
review our current understanding of the communication of best-practice and other
knowledge in organizations generally, in organizational networks more specifically, and
in organizational networks in the fields of health care and public health in particular.
Throughout the chapter, I connect the review of these three interrelated areas of
scholarship to the sociomaterial ontology described in Chapter 1.
Communicating Knowledge in Organizations
A comprehensive model of knowledge communication in organizations,
incorporating research from a range of disciplines and theoretical perspectives, would
include variables at the individual, group, organizational, inter-organizational, and
institutional levels, and would encompass characteristics of the knowledge sources and
producers, the knowledge itself, the people who use the knowledge, and the
organizational context in which the knowledge is communicated. The purpose of the
present research is not to propose such a global model. Instead, building on the
sociomaterial ontology described in Chapter 1, and on propositions from organizational
learning and institutional theories, this dissertation takes up the more modest task of
examining how three characteristics of the human and material actors in an organizational
knowledge network—communication load, legitimacy, and credibility—influence and
39
become entangled in the structure of knowledge communication. In this section I
examine two foundational aspects of organizational knowledge: types of knowledge and
knowledge behavior.
Knowledge typologies and epistemologies
In considering knowledge typologies, it is advisable to briefly attend to two
preliminary tasks. The first is to identify major conceptual perspectives, or
epistemologies, of knowledge in organizations, and the second is to define what is meant
by “organizational knowledge.”
Given that knowledge is a concept of relevance in many empirical and theoretical
contexts, scholarly perspectives and propositions about knowledge are numerous, even if
one narrows the scope of what is meant by “knowledge” to “organizational knowledge.”
A number of scholars (Cook & Brown, 1999; Easterby-Smith & Lyles, 2011a; Hayes,
2011; Iverson & McPhee, 2002; Kuhn & Porter, 2011; Orlikowski, 2002) have identified
two major conceptual perspectives or epistemologies on knowledge in organizations and
organizing. The first perspective has been referred to variously as a content (Hayes,
2011), possession (Cook & Brown, 1999), taxonomic (Orlikowski, 2002; Tsoukas, 1996),
informational (Iverson & McPhee, 2002), or cognitivist (Argote & Miron-Spektor, 2011;
Tsoukas, 2011) perspective. According to this perspective, it is possible and desirable to
“maximize the efficient use of knowledge in organizations” (Easterby-Smith & Lyles,
2011a, p. 12)—although typically this perspective also recognizes that in practice the use
of knowledge is not always, or even often, rational. According to the “cognitivist”
perspective, knowledge can be possessed, stored, and shared—it can take the form of an
entity or “thing.” Examples of a more cognitivist approach include the work of Nonaka
40
and Takeuchi (Nonaka, 1994; Nonaka & Takeuchi, 1995); von Hippel (1994) and
Szulanski (Jensen & Szulanski, 2004; Szulanski, 1996); Kane and Alavi (2007); and
Hansen and Haas (Haas & Hansen, 2007; Hansen, 1999, 2002; Hansen & Haas, 2001).
The second perspective has been referred to variously as a social constructionist
(Easterby-Smith & Lyles, 2011a; Tsoukas, 1996), relational (Hayes, 2011), interactional
(Iverson & McPhee, 2002), or practice-based (Argote & Miron-Spektor, 2011; Cook &
Brown, 1999; Kuhn & Porter, 2011) perspective. According to this “practice-based”
perspective, knowledge is situated in social processes, communities, and specific
organizational contexts, and “cannot be passed from person to person as if it were a
physical object” (Easterby-Smith & Lyles, 2011a, p. 9). It is neither static nor an object at
all, but rather a dynamic practice or action, and to reflect this, “knowing” is a more
appropriate term than “knowledge.” Those who take a more practice-based approach
include Blackler (1995); Cook, Brown, and Duguid (J. S. Brown & Duguid, 2001; Cook
& Brown, 1999); Orlikowski (2002); Lave and Wenger (Lave & Wenger, 1991; Wenger,
1998); and Spender (1996).
“Organizational knowledge” is knowledge that is “generated, developed, and
transmitted by individuals within organizations” (Tsoukas & Vladimirou, 2001, p. 979).
More specifically, organizational knowledge is “multifaceted, grounded in various
aspects of human experience, and individual as well as collective” (Canary, 2011, p.
247). If one considers both conceptual perspectives outlined above, organizational
knowledge can refer to both a noun and a verb, a thing and a process, knowledge and
knowing. In summary, “organizational knowledge is the capability members of an
organization have developed to draw distinctions in the process of carrying out their
41
work, in particular concrete contexts, by enacting sets of generalizations whose
application depends on historically evolved collective understandings” (Tsoukas &
Vladimirou, 2001, p. 973). Tsoukas and Vladimirou’s (2001) definition recognizes that
knowledge is both individual, involving “members of an organization,” and “collective”;
“context[ualized]” but also “generaliz[able]”; a latent “capability” or thing, and a
“process” or action.
Scholars have identified a number of different types of knowledge in an effort to
capture the complexity of knowledge and/or to summarize the ways in which knowledge
has been discussed in the research literature (Canary, 2011). These include dichotomous
typologies such as explicit knowledge vs. tacit knowledge (Polanyi, 1966) and sticky
information vs. non-sticky information (Jensen & Szulanski, 2004; Szulanski, 1996; von
Hippel, 1994); as well as more elaborate typologies based on the perceived location of
knowledge (i.e., in bodies, brains, routines, etc.) (Blackler, 1995; Collins, 1993) and on
the combined dichotomies of explicit/tacit and individual/collective (Spender, 1996). A
number of these typologies have been associated with a more cognitivist epistemology
(Tsoukas, 1996), although many scholars (e.g., Cook & Brown, 1999; Spender, 1996)
taking a more practice-based approach have found use for—or have been unable to
completely avoid—knowledge typologies as well.
As Canary (2011) observes, many of these knowledge typologies build upon
Polanyi’s (1966) explicit/tacit typology, and as such an elaboration of this concept is
warranted. Although there has been much scholarly debate about the definitions,
operationalizations, and most faithful interpretations of “explicit” and “tacit,” (Cook &
Brown, 1999; Corman & Dooley, 2011; Nonaka, 1994; Nonaka & Takeuchi, 1995;
42
Tsoukas, 1996, 2011; Tsoukas & Vladimirou, 2001) explicit knowledge is often
explained as more abstract, “know-what” knowledge that can be codified and that is
accessible from artifacts or people. Tacit knowledge is often explained as more situated,
“know-how” knowledge embedded in experience and specific contexts and obtainable
from other people or through one’s own experience. For example, someone looking for
explicit knowledge in health care delivery might ask, “What is the current best practice
for flu vaccination?”, whereas someone looking for tacit knowledge might ask, “How
should I operationalize flu vaccination best practices in my outpatient clinic, with my
particular patient population?” While earlier applications of the explicit/tacit distinction
emphasized an either/or quality and often valued one dimension over the other, more
recently scholars have argued for a less binary view and a recognition of the importance
of both dimensions (J. S. Brown & Duguid, 2001; Corman & Dooley, 2011; McPhee,
Canary, & Iverson, 2011; Tsoukas, 2011). As these scholars have noted, this more
nuanced interpretation of Polanyi’s precepts corresponds with Polanyi’s (1966) own
original conceptualization of the dimensions, which he describes as two aspects of
knowing; “neither is present without the other” (1966, p. 7).
In public health and health care, much attention has been devoted to a type of
knowledge referred to as “best practices.” As many observers have noted, a topic of
particular interest has been the communication of best-practice knowledge and evidence-
based medicine (Dearing, 2004; Green, Ottoson, García, & Hiatt, 2009; Greenhalgh,
Robert, Bate, Macfarlane, & Kyriakidou, 2005; Grol, 2001; Murphy & Eisenberg, 2011;
Raman & Bharadwaj, 2012), and also, more generally, the transfer of new research
evidence from scholars to practitioners (Cash et al., 2003; Lavis, Robertson, Woodside,
43
McLeod, & Abelson, 2003; Nutley, Walter, & Davies, 2007) and the circulation of
fashionable and/or legitimated best practices or “management knowledge” within
organizational communities and industries (Abrahamson, 1996; Calhoun, Starbuck, &
Abrahamson, 2011; Orlikowski, 2002; Orlikowski & Scott, 2009; Sahlin-Andersson &
Engwall, 2002; Sahlin & Wedlin, 2008; Szulanski, 1996). As Strang and Macy observe,
“Increasingly, competitive benchmarking has been adopted as a mode of corporate
planning and innovation, where firms learn from the practices of the ‘best in class’”
(2001, p. 175). Best-practice knowledge has been conceptualized as having “empirically
demonstrated advantages over other means of achieving the same ends,” (Dearing, 2004,
p. 23), and as “a set of interrelated work activities repeatedly utilized by individuals or
groups that a body of knowledge demonstrates will yield an optimal result,” (Tucker,
Nembhard, & Edmondson, 2007, p. 894). Wellstein and Kieser (2011) provide a
description of best-practice knowledge in health care contexts:
Medical good or best practices are usually described in great detail and are made
available to everybody in this field who is involved in the respective treatment.
The usual process is that experts appointed by medical associations screen and
evaluate available empirical evidence and establish consensus on good or best
practices and recommend them to hospitals or practicing physicians. . . . There is
a general interest across the field to standardize and spread good practices of
treatment and patient care. (p. 685)
It should be noted here that “best practice” is, as are many terms dressed with the
“best” adjective, a contested term. Critics question whether it is possible for members of
one organization to identify and utilize the best-practice knowledge of members of other
44
organizations, given the context-specific, interdependent nature of organizational
practices (e.g., Orlikowski, 2002; Rycroft-Malone et al., 2004; Wellstein & Kieser,
2011). As Orlikowski (2002) argues, “Leaving aside the problematic notion of who
decides what ‘best’ means, practices are, by definition, situationally constituted. They are
not discrete objects to be exchanged or stable processes to be packaged and transported to
other domains” (p. 271). This view suggests that individual and organizational learning
based upon knowledge acquired from external organizations, networks, institutions, and
other sources—as opposed to learning based upon knowledge acquired through one’s
own experience—is a questionable enterprise. I used the concept of best-practice
knowledge in this dissertation not for the purposes of endorsing its merit or otherwise
engaging in its politics, but because it is a term widely used in the health care empirical
context of this study, by both the designers and creators of such knowledge, who wish for
it to be adopted, and by its potential consumers and adopters. “Best practice” serves as a
convenient and empirically-grounded descriptor of the type of knowledge of interest in
this study: knowledge that is accessed from external sources and not from accumulated
experience, that has been legitimated to some degree, and that is recognized by health
care organizations as being of potential use and value.
Terminology for the concept of “knowledge” itself also merits brief examination
here. The terms “information” and “knowledge” have been used similarly and
interchangeably in the research literature. When scholars take care to distinguish between
the two, it is often with a hierarchy (Case, 2008; Davenport & Prusak, 1998; Nonaka,
1994), where knowledge is a higher form of information and a “phenomenon of the
human mind,” and information is more basic, “something that either reduces uncertainty
45
or changes one’s image of reality” (Case, 2008, pp. 65-66). I used the term “knowledge”
here, rather than “information,” because in the case of health care best practices, someone
has by definition decided that these practices are “best,” implying that they are closer to,
in Case’s (2008) words, a phenomenon of the human mind. In addition, I used
“knowledge” rather than the term “innovation,” because “innovation” is sometimes
associated more with new, original products or practices that are created, rather than
those best practices that were developed in the past, evaluated, proven superior to
alternatives, and that are now being disseminated (Greenhalgh, et al., 2005). Relative to
other industries where innovation for competitive advantage may be of greater interest,
best-practice knowledge is emphasized among health professionals, who often strive to
conform to standard practices of health care delivery endorsed and mandated by various
institutions (Provan, Beagles, Mercken, & Leischow, 2012).
Returning to the question of epistemology, this study took a cognitivist approach
to knowledge, in the sense that if one conceptualizes knowledge as something that can be
embodied in material forms, this implies that knowledge can be viewed as a “thing.”
Further, the focus on “best-practice” knowledge suggests a more cognitivist orientation
toward the content of knowledge; an assumption that knowledge can be better or worse,
more or less efficient; and a belief that knowledge can be communicated, transferred, and
shared. On the other hand, this study referenced a practice-based epistemology in that it
focused, as Blackler (1995) argues for, on “the systems through which knowing and
doing are achieved” (p. 1040). Here, the “system through which knowing is achieved”
was conceptualized as a sociomaterial knowledge network, in which knowing was
achieved through the interaction of human and nonhuman actors. Additionally, the
46
sociomaterial ontology of this dissertation assumed that social and the material are
entangled and mutually constitute social structures and phenomena, such as knowledge
communication and knowing. It conceptualized human and nonhuman actors as sources
of knowledge, not in isolation, but as situated in a network of communication
relationships, which recalls the practice-based perspective that knowledge is situated in
social processes and communities. Tsoukas and Vladimirou (2001) write that “managing
organizational knowledge does not narrowly imply efficiently managing hard bits of
information but, more subtly, sustaining and strengthening social practices” (p. 991). To
express the view the present study took, I would amend this statement just slightly,
changing “social practices” to “sociomaterial practices.”
Indeed, the cognitivist and practice-based perspectives on knowledge are not
necessarily irreconcilable, although they are sometimes portrayed as such. Ironically, the
term “best practice” is associated more with the cognitivist approach and is critiqued by
some who take a practice-based view, but the term itself contains the word “practice,”
seemingly acknowledging that knowledge (or knowing) is embedded in practice. This
irony perhaps speaks to the elusive nature of knowledge as a topic of scholarly attention,
to the fact that knowledge is difficult to pin down as definitively one thing and not
another. Accordingly, Cook and Brown (1999) propose that scholars can better
understand organizations “if knowledge and knowing are seen as mutually enabling (not
competing)” (p. 381), calling “what is possessed, ‘knowledge,’ and what is part of action,
‘knowing’” (p. 382). Similarly, in articulating her practice-based orientation, Orlikowski
(2002) writes that she intends her perspective to be a complement, not a substitute, to
other approaches—in other words, that neither perspective is “right” or “wrong.”
47
Best-practices and other types of knowledge become embodied in material
artifacts when this knowledge is written down in texts or embedded in technologies such
as software and electronic databases by people and organizations (D'Adderio, 2011). In
doing so, people and organizations exercise what might be conceptualized as, to borrow
Cooren’s (2006) term, “upstream” sociomaterial agency. This is to be distinguished from
“downstream” sociomaterial agency, that is, the agency that is exercised when people use
these knowledge artifacts, which is the subject of the next section.
Knowledge behavior
Scholars use various terms to characterize the human behavior of interest when
studying organizational knowledge, depending on the research question, of course, but
also on disciplinary affiliation, theoretical orientation, and current fashion in the research
literature. Thus literature may refer to knowledge creation (e.g., Lee & Cole, 2003;
Nonaka, 1994); knowledge seeking (typically used with the noun “information,” e.g.,
Borgatti & Cross, 2003; Case, 2008); knowledge sharing (e.g., Cummings, 2004; Dyer &
Nobeoka, 2000; Haas & Hansen, 2007; Hansen, 1999; Sharratt & Usoro, 2003; Wasko &
Faraj, 2005) or transfer (favored by organizational learning scholars and many others,
e.g., Argote, Ingram, Levine, & Moreland, 2000; Phelps, Heidl, & Wadhwa, 2012;
Reagans & McEvily, 2003; Szulanski, 1996); knowledge adoption (often used with the
term “innovation,” e.g., Burns & Wholey, 1993; Rogers, 2003; Schuster et al., 2006;
Westphal, Gulati, & Shortell, 1997); knowledge management (a term representing an
entire subfield of management literature, e.g., the Journal of Knowledge Management;
Davenport & Prusak, 1998; Iverson & McPhee, 2002; Lehr & Rice, 2002; Shumate,
48
2011); and so on. A smaller body of research addresses knowledge ignoring and
avoidance (Brashers, Goldsmith, & Hsieh, 2002; Case, 2008; Eppler & Mengis, 2004).
Some of these behaviors are more conceptually distinct than others. Creating
knowledge, for example, seems reasonably different from managing already-existing
knowledge. The distinctions between knowledge sharing and knowledge transfer appear
less clear, however—when is transfer not sharing, and vice versa? Further, the difference
between some terms seems one of perspective, rather than an empirically observable
distinction. For example, knowledge seeking could result in the same net behavioral
outcome as knowledge sharing, but seeking places behavioral agency with the knowledge
recipient, and sharing with the knowledge source. In the present project I conceptualized
all of these knowledge behaviors as falling under a common umbrella of knowledge
communication activity.
There were several reasons to do this. First, all of the above behaviors require
communication of one sort or another, and using an umbrella term thus included all.
Unlike some of the behaviors discussed above, knowledge communication makes few
assumptions about the nature, behavior, relationship, agency, or hierarchy of the actors
involved in the knowledge behavior.
Second, in this dissertation I have echoed the CCO sociomaterialists’ argument
for the primacy of communication in organizing, by choosing to describe all of the
relations in an organizational knowledge network as communication relations. “One
cannot not communicate,” according to an axiom of communication scholarship
(Watzlawick, Beavin, & Jackson, 1967, pp. 48-49), and it is difficult to imagine
knowledge existing among members of organizations, or among organizations
49
themselves, without the presence of various forms of communication. A number of
scholars have presented claims for the inherently communicative constitution of
knowledge, arguing that “organizational knowledge is developed, manifested, managed,
and/or utilized through communication” (Canary & McPhee, 2011, p. 1), and for
knowledge to be viewed as a “resource/outcome of communication” (McPhee, et al.,
2011, p. 307).
Communicating Knowledge in Organizational Networks
Having briefly outlined some of the scholarly definitions and epistemologies that
ground our understanding of how knowledge is communicated in organizations, I turn
now to the more specific context of how knowledge is communicated in organizational
networks. The network perspective on organizational knowledge allows for the
consideration of how the attributes of actors affect knowledge communication, and also,
importantly, of how the characteristics of the relationships between those actors affect
knowledge communication. Research on organizational knowledge networks has
encompassed all levels of analysis, from advice relationships between colleagues, to
collaborative innovation within a team, to knowledge transfer between organizations. In
the following sections, I briefly review how scholars have studied social-only knowledge
networks, and then outline how they have studied, to a much more limited extent,
sociomaterial knowledge networks, providing the basis for the current project.
Social knowledge networks
Organizational and professional knowledge networks composed of human actors
have been a topic of enduring interest in the social network, organization science, and
organizational communication literatures. In fact, two of the most studied, most cited
50
theories of social networks (Borgatti & Halgin, 2011)—Granovetter’s strength of weak
ties theory (1973) and Burt’s structural holes theory (1992)—were originally motivated
by research questions about knowledge communication among people in professional and
organizational contexts. Granovetter was interested in how job seekers acquired
knowledge about the job market from their interpersonal networks. Burt was interested in
how professionals who connected otherwise unconnected people in their professional
networks acquired competitive advantage by doing so.
Over the past 15 years, there have been a number of reviews of the scholarship on
knowledge communication in organizational networks specifically (e.g., Phelps, et al.,
2012; van Wijk, Jansen, & Lyles, 2008; Van Wijk, Van Den Bosch, & Volberda, 2011),
and on behavior in organizational networks more generally (e.g., Baker & Faulkner,
2002; Borgatti & Foster, 2003; Brass, Galaskiewicz, Greve, & Tsai, 2004; Carpenter, Li,
& Jiang, 2012; Gulati, Dialdin, & Wang, 2002; Ibarra, Kilduff, & Tsai, 2005; Kilduff &
Brass, 2010; Kilduff, Tsai, & Hanke, 2006; P. R. Monge & Contractor, 2003; Owen-
Smith & Powell, 2008; Provan, Fish, & Sydow, 2007; Rank, Robins, & Pattison, 2010;
Shipilov, Gulati, Kilduff, Li, & Tsai, 2014; Shumate & Contractor, 2013). These reviews
highlight dominant lines of research and key findings on the characteristics of the actors
and relations in organizational knowledge networks where the nodes are human only.
Because the literature on such organizational knowledge networks has been reviewed
elsewhere, and because another thorough review of this literature is beyond the scope of
this dissertation, I confined my discussion here to findings relevant to the empirical
context of this dissertation, intraorganizational knowledge communication networks.
Additionally, because I review literature on relevant actor characteristics in Chapters 3
51
and 4, I primarily focus in this section on relevant structural characteristics of social-only
intraorganizational knowledge networks. These include the dyadic structural
characteristics of reciprocity, multiplexity, and spatial proximity; the dyadic actor
characteristic of homophily;
3
and the broader (triad or greater) structural characteristics
of centrality, brokerage, and closure.
Reciprocity refers to the tendency for network relationships to be bi-directional,
rather than uni-directional. It has been a subject of consistent focus in the scholarship on
organizational knowledge networks. Generally speaking, researchers have observed that
reciprocal knowledge ties are more common in organizational contexts (a) with less
social status or formal hierarchy, where giving and seeking knowledge has little effect on
status; and (b) where equitable, balanced exchange—that is, repaying colleagues who
share resources—is possible both because everyone possesses some knowledge of value
to the group, and because doing so is the social norm (Agneessens & Wittek, 2012;
Lazega, Mounier, Snijders, & Tubaro, 2012; Nebus, 2006). So, for example, reciprocity
was a significant tendency in the longitudinal advice relationships of employees of a
public housing organization (Agneessens & Wittek, 2012), in the advice ties of managers
in two multinational corporations (Rank, et al., 2010) and in a multiunit industrial
company (Lomi, Lusher, Pattison, & Robins, 2014), and in the knowledge transfer
relationships of employees in two multinational consulting firms (Su, Huang, &
Contractor, 2010). In the context of health care professionals, Keating and colleagues
3
I discuss homophily here, although it is not a structural characteristic, because it is one of the most studied
features of organizational knowledge networks (and social networks generally), and because although I
control for a number of homophily effects in my analysis, I do not focus on homophily in my research
hypotheses and thus do not discuss it in Chapters 3 and 4.
52
(2007) found that reciprocity was a significant predictor of knowledge ties and Zappa
(2011) found that reciprocity characterized 93% of such relationships.
Conversely, reciprocal knowledge ties are less common in contexts where the
status or formal hierarchy differential is substantial, knowledge-seeking behavior may
decrease status, and/or some actors in the network have little knowledge to provide and
thus cannot reciprocate. For example, Lazega and van Duijn (1997) observed that advice
was not reciprocated between senior attorneys and their more junior counterparts in a law
firm. Direct, tit-for-tat reciprocity may also be less common in contexts where indirect,
generalized reciprocity—that is, where a gift is repaid not by the direct recipient, but by
someone else in the social group—has emerged as the more dominant social norm. For
example, Brennecke and Rank (in press) observed that among research-and-development
employees who worked in the same project teams, advice relationships were mostly but
not uniformly characterized by reciprocity. The authors hypothesized that reciprocity was
not present in the advice relationships in some teams because in those teams, the norm of
generalized reciprocity, rather than direct reciprocity, was at work.
A second dyadic structural characteristic of interest in the literature on
organizational knowledge networks is multiplexity. Multiplexity refers to the presence of
more than one type of tie between two actors. Probably the most commonly reported
finding regarding multiplexity is that affective or expressive ties often co-occur with
more instrumental knowledge communication ties. This is because most of us, all else
being equal, prefer to obtain knowledge from people we already know and like (Nebus,
2006). For example, a number of studies have found that friendship facilitates both
intraorganizational (Lazega & Pattison, 1999; Morrison, 2002; Saint-Charles &
53
Mongeau, 2009) and interorganizational (Bell & Zaheer, 2007; Coleman, Katz, &
Menzel, 1966; Ingram & Roberts, 2000) knowledge transfer relationships. Similarly, Tsai
(2002) found that co-attendance at intraorganizational social events facilitated knowledge
sharing in a large multi-unit corporation. In a related stream of research, theory and
empirical evidence suggests that knowledge ties are more likely to occur when any sort of
positive affect ties (not just friendship ties) are also present, and much less likely to occur
when negative affect is present (Casciaro & Lobo, 2008; Labianca & Brass, 2006).
A substantial body of literature indicates the importance of a third dyadic
structural characteristic, spatial proximity, or propinquity, to organizational
communication of all sorts, including knowledge communication (P. R. Monge &
Contractor, 2003; Phelps, et al., 2012). Spatial proximity is influential because it
increases the efficiency and efficacy of knowledge transfer, but the knowledge
transferred is likely to be less novel (e.g., Bell & Zaheer, 2007). Empirical research has
supported a positive association between propinquity and knowledge network relations in
the case of advice relationships between employees of a public housing organization
(Agneessens & Wittek, 2012), information seeking ties between researchers at two
different pharmaceutical companies (Borgatti & Cross, 2003), and knowledge sharing ties
between primary care physicians working in separate clinics within a teaching hospital
(Keating, et al., 2007).
Whereas spatial proximity describes our preference for knowledge
communication with actors physically close to us, homophily refers more broadly to our
preference for knowledge communication with those with whom we feel “close” because
they share one or more personal characteristics with us. Numerous studies in the field of
54
social networks support the proposition that homophily facilitates all sorts of
communication relationships (McPherson, Smith-Lovin, & Cook, 2001). Research
suggests that knowledge communication between actors who are more homophilous will
be easier than knowledge communication between actors who are less homophilous,
because actors with similar characteristics are likely to share some existing common
knowledge and thus less translation of the knowledge is required in a homophilous
context (Szulanski, 1996; von Hippel, 1994). For example, a student studying the field of
communication may find it easier to learn about statistical analysis from a
communication professor than from an economics professor, even if the statistical
concepts taught by both professors are the same. Knowledge seekers are also more likely
to select knowledge sources that are homophilous simply because they are easier to find
(Darr & Kurtzberg, 2000) and are perceived to be more likely to be helpful or willing to
help. Among health professionals, for example, social and cognitive boundaries between
different professions with different education, training, and so forth, have been identified
as barriers to knowledge communication (Ferlie, Fitzgerald, Wood, & Hawkins, 2005).
Extending one’s gaze beyond the level of the dyad, scholars have also focused on
the association between knowledge communication and network centrality. A meta-
analysis of research on knowledge transfer in intraorganizational networks found a
positive relationship between centrality and knowledge transfer (van Wijk, et al., 2008).
A central position in a knowledge network affords knowledge seekers access to a greater
quantity of knowledge sources (Morrison, 2002), and allows knowledge sources to be
more accessible and to appear more trustworthy to their colleagues (Nerkar & Paruchuri,
2005). In studies of knowledge diffusion, scholars have observed that central knowledge
55
sources may act as opinion leaders, driving the spread of knowledge through the network
(Valente, 2012). Greater centrality also has its costs, however. The greater the number of
relationships the focal central actor forms and maintains, the greater the time and social
capital the central actor must spend on these relationships (Phelps, et al., 2012). Note also
that greater centrality as a knowledge seeker (that is, seeking knowledge from relatively
more sources as compared to one’s colleagues) may not be associated with greater
centrality as a knowledge source (that is, being identified as a source more frequently as
compared to one’s colleagues). In their study of advice relations in a public housing
organization, Agneessens & Wittek (2012) found that more active advice seekers were
less likely to be sources, perhaps because active advice seeking signals a lack of expertise
and discourages colleagues from viewing the seeker as knowledgeable.
Empirically, it is difficult to identify any particular trend in the extant literature
indicating that centralization is more or less commonly observed in organizational
knowledge networks. Both centralized and decentralized structures have frequently been
observed in knowledge networks. Likewise, it is difficult to claim that an organizational
network that is characterized by either greater or lesser degrees of centralization will be
more or less conducive to knowledge communication. As Valente (2010) points out, the
effect of network centralization on diffusion outcomes depends on the characteristics of
the actors holding the central positions.
Scholars have also investigated, à la Burt (1992), mentioned above, the
association between knowledge communication and the presence of structural holes
between nodes in the network. A structural hole exists when two actors do not share a tie,
and the “hole” is filled when a third “bridge” or “broker” actor forms ties with the first
56
two actors, thereby indirectly connecting them. Actors who occupy central positions, for
example, can act as knowledge brokers by virtue of the greater numbers of alters
connected to them, and can locate relevant and diverse knowledge and then exchange it
with others. But less central actors may also act as brokers, as in a person peripherally
connected to two different groups who acts as a bridge between the two. For the actor
who serves as the broker, being connected to unconnected others may provide access to
more, and more diverse, knowledge (Reagans & McEvily, 2003), and may confer status
and power (Burt, 1992). For the two actors whom the broker connects, there are both
advantages and disadvantages to obtaining each other’s knowledge indirectly through a
broker (Hansen, 2002). On the one hand, an indirect tie to a knowledge source provides
access to knowledge that might otherwise be unobtainable, and reduces the time and
social capital costs associated with searching for a knowledge source and then
maintaining a direct relationship with that source. On the other hand, an indirect tie to a
knowledge source may result in a loss of knowledge fidelity that may occur in the
process of the transfer. This disadvantage is classically exemplified by the old children’s
game of telephone, where a message becomes increasingly distorted as it is relayed
through an increasing number of intermediaries. An indirect knowledge tie may also be
particularly ill-suited for the communication of ambiguous and tacit knowledge that
requires extended interaction, discussion, and clarification between the knowledge source
and knowledge seeker (Hansen, 2002).
A third extra-dyadic structural characteristic of interest in the literature on
organizational knowledge networks is triadic closure. Triadic closure occurs when three
actors are connected to each other, with the ties between them forming a closed triangle
57
structure. Closure has been positively associated with knowledge communication in a
number of intra- and interorganizational contexts (e.g., Morrison, 2002; Reagans &
McEvily, 2003; Uzzi, 1996, 1997). Scholars have at times viewed brokerage and closure
as mutually exclusive structural phenomena achieving opposite ends in knowledge
networks, with brokerage associated with the communication of more novel, diverse
knowledge, and closure associated with the communication of more trustworthy
knowledge in a more reliable fashion (Van Wijk, et al., 2011). But as van Wijk and
colleagues (2011) point out, they need not be viewed in such oppositional terms: both
brokerage and closure are associated with the facilitation of knowledge communication
and with positive knowledge-related outcomes such as performance improvement and
innovation, because they both “provide access to different types of knowledge that may
be unavailable otherwise” (p. 484)
There are two main forms of triadic closure, depending on the directionality of the
relationships between the three actors (Rank, et al., 2010). Cyclic closure occurs when all
three actors are in equivalent structural positions, with each node acting as both a
knowledge seeker and a knowledge source, for example, and with the directionality of all
three ties the same. Cyclic closure may indicate the presence of generalized or indirect
reciprocity (rather than tit-for-tat, direct reciprocity) in knowledge networks. Transitive
closure occurs when the three actors each hold different structural positions: for example,
one actor is a knowledge source for the other two, one actor seeks knowledge from the
other two, and one actor is both a source and seeker. Transitive closure may suggest a
hierarchical structure of knowledge seeking and providing in organizational networks. In
the empirical literature on organizational knowledge networks, a majority of studies have
58
observed transitive closure and not cyclic closure (Agneessens & Wittek, 2012;
Brennecke & Rank, in press; Lazega, et al., 2012; Lomi, et al., 2014; Rank, et al., 2010).
In summary, relationships in social knowledge networks tend to be reciprocated in
organizational contexts with less hierarchy and status differentiation; they are facilitated
by spatial proximity, actor homophily, and co-occurring positive affective relationships;
and the effects of the broader structural characteristics of centrality, brokerage, and
closure on knowledge communication may be advantageous or disadvantageous,
depending on the organizational context and type of knowledge involved. As this brief
review of the extant literature illustrates, scholars have developed a rich understanding of
the structure of knowledge communication in organizational networks. The typical
empirical network upon which this understanding is based, however—consisting of
human actors only—is, as Butts noted, “ quite restrictive” (2009, p. 414) in its ability to
represent reality. It ignores the vast array of material artifacts available to and used by
professionals as knowledge sources. As a result, a number of aspects of knowledge
communication in organizational networks remain less explored and less understood,
including how various forms of sociomaterial agency influence knowledge networks.
If one looks outside the literature on organizational networks, there are substantial
lines of research on people’s use of both other people and material artifacts as knowledge
sources, and the inclusion of artifacts in a study of knowledge communication would
hardly be seen as novel or radical. See, for example, work on information seeking in the
information science literature (Case, 2008), and media effects research in the
communication literature (J. Bryant & Oliver, 2009). However, these streams of research
have largely been conducted without explicit use of sociomateriality as an ontological
59
lens, and in non-network contexts, such that network structure has not been considered as
either an antecedent or consequence of knowledge communication in such research. Thus
the network literature has been deficient in its consideration of sociomaterial knowledge
communication, and other fields that do examine sociomaterial knowledge
communication, either implicitly or explicitly, have been deficient in their consideration
of network contexts.
Sociomaterial knowledge networks
Enter, then, the study of sociomaterial knowledge networks. In Chapter 1, I
argued for the application of sociomateriality to the study of social networks, in order to
ontologically support the conceptualization and analysis of the numerous networks of
human and nonhuman actors that exist in everyday organizational life. With its
recognition of material and sociomaterial agency, sociomateriality enables consideration
of how the patterns of relations among human and nonhuman actors contribute to the
communication of knowledge in organizations.
In Chapter 1, I argued that in addition to expanding our ontological
conceptualizations of the types of actors and relations that exist in networks,
sociomateriality also holds potential to expand our understanding of the structure of
empirical networks. The rich body of literature reviewed above, on common structural
patterns found in social-only organizational knowledge networks, may be extended with
research on the structure of sociomaterial networks (Contractor, et al., 2011). And, in
fact, there is a small but growing corpus of scholarship in which sociomaterial knowledge
networks have been empirically investigated using theoretically-derived hypotheses and
quantitative analysis.
60
One example of such work is a study of the knowledge network existing among
management consultants in two consulting firms (Su & Contractor, 2011). In this study,
the authors conceptualized the network as being composed of three types of actors:
consultants who sought task-related information, consultants who were information
sources, and a corporate intranet that served as a nonhuman information source. Su and
Contractor used transactive memory theory to investigate how the consultants’ formation
of new information-seeking relationships was affected by the existing structure of the
knowledge network and by the characteristics of the network actors and of the
information being sought. Their analysis yielded several findings: (1) the consultants’
information seeking was positively influenced by source expertise and source
accessibility, but only when the sources were human; (2) the consultants’ information
seeking was positively influenced by the quantity of information held by the source and
by whether their colleagues used the source, but only when the sources were nonhuman;
and (3) the information seeking was negatively influenced by the complexity of the
information being sought, again only when the sources were nonhuman.
The two primary structural findings of the study offer an intriguing addendum to
the conventional wisdom on the structure of social-only knowledge networks,
summarized above, and merit additional discussion. The first structural result was that
source accessibility facilitated information seeking relationships when the sources were
human. This is a structural result because the authors measured human source
accessibility structurally, in terms of relational multiplexity: a human source was defined
as more accessible to an information seeker if the seeker also reported collaboration and
frequent social communication relationships with the source. This finding suggests that in
61
sociomaterial knowledge networks, as in social-only knowledge networks, multiplexity
facilitates knowledge ties among people. It also raises the question of why accessibility
did not also encourage the use of artifact information sources. Notably, Su and Contractor
measured artifact source accessibility differently from human source accessibility, with a
Likert rating scale instead of with relational multiplexity, which may account in part for
why the accessibility variable was a significant source characteristic only for human
sources. We do not know if multiplexity may also facilitate knowledge ties between
people and artifacts, and this would be interesting question to investigate. It seems
reasonable to hypothesize that if people perceive colleagues with whom they already
have other types of relationships as more accessible information sources, people would
also find media artifacts that they already access for other purposes (such as
entertainment) to be more accessible information sources as well.
The second structural result of note was that in the case of nonhuman sources, a
consultant’s information seeking was facilitated if his or her colleagues also used that
source. This is a structural finding because a colleague’s use of an artifact source,
together with the focal consultant’s use of that source, represents a form of triadic closure
with the focal consultant, the colleague, and the artifact source as the three vertices in the
triangle. The two consultants share a co-worker relationship, which constitutes one edge
of the triangle; the “colleague” or alter consultant has sought information from a
particular artifact source, which constitutes a second edge of the triangle; and as a result
of the colleague’s information seeking choice, the focal consultant decides to seek
information from that source as well, creating the third, closing edge of the triangle. Su
and Contractor’s finding—that closure occurs significantly more often when consultants
62
seek information from artifacts than what might be expected by chance—replicates the
finding discussed in the previous section, that closure is a significant structural signature
in many social-only knowledge networks. The triadic closure observed in Su and
Contractor’s study may be the result of an explicit brokerage mechanism, whereby the
colleague recommends that the focal consultant use the artifact source, or even helps the
consultant access the source. Or the closure may be the result of social influence (e.g., the
consultant hears that his or her colleagues are using the source, and wants to do what
others are doing), satisficing, (e.g., the consultant hears that his or her colleagues are
using the source, and decides not to look for any alternative sources due to a lack of
time), or a combination of these and/or other factors. Su and Contractor’s analysis does
not pinpoint the exact mechanism(s) driving the structural closure effect they observe,
nor does it identify the exact reason why closure is observed in the case of artifact
sources but not human sources.
4
Future sociomaterial network research could build on
their study, however, by investigating both questions.
In another example of an empirical investigation of a sociomaterial knowledge
network, Keegan and colleagues (Keegan, Gergle, & Contractor, 2012) conceptualized a
network composed of two types of actors: Wikipedia authors and the articles they
collaboratively write and edit. The ties in the network represented authorship
relationships between the people and articles. Unlike the knowledge-seeking and -sharing
networks of interest in the present dissertation and in the above study by Su and
Contractor, Keegan and colleagues’ work focused on a different type of knowledge
4
The authors speculated that the vast quantity of information that is available from the artifact sources they
studied may have prompted the consultants to experience “cognitive overload” and to use the behavior of
their colleagues as a guide in seeking information from artifacts—a hypothesis to which I will return later,
in Chapter 3.
63
network, one in which people contribute knowledge to a collaborative product, outside of
a traditional organizational context. For their study, Keegan et al. examined how the
characteristics of both authors and articles affected who collaborated with whom, and on
which articles. They found that (1) highly experienced Wikipedia editors were more
likely to make numerous contributions to a few select articles about contemporary events;
and (2) editors with less or no experience were more likely to make contributions to
many articles about breaking news and historical events. In other words, in the case of
more experienced editors and articles about contemporary events, the knowledge network
was more centralized, whereas in the case of less experienced editors and articles about
breaking and historical events, the knowledge network was less centralized. Keegan and
colleagues also found that overall, the attributes of editors exerted greater influence on
the structure of the collaboration network than the attributes of articles.
The research by Keegan et al. suggests that as with social-only knowledge
networks, both centralized and decentralized structures may be observed in sociomaterial
knowledge networks and may be conducive to knowledge communication and
production. Keegan et al.’s study goes beyond the findings of previous scholarship on
online collective intelligence systems, however, by considering how the social and
material features of such systems simultaneously influence how knowledge, in the form
of encyclopedia articles, is collaboratively produced.
It is to this small body of work on sociomaterial knowledge networks,
exemplified by the two studies described above, that the present research aimed to
contribute. Like those two studies, a primary objective of this dissertation was to extend
64
our understanding of the structure of organizational knowledge networks by considering
them through the ontological lens of sociomateriality.
Communicating Knowledge in Health Contexts
Having briefly reviewed the literature on the empirical contexts of communicating
knowledge and “best-practice” knowledge in organizations, and doing so in social and
sociomaterial networks, this chapter concludes with a review of the literature on doing so
in health contexts. It introduces the empirical rationale for focusing on communication
load, legitimacy, and credibility as key sociomaterial influences on the structure of health
care best-practice communication.
In the fields of public health and health care there is acute concern over the need
for the communication of best practices, and at the same time widespread recognition of
the challenges of communicating such knowledge (Dearing, 2004; Glasgow & Emmons,
2007; Green, et al., 2009; Institute of Medicine, 2011; Lavis, et al., 2003; Nembhard,
Alexander, Hoff, & Ramanujam, 2009; Sussman, Valente, Rohrbach, Skara, & Pentz,
2006; Timmermans & Mauck, 2005; Van Zandt, 2004). Timmermans and Mauck (2005)
summarize the zeitgeist in this way:
So many parties have jumped on the EBM [evidence-based medicine] bandwagon
and so many clinical practice guidelines are churned out by individuals,
professional organizations, insurers, and others that the benefits of [clinical
practice] uniformity may disappear in the cacophony of overlapping, conflicting,
and poorly constructed guidelines. (p. 19)
Timmermans and Mauck’s description of prevailing sentiment regarding best-practice
guidelines reflects two key, interrelated challenges that scholarly and lay literature on
65
health care best practice diffusion and dissemination repeatedly identify: one challenge
involving the quantity of knowledge available, and the other involving the quality of that
knowledge.
The first challenge of concern for stakeholders in the communication of health
care best practices lies in the large volume of knowledge available (Balas & Boren, 2000;
Shaneyfelt & Centor, 2009; Shaneyfelt, et al., 1999). For organizational and institutional
sources and suppliers of knowledge, such as the American College of Physicians or the
Centers for Disease Control and Prevention, it is difficult to attract the attention of target
audience members in a crowded marketplace of best-practice knowledge. For potential
users or seekers of this knowledge, it is difficult to identify the most appropriate source
among so many candidates. For example, in May 2015, the National Guideline
Clearinghouse maintained by the U.S. Agency for Healthcare Research and Quality
(www.guideline.gov) contained summaries of 2,399 different best-practice guidelines
published by 104 different organizations (not counting those guidelines that did not meet
the Clearinghouse’s detailed quality standards or those that were older than five years).
Zooming in to consider best practices for a specific health condition, ten separate sets of
clinical guidelines were available in 2007 for appropriate clinical treatment of sore throat
in adults (Shaneyfelt & Centor, 2009).
A meta-analysis of 76 studies of barriers to physician adherence to best-practice
guidelines found that basic awareness of the best practice in question was the barrier most
commonly identified in surveys—with guideline accessibility, physicians’ lack of time,
and overall volume of information available identified as key factors contributing to lack
of awareness (Cabana et al., 1999). Information volume becomes particularly problematic
66
for potential consumers when some of the information about a particular set of best
practices conflicts (Cabana, et al., 1999), as is likely the case when there are, to refer
back to the above scenario, ten separate sets of guidelines for treatment of sore throat.
Researchers estimate that the average primary care physician would need 18 hours per
day to carry out current best-practice guidelines for preventive care and chronic disease
management for a typical panel of patients (Bodenheimer, 2006; Østbye et al., 2005;
Yarnall, Pollak, Østbye, Krause, & Michener, 2003).
Given the volume of health care best-practice guidelines published and
disseminated, it is not surprising that many observers express concern for the potential for
health care professionals to feel overwhelmed or overloaded by the quantity of
knowledge available. “For a physician, trying to process this rush of inputs can feel like
drinking from a fire hose,” notes one editorial (Avorn, 2013, p. A27). As Haas and
Hansen (2001) observe, “the main constraint for successful knowledge dissemination in
the interorganizational market is the limited amount of time people have to consult this
knowledge” (pp. 25-26).
A second challenge of communicating and learning about best-practice
knowledge lies in evaluating the quality of the knowledge. Stakeholders have identified
several aspects of the quality of health care best practice guidelines that may compromise
their communication and diffusion among physicians and other health care professionals.
One commonly expressed concern involves the process by which guidelines are
formulated. Health care best-practice guidelines are typically written by a panel of
experts convened by a professional, not-for-profit, or government agency, and observers
have questioned the scientific evidence these experts use to formulate best-practice
67
recommendations, as well as the system by which the experts review the evidence and
reach consensus on how to apply it (Institute of Medicine, 2011).
A second, related concern involves the potential for bias and conflicts of interest
among guideline authors. The potential for a particular set of best-practice
recommendations to be influenced by commercial interests, such as pharmaceutical
companies, that stand to benefit from a particular treatment recommendation, has been
well-documented, especially when some of the experts who write the guidelines in
question serve as paid pharmaceutical consultants (Avorn, 2013; Lenzer, Hoffman,
Furberg, & Ioannidis, 2013). Uncertainty about the process by which best-practice
guidelines are developed, and worry about the potential for bias to exist in guidelines,
contribute to health care professionals’ uncertainty over the legitimacy of best-practice
authors and the credibility of their work. The previously-cited meta-analysis of barriers to
best-practice adoption identifies physicians’ overall lack of confidence in guideline
authors as a key attitudinal barrier (Cabana, et al., 1999).
In a third type of concern over best-practice quality, many observers identify a
distinction, or gap, between a physician acquiring awareness and knowledge of a best
practice in the abstract, and that physician acquiring the know-how and knowledge
needed to implement the practice in his or her own work context. Grol (2001), for
example, discusses how efforts to transfer evidence-based medicine are often ineffective
because information about the best practice may be ambiguous, may address only part of
the sequence of steps needed to apply the practice to one’s own work, and may require
broader systemic changes not feasible for many individual practitioners. Dearing (2004)
explains that among other factors, barriers to transfer of best-practice knowledge may
68
include the academic, rather than practical, orientation of some best-practice knowledge
sources. Shaneyfelt and colleagues (2009; 1999) point out that the values and priorities
that inform the formulation of some best practices may be at odds with the values and
priorities of potential users of this knowledge (e.g., the values of cost effectiveness versus
risk reduction). Writing of the gap between research evidence of best practices and the
practices actually implemented by practitioners, Green and colleagues (2009) observe,
The production and dissemination of [health best-practice] evidence is organized
institutionally with highly centralized (mostly federal and national) funding,
storing, indexing, synthesizing, and disseminating of science, whereas the
application of that science is highly decentralized . . . The gap [between research
evidence and practice] is then partly one of social distance between the supply
and the demand sides of science in geography as well as in organizational and
professional or personal self-identities. (pp. 154-155)
In response to the challenges of a potentially overwhelming quantity of best-
practice knowledge available of potentially questionable quality, a number of
organizations have created guidelines for the guidelines (Timmermans & Mauck, 2005);
see for example, the U.S. Institute of Medicine’s recent report titled, “Clinical Practice
Guidelines We Can Trust” (2011). Such efforts aim to provide guidelines and standards
for best-practice authors on how best to craft high-quality guidelines, and guidelines for
knowledge consumers on how to best evaluate guideline quality, given the volume of
information available on clinical best practices.
It is in this empirical context of health care professionals, organizations, and best-
practice guidelines, in which best-practice communication is both highly valued and
69
highly problematic, that the primary research question of this dissertation is investigated:
how do the characteristics of both people and knowledge artifacts affect the structure of
best-practice knowledge communication in a health care organizational network? Given
ongoing concern in the public health and health care literatures about how the quantity
and quality of best-practice information affects its diffusion, I focused in this project on
how three forms of sociomaterial agency—the communication load of knowledge
consumers, and the legitimacy and credibility of knowledge sources—affect the structure
of communication in a health care organizational network. Communication load
addresses the challenge of quantity, but also, quality. Legitimacy and credibility also
address the challenge of quality. To investigate the effects of these forms of sociomaterial
agency on knowledge network structure, I used two theories. I used organizational
learning theory to hypothesize how physician communication load affected the structure
of the knowledge network, and institutional theory to understand how knowledge source
legitimacy and credibility affected the structure of the network. Chapter 3 and Chapter 4
elaborate on these theories and the hypotheses derived from them.
70
CHAPTER 3: ORGANIZATIONAL LEARNING THEORY AND THE
SOCIOMATERIALITY OF COMMUNICATION LOAD
In the previous two chapters, my objectives were to establish the ontological and
then empirical foundations for this project’s overarching research question, examining
how the characteristics of both people and artifacts affect the structure of best-practice
knowledge communication in a health care organizational network. In the present chapter
and in Chapter 4, I explain how I investigated this research question from two theoretical
perspectives: organizational learning theory and institutional theory.
There are several advantages to adopting a multitheoretical perspective for the
present project. First and most generally, a multitheoretical perspective offers the
potential to increase the explanatory power of a research study by increasing the amount
of variance that is accounted for in the phenomenon of interest (Mayer & Sparrowe,
2013; P. R. Monge & Contractor, 2003). Few social phenomena can be fully explained by
a single causal mechanism operating at a single level of analysis. For the task of
acquiring a better understanding of the structure of relationships between human and
nonhuman actors in an organizational knowledge network, more than one theory is likely
to be useful.
Second, a multitheoretical perspective offers the opportunity to compare or
integrate complementary theories. Organizational learning theory and institutional theory
have some useful overlap between, and points of departure from, one another (H. Aldrich
& Ruef, 2006). Both are broadly concerned with a similar phenomenon, the influence of
social forces on changes in organizational knowledge and practices (Haunschild &
Chandler, 2008). The concepts of vicarious learning in organizational learning theory,
71
and institutional isomorphism in institutional theory, similarly examine the diffusion and
adoption of such knowledge and practices throughout organizational communities,
although the two concepts diverge in their accounting of organizations’ motives for
adoption. Vicarious learning emphasizes organizations’ desire to successfully compete
with one another, and isomorphism emphasizes organizations’ desire for legitimacy.
Scholarship on both theories also pays considerable attention to sources and consumers of
organizational knowledge, and how characteristics of both affect knowledge behaviors
and processes (H. Aldrich & Ruef, 2006).
At the same time, while some of the key concepts and explanatory variables
proposed by the two theories are similar, the level of analysis in each theory is often
different. As Aldrich and Ruef (2006, pp. 34-60) discuss, both theories are readily
applicable at multiple levels of analysis, and yet the scholarship on each has been
particularly focused on illuminating phenomena at distinct levels. Scholars have
employed organizational learning as a means for understanding more meso-level
processes within one organization or between several organizations, while institutional
theory has been typically used to provide a more macro view of the field and societal
levels. This difference in level of analysis indicates that the two theories could be used
together in complementary fashion, as Haunschild and Chandler (2008) have noted.
Additionally, institutional theory has been criticized for neglecting the micro level and
the microprocesses of organizational change that occur there (Felin, Foss, & Ployhart,
2015; Gondo & Amis, 2013; R. Greenwood, Oliver, Sahlin, & Suddaby, 2008; Powell &
Colyvas, 2008). Bechky (2011), for example, calls for more grounding of organizational
theories in the “individual action of people in organizations” (p. 1157). As a result of
72
these critiques, Suddaby and colleagues (2010) identify in institutional theory scholarship
an “emerging focus on the role of individual actors and their engagement with, and
reaction to, institutional pressures” (p. 1237).
This similarity in both theories’ relative lack of attention to the micro level
suggests that both could benefit from a similar remedy, the application of a sociomaterial
perspective. Sociomateriality, with its emphasis on how two different types of actors,
humans and artifacts, interact, and to what effect, often encourages or necessitates
attention to the micro level, and thus shows promise as a useful ontological companion to
both organizational learning theory and institutional theory. Further, neither theory has
given much consideration to the role that material artifacts play in the diffusion and
adoption of organizational knowledge and practices. Czarniawska (2008) notes that
technology has been mostly overlooked in institutional theory. In their review of
organizational learning scholarship, Argote and Miron-Spektor (2011) similarly observe
that “research on how tools affect knowledge creation and organizational learning is in its
infancy” (p. 1129), with organizational learning scholars typically focusing on how
knowledge is used and embodied by people and collectives of people, rather than on how
knowledge is embodied by artifacts and subsequently used by people.
Fourth, both organizational learning theory and institutional theory have proven
useful in explaining the empirical phenomenon of interest in this project, the structure of
relations in social networks, and more specifically, in organizational knowledge
networks. In their article about the intersection of the organizational learning and
institutional theory perspectives, Haunschild and Chandler (2008) note that both
perspectives are concerned with organizations’ interactions with their environments,
73
including their social networks. Extensive scholarship indicates that social networks play
an important role in organizational learning and in institutional isomorphism (for reviews,
see Owen-Smith & Powell, 2008; Van Wijk, et al., 2011).
There are also disadvantages, of course, to adopting a multitheoretical perspective
to investigate one’s research question. A key disadvantage is that for the breadth that is
achieved, depth is sacrificed. Because I used two theoretical perspectives with two
separate bodies of literature, and united them under a third ontology with its own distinct
literature, I necessarily skimmed over many aspects of each theory that are not directly
relevant to the task at hand. Devotees of each of these two theoretical literatures may be
dismayed that I did not apply all or even most propositions of each theory. While the
potential to explain the empirical phenomena and problems of interest may increase with
the use of multiple theoretical lenses, one’s ability to advance or further develop any
single theory may decrease. This is not a research project that aimed for theoretical depth.
I did not employ one theory exclusively and comprehensively, although I did attempt to
apply particular constructs and propositions from several different theoretical
perspectives faithfully.
Accordingly, I used organizational learning to investigate organizational
members’ communication load, which I conceptualized as one form of sociomaterial
agency influencing an organizational knowledge network. I used institutional theory to
investigate the legitimacy and credibility of human and nonhuman knowledge sources,
which I conceptualized as additional forms of sociomaterial agency. I focused on
communication load, legitimacy, and credibility because, as explained in Chapter 2, and
74
elaborated upon in Chapter 5, these three variables have been repeatedly cited as
important in empirical examinations of the communication of health care best practices.
Organizational Learning Theory and Communication Load
Organizational learning theory encompasses a range of propositions concerned
with knowledge and learning in organizational contexts (Argote & Miron-Spektor, 2011;
Huber, 1991; Levitt & March, 1988). As a phenomenon, organizational learning may be
defined as a “change in the organization’s knowledge that occurs as a function of
experience” (Argote & Miron-Spektor, 2011, p. 1124), with experience being either
directly acquired by an organization and its members or indirectly transferred from
organizations and people external to the focal organization. As a theory, organizational
learning is intended to address, as defined by Easterby-Smith and Lyles, “the learning
processes of and within organizations” (2011a, p. 3). The prepositional phrases are
important here: organizational learning encompasses both the learning processes of an
organization as a whole, and those of the people who work within the organization. The
prepositional phrase that is missing in this definition is also important: while
organizational learning scholars very often consider external environmental influences,
they typically place their spotlight, as noted above, on how learning processes occur
within and affect one organization or organizational subunit, rather than how learning
occurs between and affects multiple organizations.
The term “learning processes” in Easterby-Smith and Lyles’ (2011a) definition
also merits comment because it conveys two additional important characteristics of
organizational learning theory. First, as Easterby-Smith and Lyles observe, the primary
outcome of interest in organizational learning research is learning, a phenomenon that
75
occurs over time. As such, a study that examines organizational learning as an outcome
requires longitudinal data and a measure of some sort of change in experience or
knowledge (Argote & Miron-Spektor, 2011; Easterby-Smith & Lyles, 2011a). Second,
the term “learning processes” emphasizes that there are multiple sub-processes involved
in learning. Huber (1991) also uses the term “processes” in the title of his review of
organizational learning theory, referring to the “contributing processes” of organizational
learning—that is, the many concepts, facets, and steps involved in learning. In the present
research I examined cross-sectional data and therefore did not attempt to observe or
measure organizational learning as an outcome per se. Instead, I applied propositions
from organizational learning theory to one phenomenon, sociomaterial knowledge
communication, a “contributing process” that can lead to the eventual outcome of
organizational learning.
Huber (1991) identifies four primary processes of organizational learning:
knowledge acquisition, distribution, and interpretation, and organizational memory. Of
these, the present research focuses on knowledge acquisition and distribution—
essentially, the flow or communication of knowledge. One of the sub-processes of
knowledge acquisition that Huber (1991) describes in his review is “vicarious learning,”
in which organizations and their members observe other organizations and imitate
successful practices and routines. Another term in the literature that expresses the same
sort of concept as vicarious learning is knowledge transfer (Argote, et al., 2000; Argote,
McEvily, & Reagans, 2003; Argote & Miron-Spektor, 2011), which has been used to
refer to both intra- and interorganizational transfer (Easterby-Smith, Lyles, & Tsang,
2008; van Wijk, et al., 2008). As Miner and Mezias (1996) point out, best-practice
76
knowledge transfer and organizational benchmarking represent forms of vicarious
learning. The vicarious learning of one organization may lead to the field-level
isomorphism examined in institutional theory, a relationship that Miner and Mezias
(1996) note:
Understanding vicarious learning by individual organizations is especially
important because it can produce systematic population level outcomes. The
frequency and accuracy of vicarious learning has powerful implications for
patterns of behavior in populations or organizations. (p. 93)
Other scholars (R. Greenwood, et al., 2008; Huber, 1991; Levitt & March, 1988;
Spender, 1996) have also noted the similarity between the concepts of institutional
isomorphism and vicarious learning.
One of the essential claims that organizational learning theory makes about
vicarious learning—and, indeed, all processes associated with knowledge communication
in organizations—is that such communication is fraught with difficulty and messiness,
complicated by a variety of barriers, and generative of errors and unintended
consequences. As Stephens (2012) points out, scholars of organizational learning have
long been interested in factors that may cause learning and decision-making to occur in
non-optimal conditions, such as bounded rationality (Simon, 1957, 1979), sticky
information (Jensen & Szulanski, 2004; Szulanski, 1996; von Hippel, 1994), and
information and communication overload (Huber, 1991). Each of these concepts
expresses skepticism about organizational learning and points to factors that limit, rather
than facilitate, learning (Friedman, Lipshitz, & Popper, 2005).
77
In early work on organizational learning, the concepts of information overload
and communication load were discussed by Cyert and March (1963), Simon (1957,
1979), Galbraith (1977), and others in the context of how they affected, and
compromised, organizational decision-making (Case, 2008). Although the concept of
overload is briefly mentioned in some reviews of organizational learning scholarship,
particularly earlier review articles (Fiol & Lyles, 1985; Huber, 1991; Miner & Mezias,
1996), it is not featured in many other reviews (Argote, 2005; Argote, et al., 2003;
Easterby-Smith, et al., 2008; Levitt & March, 1988) or books on the subject (Easterby-
Smith & Lyles, 2011b), especially in comparison to similar concepts that identify
challenges and barriers to organizational learning, such as bounded rationality.
Since its appearances in the early organizational research of Carnegie School
theorists, the concept of information or communication load, or overload, has been of
interest and use to other scholars, however, particularly those working in information
science, human resource management, marketing, accounting, and communication. In the
research literatures of these fields, the concept has been referred to with various
combinations of the words “load” or “overload,” and “information” or “communication”
(Ballard & Seibold, 2006). The first distinction, between “load” and “overload,” has been
a relatively minor one in the literature, with researchers using both words tending to
focus on the perception of too much (rather than too little or an appropriate amount of)
information (Stephens, 2012). The second distinction, between “information” and
“communication,” also has been subtle, but somewhat more substantive.
When “information” has been used in the literature, scholars have tended to focus
solely on the characteristics of the information or knowledge in question, and particularly
78
on the quantity of the information. Scholars have defined information overload as
occurring “when people receive more information than they can process,” (Chen & Lee,
2013, p. 729). Bawden and Robinson (2009) define information overload as occurring
when “an individual’s efficiency in using information in their work is hampered by the
amount of relevant, and potentially useful, information available to them,” and when
“information received becomes a hindrance rather than a help, even though the
information is potentially useful” (pp. 182-183). Similarly, scholars interested in
decision-making and in viewing organizations as information processing systems
(Galbraith, 1974, 1977; Tushman & Nadler, 1978) have defined information overload as
occurring when the information supply or the information processing requirements—that
is, the amount of information that must be processed within a certain period of time—
exceed an individual’s information processing capacities (Eppler & Mengis, 2004; Karr-
Wisniewski & Lu, 2010).
When “communication” overload has been used in the literature, scholars have
tended to define overload more expansively, as encompassing characteristics of the
information as well as other content that can be communicated, such as emotion (Chen &
Lee, 2013; Cho, Ramgolam, Schaefer, & Sandlin, 2011); and as encompassing the social
context of knowledge communication, such as the attributes of the knowledge sources or
communication partners, and one’s relationships to them, in addition to characteristics of
the knowledge itself (Stephens, 2012). Scholars have defined “communication load” as
“a measure of the extent to which, in a given period of time, an organization’s members
perceive more quantity, complexity, and/or equivocality in the information than the
individual desires, needs, or can handle in the process of communication” (Ballard &
79
Seibold, 2006; Cho, et al., 2011; C. J. Chung & Goldhaber, 1991, p. 8). Communication
overload is “the rate and complexity of communication inputs to an individual” (Farace,
Monge, & Russell, 1977, p. 202), and occurs “when people feel overloaded by a vast
amount of complex communication input from diverse sources, multiple channels, with
rapid turnaround time, which can lead to stress and depression” (Chen & Lee, 2013, p.
729). Stephens (2012) conceptualizes communication overload sociomaterially, as being
constituted by material, individual, and social factors, and as a dynamic condition “that is
constantly re-negotiated between people and their environment” (p. 11). Because of its
more expansive definition and greater consideration of the social context of knowledge
communication, I used the term “communication load” in the present project.
Communication Load as a Form of Sociomaterial Agency
Empirically, as a number of accounts attest (Canary, 2011; Esterling, 2004;
Hansen & Haas, 2001; Nutley, et al., 2007), health care is but one of a number of
knowledge-intensive industries and fields characterized by an over-supply or “overload”
of best-practice and other forms of organizational knowledge. These accounts often link
the over-supply of this knowledge to material artifacts. For example, Hansen and Haas
(2001) note that “the relatively recent explosion of information available in electronic
forms makes attention, rather than information, the scarce resource in organizations” (p.
1). Similarly, Van Zandt (2004) observes that “the cost of physical communication
resources [e.g., the postal service, data networks] has fallen so much that the relatively
scarce resource is now the human attention needed to process and understand
information” (p. 542). Write Whelan and Teigland (2013), “With the increasing
processing power and plummeting costs of information and communication technologies,
80
the ability of employees to ubiquitously access and disseminate information grows.
However, emerging research shows that individuals are struggling to process information
as fast as it arrives” (p. 177).
At the same time, scholars have documented reports and perceptions of overload
dating from ancient times to the present day (Bawden & Robinson, 2009; Blair, 2003,
2010), suggesting that the mechanism at work here may not be simply particular artifacts
themselves, such as an overflowing e-mail in-box or a stack of reports and articles on a
desk, but also people’s responses and adaptations to ICTs. That is, the mechanism at
work may be sociomaterial, involving the interaction between and entanglement of
people and artifacts (Stephens, 2012). As Blair (2010) explains,
These days we are particularly aware of the challenges of information
management given the unprecedented explosion of information associated with
computers and computer networking. . . . But the perception of and complaints
about overload are not unique to our period. Ancient, medieval, and early modern
authors and authors working in non-Western contexts articulated similar concerns,
notably about the overabundance of books and the frailty of human resources for
mastering them (such as memory and time). The perception of overload is best
explained, therefore, not simply as the result of an objective state, but rather as the
result of a coincidence of causal factors, including existing tools, cultural or
personal expectations, and changes in the quantity or quality of information to be
absorbed and managed. (pp. 2-3)
Like Blair (2010), Stephens (2012), and others (e.g., Barley, et al., 2011), in this
dissertation I viewed communication load, or overload, as a form of sociomaterial agency
81
affecting the structure of knowledge networks. Recall from Chapter 1 that agency is the
ability of one actor to “modify a state of affairs by making a difference” (Latour, 2005, p.
71) in another actor’s actions. Also recall the observation that in the abstract, it may be
relatively easy to distinguish between human agency, the agency that people exercise;
and material agency, the agency that nonhuman things exercise. In an empirical network
of human and nonhuman actors, however, it is often difficult to disentangle the two types
of agencies when they are jointly engaged in relationships (e.g., knowledge
communication) that contribute to a particular social phenomenon (e.g., knowledge
dissemination, knowledge diffusion). Recall that in the example of gun violence, it is
difficult to blame this phenomenon solely on guns and material agency (à la “guns kill
people”) or solely on humans and human agency (à la “people kill people”), and that it
seems more reasonable to attribute the violence to the interaction of the two forms of
agency—in other words, to attribute it to sociomaterial agency. In the case of
communication load and its influence on the structure of knowledge communication, I
argue that the same holds true. Communication load is an influence on knowledge tie
formation that can be attributed both to people and knowledge artifacts. Following
similar sociomaterial conceptualizations of communication and information overload by
Stephens (2012) and Barley and colleagues (2011), here I conceptualize communication
load as a form of sociomaterial agency that influences the structure of relationships in a
knowledge communication network.
Returning to another example in Chapter 1, consider again the physician who
happens to hear a radio news report on her drive home from work about new best-
practice guidelines for the treatment of high cholesterol. Imagine that on this particular
82
day, she met with many patients, attended a staff meeting with colleagues, and waded
through an overflowing in-box of e-mails. As a result, the physician feels a high level of
communication load, and because of this sense of overload she decides to change the
radio station before listening to the full news report and learning any of the details about
the cholesterol guidelines. Here the decision not to form a “knowledge tie” with a
potential source of best-practice knowledge is influenced by the physician’s perception of
communication overload. One could conclude that this decision not to form a tie should
be attributed to the physician’s human agency, since she is the actor who is feeling the
communication overload. But given that her e-mail contributed to her sense of overload,
and that she is concerned that the radio news report will exacerbate this feeling, I would
argue instead that the decision not to form a tie should be attributed to the sociomaterial
agency of communication load—that communication load should be conceptualized as a
form of sociomaterial agency, rather than as a form of human agency only.
The Role of Communication Load in Sociomaterial Knowledge Networks
Empirical research has examined causes and effects of communication overload.
In the remainder of this chapter, I draw upon this work to develop three sets of
hypotheses about how communication load acts as a form of sociomaterial agency
affecting knowledge network structure.
A common element in the definitions of communication and information load
discussed above, as well as in empirical research on load, is that overload is related to the
quantity of communication or information received, both in terms of the frequency of
communication and the number and diversity of communication partners or knowledge
sources (Eppler & Mengis, 2004). For example, in their study of information overload
83
and the use of ICTs among high-tech workers, Barley and colleagues (2011) observed a
positive association between the time employees spent sending, processing, and
responding to e-mail and the degree to which they felt overloaded, independent of the
overall number of hours the employees spent on any type of work.
Research suggests that when the quantity of communication an actor experiences
becomes great enough that the actor begins to feel overloaded, he or she may experience
a number of negative consequences. Overload may adversely affect people’s ability to
interpret information that is communicated to them (Huber, 1991) and to make decisions
(Eppler & Mengis, 2004; O'Reilly, 1980). Overload may decrease productivity (Hansen,
2002; Karr-Wisniewski & Lu, 2010) and overall job performance (O'Reilly, 1980;
Oldroyd & Morris, 2012). Overload has also been associated with reduced self-esteem
(Chen & Lee, 2013), stress and burnout (Eppler & Mengis, 2004), and turnover (Oldroyd
& Morris, 2012; Soltis, Agneessens, Sasovova, & Labianca, 2013). These effects of
overload on individuals may ripple out to affect an entire organization when, for
example, the person experiencing overload is highly centralized in the organization’s
knowledge networks, and his or her overload results in a bottleneck in the flow of
organizational knowledge (Oldroyd & Morris, 2012). One or two individuals’ overload
also may affect an entire organization when those people experiencing overload are of
average centrality but their particular overload “symptoms,” such as stress and burnout,
become contagious and affect their colleagues as well.
The fact that communication overload puts a person at risk for a range of negative
consequences suggests that actors will avoid experiencing overload and, by implication,
will avoid the large quantity of communication that prompts feelings of overload. That is,
84
in order to prevent or mitigate overload and its detrimental effects, people will begin to
avoid or reject additional communication among existing or potential ties. Indeed,
research indicates that people who experience a high quantity of communication and who
feel overloaded are more likely to satisfice, reduce their information search or sharing
efforts, and/or avoid additional communication relationships entirely. In a recent
application of the same type of multilevel network analysis conducted in the present
study, Brennecke and Rank (in press) found that among research-and-development
workers in a large high-tech company, those who were members of a greater number of
project teams, requiring a greater number of communication relationships, were less
likely to seek or provide advice to their colleagues. This result suggests a negative
relationship between communication load and formation of additional knowledge ties.
Similarly, a meta-analysis of research on organizational knowledge transfer (van Wijk, et
al., 2008) found that in networks of organizations or organizational units, there was a
significant and negative relationship between an actor’s degree (number of relationships)
and knowledge acquisition. The study authors interpreted this result as suggesting that
organizations connected to a large number of knowledge sources experience difficulty
managing the quantity or diversity of those sources and thus avoid acquiring additional
knowledge.
The association between communication quantity and overload is further
supported by a separate but similar line of research on the concept of carrying capacity.
As Monge and colleagues (P. Monge, Heiss, & Margolin, 2008) note, there are
similarities between the concept of communication overload and the concept of carrying
capacity in organizational ecology and evolutionary theory. According to the ecological-
85
evolutionary perspective of organizational change, carrying capacity refers to the
maximum number of organizations a particular environmental resource niche can support
or “carry,” given finite resources (H. Aldrich & Ruef, 2006; Hannan & Freeman, 1989).
Monge and colleagues (2008) apply the concept of niche carrying capacity to networks of
organizations, proposing a “relational carrying capacity” that refers to the maximum
density, or number of relations, that can be sustained by a single node (i.e., an
organization) in a network, or by the network of organizations as a whole. Unlike the
literature on carrying capacity, the literature on information and communication overload
does not explicitly include the concept of a tipping point at which communication load
reaches the maximum level an actor can maintain, given his or her coping ability, before
it becomes overload. It seems reasonable, however, to equate an actor’s relational
carrying capacity in a communication network with the point at which an actor’s high
communication load becomes communication overload.
Modeling of empirical networks suggests that the relational carrying capacity of
human communication networks (e.g., e-mail networks) is significantly lower than some
other types of networks (e.g., academic citation networks) (Leskovec, Kleinberg, &
Faloutsos, 2007). Monge and colleagues (2008) propose that the lower relational carrying
capacity in human communication networks could be the result of nodes choosing not to
acquire new relations after a certain point because of diminishing returns, or of nodes
seeking to limit the effort associated with acquiring and maintaining additional relations.
In other words, nodes are likely to try to avoid reaching their maximum carrying capacity
or the point at which they experience communication overload.
86
In a related theoretical article, Oldroyd and Morris (2012) propose that the effects
of high social capital are curvilinear for “star” employees, defined as members of
organizations who demonstrate superior performance relative to their colleagues and who
are highly visible in the labor market. The authors propose that after a certain point for
such employees, increases in social capital begin to have a deleterious effect on
performance due to information overload, rather than continuing to have the traditionally-
hypothesized advantageous effect due to greater access to information and opportunities
(Burt, 1992). Such star employees with very high social capital have greater access to
knowledge sources and are also more likely to be sought after as knowledge sources
themselves, increasing the number of relationships they carry in a rich-get-richer cycle,
but also increasing their risk of reaching their relational carrying capacity and
experiencing communication overload. Oldroyd and Morris (2012) hypothesize that when
star employees become overloaded, they may cause bottlenecks in the flow of knowledge
in their organizations, and they may leave their organizations to escape the heavy
communication burden.
For the present project, longitudinal data were not collected, so the process by
which actors reach relational carrying capacity or communication overload could not be
observed. In a cross-sectional study, however, where perceived communication load is
conceptualized as an independent variable, and the structure of the knowledge
communication network is the dependent variable, a negative association between an
actor’s communication load and out-degree statistic (i.e., number of outgoing ties with
alters) may suggest that actors operate with a relational carrying capacity. This leads to
87
the first set of hypotheses about how communication load may affect the structure of a
sociomaterial knowledge network:
Hypothesis 1: Physicians with higher communication load will be less likely to
seek or receive best-practice knowledge from (a) their colleagues, (b) artifacts,
and (c) both colleagues and artifacts.
5
Scholars have found that in addition to the quantity of communication,
communication overload is also associated with the quality of communication, with
quality referring to novelty, complexity, ambiguity, tacitness, and relevance (Eppler &
Mengis, 2004). People tend to perceive overload both when they are confronted with a
large amount of knowledge being communicated, and when that large amount of
knowledge is more novel, more complex, more ambiguous, more tacit, and of uncertain
relevance. This suggests that in order to avoid communication overload, workers will
avoid forming ties with knowledge sources of lower quality—that is, whose knowledge
communication will in some way be more ambiguous, more tacit, less relevant, etc.
Pursuing this aspect of communication load, scholars have investigated which
types of knowledge sources are most likely to be perceived as being of lower quality and
thus as less desirable communication partners. In particular, researchers have focused on
the conditions under which human versus nonhuman knowledge sources will be more or
less desirable to actors wishing to prevent or mitigate overload.
5
Note that each part (i.e., 1a, 1b, and 1c) of the first hypothesis represents the hypothesis being tested at a
different level of the multilevel sociomaterial network, where each “level” corresponds to a different type
of node. There is one level for physician-to-physician relations, a second level for physician-to-artifact
relations, and a third level that considers both types of ties simultaneously. Chapter 5 elaborates on the
multilevel conceptualization I used for the network analysis. For now, it is sufficient to note that this
multilevel analysis necessitated the third part of the above hypothesis, despite its seeming redundancy with
the first two parts. The same holds true for Hypotheses 6 and 8 proposed in Chapter 4.
88
Numerous observers have identified artifacts, particularly newer ICTs such as the
Internet, as advantageous sources of knowledge because they possess features that enable
professionals to easily and quickly access a multitude of knowledge. Scholars have also
identified features of human knowledge sources that may be viewed as disadvantageous,
and circumstances in which a person might prefer to obtain knowledge from an artifact
rather than a coworker. For example, members of organizations may prefer nonhuman
knowledge artifacts under conditions of time and geographic constraints, which research
indicates make knowledge communication with human sources more difficult (Fulk,
Heino, Flanagin, Monge, & Bar, 2004). Haas and Hansen (2007) observed that teams of
management consultants who relied on colleagues as knowledge sources, rather than
electronic documents, required more time to complete projects. Nonhuman artifacts may
also be preferred when communication with human knowledge sources involves high
levels of reciprocity and relationship maintenance (Fulk, et al., 2004; Hansen, 2002).
Other potential costs of accessing human knowledge sources include loss of face or
status, as in when knowledge seekers are embarrassed about their lack of knowledge or
about the type of knowledge they seek; and the generation of negative affect, as when
knowledge seekers are forced to ask for knowledge from a colleague whom they dislike
(Borgatti & Cross, 2003; Casciaro & Lobo, 2008; Nebus, 2006).
On the other hand, reports raise questions about the utility and effectiveness of
ICTs and other artifacts for knowledge communication (e.g., Corman & Dooley, 2011;
Flanagin, 2002; Flanagin & Bator, 2011). As Canary and McPhee (2011) note, “many
organizational practitioners have lionized technology as the panacea for all problems
related to knowledge management, use, and sharing, without taking into account social
89
processes that work concomitantly with technology in organizations” (pp. 10-11).
Research indicates that knowledge novelty and relevance can be put into context by a
person’s colleagues, and those colleagues can also help a person make sense of
knowledge complexity, ambiguity, and tacitness, in part because communication with
another human can allow for synchronous question-and-answer interaction (Deng &
Poole, 2011; Flanagin, 2002; Flanagin & Bator, 2011; Kane & Alavi, 2007; Su &
Contractor, 2011). Peer-to-peer communication may also afford knowledge-seekers the
opportunity to learn best-practice knowledge from colleagues with in-the-trenches
experience who operate in similar work contexts, rather than from experts or institutional
actors operating from afar. Peer-to-peer interaction may afford knowledge-seekers the
ability to acquire best-practice knowledge from other people whom they trust or to whom
they feel emotionally close, rather than from Web sites, paper reports, or other artifacts.
Studies report that professionals prefer human knowledge sources, especially those with
whom they have close relationships, when the knowledge sought is more tacit (Deng &
Poole, 2011; Hansen, 1999, 2002) or more complex (Byström, 2002; Hansen & Haas,
2001; Su & Contractor, 2011). It may be more difficult for knowledge seekers to find
highly specific knowledge tailored to their particular circumstances using knowledge
artifacts, and easier using human knowledge sources. As Jackson and Williamson (2011)
observe about the advantages of human sources over knowledge repositories,
Knowledge drawn from a repository may be valid, but it lacks the freshness of
knowledge taken from “the source”—the originator of the documentation, or a
current practitioner of a particular skill. It lacks context, currency, and, at times,
applicability. (pp. 54-55)
90
Additionally, many researchers have speculated on or actually observed a causal
link between the affordances and constraints of various types of artifacts—particularly
new technologies—and perceptions of overload (Bawden & Robinson, 2009; Eppler &
Mengis, 2004; Stephens, 2012). For example, use of highly synchronous communication
channels (e.g., face-to-face, telephone, videoconference, instant messages) as well as
asynchronous (e.g., Web sites, e-mail, memos) media have been positively associated
with communication overload, with asynchronous media having the stronger association,
perhaps because these artifacts do not allow for feedback and interaction and may
therefore increase message ambiguity or complexity (Cho, et al., 2011). Communication
artifacts that “push” communication, delivering information without a user actively
seeking it (e.g., e-mail, television), may contribute more to overload than “pull” artifacts
over which users may feel they have more control (Bawden & Robinson, 2009). Feelings
of loss of control over knowledge communication, and the interruptions in work this type
of “push” communication may cause, have been associated with perceptions of overload
(Bawden & Robinson, 2009; Oldroyd & Morris, 2012; Stephens, 2012; Su, et al., 2010).
E-mail, in fact, has been particularly singled out as contributing to overload, and a
subset of the overload literature has investigated “e-mail overload” specifically (Barley,
et al., 2011; Dabbish, Kraut, Fussell, & Kiesler, 2005; Soucek & Moser, 2010; Whelan &
Teigland, 2013). Barley and colleagues (2011) concluded that among employees holding
a variety of positions in a high-technology company, e-mail was widely recognized as a
“cultural symbol” of the overload many experienced, to the extent that this singling-out
of e-mail caused employees to overlook other causes of overload at work. Some scholars
have even conceptualized “technology overload” to be a condition in its own right,
91
distinct from but related to information overload and communication overload (Karr-
Wisniewski & Lu, 2010).
Of course, a person’s knowledge communication efforts are often, by choice or
necessity, not strictly limited to colleagues only or to artifacts only, but rather involve ties
to both types of sources. On balance, research on the association between knowledge
quality and communication load suggests that all else being equal, people are likely to
prefer receiving at least some of their best-practice knowledge from a colleague, even if
they also access artifact sources for that knowledge. Research suggests that professionals
prefer human knowledge sources even when they have easy access to sophisticated
knowledge artifacts (Case, 2008; Cross, Borgatti, & Parker, 2001; Cross & Sproull,
2004), and especially when the knowledge being sought is complex and/or tacit, as was
the case for the cholesterol treatment guidelines of interest in the present study.
6
For
example, in a study of R&D engineers in two medical device firms, Whelan and Teigland
(2013) found that the engineers frequently relied on colleagues who acted as “scouts” or
brokers of external knowledge to identify, filter, and introduce new knowledge into the
network. Research specifically on health care also suggests that physicians tend to prefer
human knowledge sources (Gorman, 1995).
Taken together, the above research suggests that in order to prevent or mitigate
communication overload caused by poor communication quality, workers (regardless of
their current degree of communication load) will exhibit a preference for using human
knowledge sources over artifact sources. This preference may impact the structure of
sociomaterial knowledge networks in several ways. First, it may cause a pattern of
6
Chapter 5 provides a detailed description of the complexity and tacitness of the cholesterol treatment
guidelines.
92
brokerage to emerge in the network, such that workers use colleagues who are connected
to artifact sources as brokers to those sources, rather than directly connecting with
artifacts themselves:
Hypothesis 2a: Among physician sources, those with ties to one or more artifacts
(i.e., those who are potential brokers to one or more artifacts) will be more likely
to be sources of best-practice knowledge for their colleagues.
Hypothesis 2b: Among physician sources, those with ties to artifacts (i.e., those
who are potential brokers to artifacts) will be more popular sources of best-
practice knowledge for their colleagues.
Second, the preference for human over nonhuman knowledge sources may cause
a pattern of closure to emerge in sociomaterial knowledge networks, such that when
workers do access an artifact source, they may seek out knowledge communication from
a colleague who is also connected to that same artifact source. Such a colleague may be
able to contextualize and provide relevance for the knowledge from the artifact source,
and through conversation help make the knowledge less complex and ambiguous. For
example, in the study of the R&D engineers cited above, Whelan and Teigland (2013)
found that engineers used an electronic knowledge management system established by
their company in the following way: they searched the system for a report containing the
information they needed, but then also approached the report author to obtain a filtered,
customized, version of that information. Therefore:
Hypothesis 3: Physicians who seek or receive best-practice knowledge from one
or more artifacts will tend to have ties to colleagues who are also connected to
those same artifacts.
93
To summarize, the first two sets of hypotheses propose that in order to prevent or
mitigate communicate overload, (1) workers with high communication load will be less
likely to access best-practice knowledge from any type of source, and (2) workers
(regardless of current level of communication load) will exhibit a preference for ties to
human sources, choosing to access artifacts indirectly when brokerage via a colleague is
possible, or directly when closure via a colleague also connected to the same artifact is
possible.
A final set of hypotheses about communication load focuses on which types of
artifacts workers may prefer when they do access nonhuman knowledge sources. I argue
that among artifact knowledge sources, workers will tend to access mass media artifacts
rather than niche media artifacts in order to prevent or mitigate communication overload.
For the purposes of this project, mass media artifacts are those targeting a broad, general
audience not limited to physicians. Niche media artifacts are those targeting a narrower
audience of professionals—in the present case, U.S. physicians for whom the treatment
of high cholesterol is part of their practice.
7
I draw on two lines of research to support the proposition that accessing
knowledge from mass media artifacts may generate less communication load and
therefore will be a preferred knowledge communication behavior among workers. The
first is the network literature on multiplexity. As reviewed in Chapter 2, multiplexity
7
Note that in the communication literature of the past two decades, scholars (e.g., Bennett & Manheim,
2006; Chaffee & Metzger, 2001) have questioned use of the distinction between “mass” and “niche” media,
given contemporary changes in ICTs and the media economy, and use of the term “mass media” entirely.
For example, Bennett and Manheim (2006) write that “one change [in media and society] is that
conventional mass media reach smaller audiences, while niche media attract increasing numbers, making it
harder to send effective generalized messages but easier to target specialized appeals” (p. 218). While
acknowledging that the traditional distinction between the two categories may no longer be valid, and that
media channels once labeled as “mass” may now reach a far narrower audience, here I used the two
categories simply to differentiate between artifacts targeting a professional physician audience, and artifacts
targeting a much more general, “lay” audience.
94
refers to the presence of more than one type of tie between two actors, and has been a
frequently observed structural signature in the literature on organizational knowledge
networks (e.g., Bell & Zaheer, 2007; Coleman, et al., 1966; Ingram & Roberts, 2000;
Lazega & Pattison, 1999; Morrison, 2002; Saint-Charles & Mongeau, 2009; Tsai, 2002).
Multiplexity is common in knowledge networks because, all else being equal, (1) people
often prefer to acquire knowledge from sources for which they feel some degree of
familiarity, positive affect, and trust, and (2) people often prefer to minimize knowledge
search costs and maximize convenience by accessing knowledge from actors with whom
they already share a tie (Granovetter, 1985; McEvily, Perrone, & Zaheer, 2003; Nebus,
2006; Uzzi, 1996, 1997).
Therefore, applying the research on relational multiplexity to workers’ use of
knowledge artifacts, we might expect workers to prefer to access best-practice knowledge
from artifacts to which they are already connected, even if the purpose of the first type of
tie may not have been to access best-practice knowledge. Since most people are
“connected to” or use mass media artifacts that provide entertainment or news about
public affairs, it seems reasonable to postulate that workers may sometimes encounter
work-related best-practice knowledge via mass media artifacts such as television
programs and newspaper articles. Given that news about recent medical advances and
treatment recommendations is frequently covered in mass media artifacts, physicians are
likely to learn about health care best-practice knowledge from these artifacts while they
also use them to stay up-to-date with world news or to be entertained. And, given their
potential to experience communication overload, physicians may learn about best-
practice knowledge from mass media artifacts more frequently than from niche media
95
artifacts, because of time constraints or the tendency to settle for a convenient source that
they are already using for another purpose. For example, a physician may not have time
to actively seek out knowledge from a medical journal, but he may encounter the
knowledge while watching the nightly news on television.
The second line of research supporting the hypothesized preference for
knowledge ties with mass media artifacts is the literature on the sister concepts of
satisficing, the principle of least effort, and the cognitive miser, primarily found in the
fields of organizational learning, information science, and cognitive science. Satisficing
(Simon, 1955, 1956) refers to when information seekers do not “make every possible
attempt to attain the most complete, accurate, and detailed information available
(optimizing) but rather gather just enough data, opinions, and impressions to feel satisfied
with the process” (Case, 2008, p. 34). The principle of least effort (Zipf, 1949) proposes
that people usually opt to expend the least amount of effort possible on a task, including
an information-seeking task. The term cognitive miser (Fiske & Taylor, 2013) refers to
“the necessary stinginess with which attention and processing time are often allocated to
stimuli in the real world,” (p. 206). All three concepts recognize the limitations in
people’s capacity to search for and process information, and emphasize that people often
prioritize adequate solutions over optimal ones. If physicians are able to access
knowledge about best-practice guidelines without intentionally or actively searching for
it, but rather simply passively encountering it when watching television, listening to the
radio, or reading the newspaper, the concepts of satisficing, least effort, and cognitive
miser suggest that they will prefer do so over purposefully seeking knowledge about the
best practice from other artifacts.
96
Building on the literature on network multiplexity and the research on satisficing,
the principle of least effort, and cognitive misers, I propose the third set of hypotheses
about how communication load affects the structure of organizational knowledge
networks. The first of these tests the simpler proposition that physicians will tend to
prefer mass media artifacts over niche media artifacts, and the second tests the
proposition that overloaded physicians will be particularly likely to exhibit this
preference:
Hypothesis 4a: Physicians will be more likely to seek or receive knowledge from
mass media, rather than niche media, artifacts.
Hypothesis 4b: Physicians with higher communication load will be even more
likely to seek or receive knowledge from mass media rather than niche media
artifacts.
In summary, I hypothesized that faced with the specter of communication
overload in everyday work life, workers who are already experiencing overload will
avoid accessing best-practice knowledge from additional sources, both human and
artifact (Hypothesis 1); that when they do access knowledge sources, workers will prefer
human sources over artifact sources (Hypotheses 2 and 3); and that when they do use
artifact sources, they will prefer mass media artifacts over niche media artifacts
(Hypothesis 4). In proposing these hypotheses, I used organizational learning theory and
related research to illustrate how communication load may act as one form of
sociomaterial agency influencing the structure of organizational knowledge networks, by
limiting the formation of knowledge ties or affecting workers’ choice of knowledge
communication partners. In Chapter 4, I draw on institutional theory to examine how
97
legitimacy and credibility may represent two additional forms of sociomaterial agency
affecting knowledge networks.
98
CHAPTER 4: INSTITUTIONAL THEORY AND THE SOCIOMATERIALITY
OF LEGITIMACY AND CREDIBILITY
Like Chapter 3, this chapter is divided into three main sections: the introduction
of the theory and the key theoretical concepts applied here; a description of how the
theory is integrated with sociomateriality, and the development of the resulting
hypotheses. The first section provides a brief overview of institutional theory, focusing
on the concept of legitimacy. It describes the relationship between legitimacy and the
concept of credibility, and explains why both are used here. These two concepts are often
defined similarly but are based in research literatures that rarely intermingle. The second
section outlines the conceptualization of legitimacy and credibility as forms of
sociomaterial agency that, like communication load, influence knowledge network
structure. The third section presents four sets of hypotheses, as well as a research
question, regarding specific structural signatures expected as a result of the influences
legitimacy and credibility exert on the network. A final hypothesis considering the effects
of all three forms of sociomaterial agency on the overall network is presented at the
conclusion of the chapter.
Institutional Theory and Legitimacy
In institutional theory, an institution is “the product of common understandings
and shared interpretations of acceptable norms of collective activity” (Suddaby, et al.,
2010, p. 1235), and is “infused with value beyond the technical requirements of the task
at hand” (Selznick, 1957, p. 17), such that it eventually assumes myth-like status (Meyer
& Rowan, 1977). Knowledge and practices that have become institutionalized are
“typically taken-for granted, widely accepted and resistant to change” (R. Greenwood, et
99
al., 2008, p. 6). When an organization and its members adopt an institution, they gain
“legitimacy, resources, stability, and enhanced survival prospects” (Meyer & Rowan,
1977, p. 340).
In the present empirical setting, I identified the institution of interest as health
care best-practice knowledge, packaged in the form of clinical guidelines for physicians.
Health care best-practice knowledge has become institutionalized over time by private
and government health insurers’ reimbursement of medical care that follows best-practice
guidelines, by professional societies’ creation and endorsement of such guidelines, and
by the advocacy for and adoption of such guidelines by health care organizations seen as
industry leaders. The cholesterol treatment guidelines that receive attention here are not
the institution itself, but one manifestation of the broader institution.
DiMaggio and Powell (1983) argue that organizations become similar to one
another because of their desire to gain legitimacy among various publics, a process called
institutional isomorphism. DiMaggio and Powell identify three isomorphic processes:
coercive, mimetic, and normative. Coercive isomorphic processes, such as government
regulation, occur when politically influential actors motivate organizations to adopt
practices using the threat of sanctions. Mimetic isomorphic processes, such as imitation
of successful competitors or adoption of practices advocated by consulting firms or
heralded in the mass media, are responses to uncertain environmental conditions.
Normative isomorphic processes are derived from professionalization, via formal
education, professional organizations, and socialization in the workplace. Organizations
take certain forms and adopt certain practices via these isomorphic processes, not
necessarily because these forms or practices are technically appropriate, efficient, or
100
empirically proven, but because they conform to socially acceptable notions of what is
appropriate and legitimate. Following these norms increases organizations’ legitimacy,
which in turn increases their likelihood of survival (DiMaggio & Powell, 1983; Meyer &
Rowan, 1977). But this comes at a cost, as Meyer and Rowan (1977) note: as institutional
rules are incorporated as “myths” within an organization, they may conflict with actual
organizational activity, and thus a phenomenon of “decoupling” occurs. Formal
organizational structure, now determined by myth, becomes detached from day-to-day
organizational activity, resulting in reduced organizational efficiency. Thus institutional
isomorphism may both help and hurt organizations: “Organizational conformity to the
institutional environment simultaneously increases positive evaluation, resource flows,
and therefore survival chances, and reduces efficiency” (Meyer & Scott, 1983, p. 141).
Institutional theory “evolved as an antidote to the overly rationalist and
technocratic perspectives” (R. Greenwood, et al., 2008, p. 29) in organizational analysis.
Greenwood and colleagues (2008) argue that the central question of institutional theory
examines why organizations adopt knowledge and practices that “defy traditional rational
explanation” (p. 31). In this way institutional theory is similar to organizational learning
theory (Haunschild & Chandler, 2008), with its insistence that although organizations
may adopt knowledge and practices for initially rational, profit-maximizing motives,
rationality is bounded and thus organizational learning and change is a process fraught
with difficulty and unintended consequences. According to institutional theory,
organizations’ actions are often motivated by institutional rules, social norms, and the
desire for legitimacy, as opposed to objectives of economic efficiency (DiMaggio &
Powell, 1983; Meyer & Rowan, 1977). The knowledge and practices that gain legitimacy
101
and are eventually institutionalized are not necessarily those that are the most efficient
and effective in the strictly rational, profit-maximizing sense of classical economic
theory.
While communication load has received a relatively modest amount of attention
in the organizational learning literature, legitimacy is a central idea in institutional theory
and appears in a majority of papers utilizing the theory (Colyvas & Powell, 2006;
Deephouse & Suchman, 2008; Haveman & David, 2008). According to Suchman’s
(1995) widely-cited definition, legitimacy is “a generalized perception or assumption that
the actions of an entity are desirable, proper, or appropriate within some socially
constructed system of norms, values, beliefs and definitions” (p. 574). Legitimacy is a
social evaluation of an entity’s right to exist (Knoke, 1985; Meyer & Scott, 1983); and of
its adherence to laws, social norms and collective values (Ashforth & Gibbs, 1990;
Dowling & Pfeffer, 1975; Fiol & O'Connor, 2006). Organizations, knowledge, and
practices with very high legitimacy are accepted, taken for granted, and unquestioned by
society (Deephouse & Carter, 2005; Meyer & Scott, 1983). In other words, they become
institutions.
In institutional theory, legitimacy is of great consequence because a key
theoretical proposition is that organizations are often motivated by the desire to attain
legitimacy, in order, in turn, to increase the likelihood of organizational survival
(DiMaggio & Powell, 1983, 1991; Meyer & Rowan, 1977; Suddaby, et al., 2010).
Legitimacy increases the likelihood of an organization’s survival by increasing the
public’s willingness to accept its practices and actions, even when they are controversial
and violate norms (Tost, 2011; Tyler, 2006). For example, in a study of public reactions
102
to organizations embroiled in controversy, Elsbach (1994) observed that organizational
spokespersons’ accounts of the controversies were judged to be more effective by
audiences when those accounts referenced institutional characteristics of the organization,
such as its legitimate, socially normative practices and structures, and were judged to be
less effective when they referenced technical characteristics, such as the organization’s
efficiency and effectiveness.
Institutional legitimacy is viewed as having two dimensions: a sociopolitical
dimension and a cognitive dimension (H. E. Aldrich & Fiol, 1994; Bitektine, 2011;
Deephouse, 1996; Suchman, 1995). Sociopolitical legitimacy refers to an organization’s
social “desirability and normativity” (Deephouse, 1996, p. 1025). An example of an
organization with low sociopolitical legitimacy is one that has experienced a controversy
or crisis, such as the U.S. company Johnson & Johnson when cyanide was discovered in
bottles of Tylenol in the early 1980s. Cognitive legitimacy refers to the degree to which
an organization is perceived as comprehensible, as taken-for-granted, and as having a
right to exist (Ruef & Scott, 1998; Suchman, 1995). An example of an organization with
low cognitive legitimacy is one that falls into a novel category, with little history or
prevalence in terms of its organizational form (Bitektine, 2011), such as the Internet
search engines Yahoo and Excite in the mid-1990s. Some scholars have conceptualized
and found empirical evidence for a temporal relationship between the two dimensions,
with cognitive legitimacy emerging only after sociopolitical legitimacy has been
established (Bitektine, 2011; Colyvas, 2007; Royston Greenwood, Suddaby, & Hinings,
2002; Tost, 2011), or conversely, with cognitive legitimacy as a base upon which other
forms of legitimacy are built (Ruef & Scott, 1998).
103
Within the dimension of sociopolitical legitimacy, scholars have identified three
subtypes.
8
The first is regulative legitimacy, which refers to the perception of an
organization’s conformity to laws, regulations, and standards created by governments,
credentialing associations, and professional bodies (W. R. Scott, 2013; Tost, 2011;
Zimmerman & Zeitz, 2002). The second is moral legitimacy (sometimes referred to as
normative legitimacy), which refers to the perception of an organization’s general benefit
to society (Suchman, 1995; Tost, 2011). Favorable media coverage and the presence of
interorganizational alliances are two common sources of moral legitimacy for
organizations. The third type of sociopolitical legitimacy is pragmatic legitimacy, which
refers to the perception of an organization’s benefit to specific constituencies (Bitektine,
2011; Suchman, 1995; Tost, 2011).
In the present research project, measurement of both sociopolitical and cognitive
legitimacy, as well all three subtypes of sociopolitical legitimacy, was not possible due to
the risk of high respondent survey burden. As a result, I chose to exclude cognitive
legitimacy and pragmatic legitimacy from investigation. There were several reasons why
I focused on sociopolitical rather than cognitive legitimacy. First, cognitive legitimacy
frequently has been applied to the evaluation of entire organizational forms, rather than to
specific organizations, which are of interest here. In this study, I was concerned with the
legitimacy of the organizations that authored the cholesterol best-practice guidelines, or
that produced artifactual sources of knowledge about the guidelines. Second, because
8
Scholars have articulated other typologies for sociopolitical legitimacy and for legitimacy in general (see,
for example, Kostova & Zaheer, 1999; Ruef & Scott, 1998; Suchman, 1995). The typology discussed
here—with sociopolitical and cognitive as the two dimensions, and regulative, moral, and pragmatic as the
three subtypes of sociopolitical legitimacy—is the most common and has inspired the greatest amount of
consensus to date, based on my review of the literature. Note also that these three subtypes of sociopolitical
legitimacy, rather confusingly, overlap somewhat, but not entirely, with DiMaggio and Powell’s (1983)
three isomorphic processes (coercive, mimetic, and normative).
104
organizations that authored the cholesterol best-practice guidelines, and that produced the
majority of artifactual knowledge sources (newspapers, academic journals, etc.) about the
guidelines, were relatively well-established, it was possible that I would not find
significant variance in these organizations’ perceived cognitive legitimacy.
There were also several reasons why I focused on the regulative and moral
subtypes of sociopolitical legitimacy, and excluded pragmatic legitimacy. Pragmatic
legitimacy is not as commonly recognized in the literature as the other two subtypes.
Additionally, its basis in the self-interest of a specific constituency conflicts somewhat
with the general conceptualization of legitimacy as an indicator of widespread social
desirability and collective social norms.
Although institutional legitimacy is conceptualized as an attribute of an
organization or organizational practice, it is not endogenous to the organization or
practice but rather is a judgment or evaluation made by others about that organization or
practice. As Suchman (1995) explains, legitimacy “represents a relationship with an
audience, rather than being a possession of the organization” (p. 594). The identity of the
actors making legitimacy judgments is therefore important. Scholars have identified the
source(s) of legitimacy judgments generally as “social actors” (Deephouse, 1996, p.
1025); the “environment” (Kostova & Zaheer, 1999, p. 64); “broader publics” (Rindova,
Pollock, & Hayward, 2006, p. 55); “the general public and relevant elite organizations”
(Knoke, 1985, p. 222); stakeholders (Bitektine, 2011); and opinion leaders (H. E. Aldrich
& Fiol, 1994). More specific sources of legitimacy have been identified as the media
(Cornelissen, Durand, Fiss, Lammers, & Vaara, 2015; Deephouse, 1996); regulators and
government officials (H. E. Aldrich & Fiol, 1994; Deephouse, 1996); and professional or
105
trade associations (R. Greenwood, et al., 2008; Zimmerman & Zeitz, 2002). In line with
these source attributions, sociopolitical legitimacy has typically been measured
retrospectively and indirectly in empirical studies, using degree of public acceptance of
an organization or its practices as reflected in popular media; amount of government
subsidies an organization or industry receives; presence of interorganizational relations,
such as interlocking directorates and strategic alliances; or simply widespread adoption of
a practice or survival of an organization over time (H. E. Aldrich & Fiol, 1994; Strang &
Soule, 1998; Zimmerman & Zeitz, 2002). (For some empirical examples of these
measurement approaches, see Deephouse, 1996; Deephouse & Carter, 2005; Tolbert &
Zucker, 1983.)
If, as discussed above, legitimacy is usually conceptualized as a characteristic of
an organization or an organizational practice, it is an attribute of a collective entity or an
artifact, rather than an attribute of a person. Organizations and their knowledge and
practices may be perceived as more or less legitimate, but legitimacy does not usually
apply to the individual members of organizations. There has been a small amount of
research in institutional theory examining how an organization’s founders, top
management teams, and other representatives may be subjects of legitimacy evaluations
and may be considered as more or less legitimate (Deephouse & Suchman, 2008;
Suchman, 1995). However, such research typically examines these people as not having
legitimacy themselves, but as conferring (or failing to confer) legitimacy upon their
organization. Little if any research has explicitly examined individual members of
organizations as more or less legitimate (for an exception, see Huy, Corley, & Kraatz,
2014).
106
In this way, institutional theory’s usage of legitimacy departs from the etymology
of the word “legitimacy,” in that legitimacy was originally applied at the individual level,
to people, to describe a child or a monarch: a legitimate child was born to parents
lawfully married to each other, and a legitimate monarch was one who acquired the
throne by hereditary right (L. E. Brown, 1993). Institutional theory’s application of
legitimacy to organizations and practices but not to people is notable here for two
reasons. First, it distinguishes legitimacy from its conceptual cousins, discussed in more
detail below. In contrast to legitimacy, similar concepts such as status, reputation, and
credibility are commonly applied to both organizations and individuals.
Second, it is important to note that while institutional theory considers legitimacy
judgments inapplicable to people, a number of institutional scholars make a point to
recognize that such judgments do not simply exist “out there” in the ether of social life,
but rather that it is individual people whose opinions form a collective legitimacy
judgment about a particular organization or organizational practice (Ashforth & Gibbs,
1990; Bitektine, 2011; Colyvas & Powell, 2006; Tost, 2011). Further, a number of
scholars argue that at the microlevel, communication, both formal and informal, sustains
legitimacy judgments and institutions (Cornelissen, et al., 2015; Lammers & Barbour,
2006; Suchman, 1995). The growing body of work on the microprocesses and
microfoundations of institutional theory, discussed in Chapter 3, has directed attention to
the role of individual actors in the institutional processes that affect organizations. As
Suddaby and colleagues (2010) point out, “institutional pressures exist only to the degree
that internal and external participants believe in them and engage in the institutional work
necessary to perpetuate them” (p. 1235). That is, there would be no institutions and
107
institutionalization without individual people making legitimacy judgments. More
researchers thus have begun to consider legitimacy from the standpoint of the actors who
make the judgments about an organization’s legitimacy, from what Bitektine (2011)
describes as the “evaluator’s perspective” (p. 151). As Tost (2011) explains,
Institutional theorists have paid relatively little theoretical or empirical attention
to the intraindividual dynamics of legitimacy judgments (i.e., the content,
formation, and change of the judgments themselves). While legitimacy is
ultimately a collective-level phenomenon, an understanding of the microlevel
dynamics of legitimacy judgments is crucial because individuals’ judgments and
perceptions constitute the ‘micro-motor’ (Powell & Colyvas, 2008) that guides
their behavior, thereby influencing interactions among individuals, which, in turn,
coalesce to constitute collective-level legitimacy and social reality. Therefore, an
understanding of the individual-level dynamics of legitimacy judgments can help
scholars to better understand not only the dynamics of institutional change but
also the critical role that individuals play in those change processes. (p. 687)
Despite the growing scholarly interest in the microprocesses of legitimacy, few
empirical studies have directly asked individuals about their perceptions of organizational
legitimacy, or examined how such perceptions are formed (Bitektine, 2011). In his
seminal article on legitimacy, Suchman (1995) notes that measurement of legitimacy has
been problematic and suggests one remedy may be to investigate the concept via
audience or constituent surveys. Bitektine and Haack (2015) also suggest that researchers
use survey research to understand individuals’ legitimacy judgments and the
microprocesses of institutional legitimacy. Echoing these sentiments, Vergne (2011)
108
argues for the use of psychometric studies of legitimacy, to elucidate the concept of
legitimacy and to parse the distinctions between legitimacy and related concepts such as
status, reputation, and credibility. Building on these observations, and the growing body
of work on microprocesses in institutional theory, I chose to measure legitimacy in the
present project by asking study participants—that is, physicians working in a health care
provider organization—for their legitimacy judgments about the organizational authors of
best-practice guidelines, and for their legitimacy judgments of the artifacts, such as
journal and newspaper articles, that they used as sources of knowledge about the
guidelines.
A final aspect of institutional legitimacy of relevance to the present project is the
handful of concepts related to but distinct from legitimacy. Scholars have identified
legitimacy as one type of evaluation made in organizational contexts; similar types of
evaluations include status and reputation, as well as credibility. Examination of the
distinctions that have been made between these concepts is helpful for understanding the
definition and role of legitimacy in institutional theory, and in the present project, for
describing the rationale for conceptualizing both legitimacy and credibility as forms of
sociomaterial agency influencing a sociomaterial knowledge network. A thorough
discussion of the similarities and differences between these concepts is beyond the scope
of this dissertation.
9
Here I briefly distinguish legitimacy from two common bedfellows
in organizational theory, status and reputation. I then compare legitimacy with credibility,
which hails from the communication and persuasion literatures.
9
For more comprehensive reviews, see Bitektine (2011); Deephouse and Carter (2005); Deephouse and
Suchman (2008); Devers and colleagues (2009); Tost (2011).
109
Status refers to an accepted, agreed-upon ranking based on an actor’s membership
in a particular group, and reflects an actor’s position relative to other actors (Deephouse
& Suchman, 2008; Washington & Zajac, 2005). A status judgment addresses the question
of how an actor fits into a ranked order of similar actors (Bitektine, 2011). Reputation is
an expectation about an actor’s future behavior based on collective perceptions of past
behavior (Deephouse & Suchman, 2008), and can be assessed on any attribute along
which the actors in question may vary (Deephouse & Carter, 2005). A reputation
judgment addresses the question of how an actor will behave in the future, relative to
other actors (Bitektine, 2011). In contrast to status and reputation, a sociopolitical
legitimacy judgment addresses the questions of whether the actor adheres to laws and
social norms, and whether the actor is beneficial to society (Bitektine, 2011).
Additionally, status and reputation are segregating evaluations, in that they emphasize
differences between actors and are used by evaluators to identify these differences;
whereas legitimacy is a homogenizing evaluation, in that it emphasizes the conformity of
an actor to social norms and ideals, and is used by evaluators to gauge an actor’s degree
of adherence to these norms and ideals (Deephouse & Carter, 2005; Deephouse &
Suchman, 2008; Devers, et al., 2009). All three types of judgments are similar, however,
in that they are inherently collective, social judgments that cannot be formed in a
vacuum, absent a broader social context. In this way, all three differ from credibility, as
described in more detail below.
Credibility
While legitimacy has been studied in the context of organizational theory and
specifically institutionalism, credibility has been studied in the context of communication
110
and persuasion, with little cross-pollination between the two sets of disciplines and
concepts. As Flanagin and Metzger (2008) note, credibility was first investigated
systematically by psychologists interested in persuasion against the backdrop of
propaganda during the world wars. Credibility judgments were identified as a key part of
the persuasion process (Wathen & Burkell, 2002). Credibility is defined as an individual
evaluation of the “believability” (Flanagin & Metzger, 2008, p. 8) of a source. Scholarly
consensus identifies two dimensions of credibility, expertise and trustworthiness
(Flanagin & Metzger, 2008; Hovland, Janis, & Kelley, 1953; Pornpitakpan, 2004).
Expertise refers to the degree to which the source is qualified, based on knowledge and
experience, to provide valid information (Hovland, et al., 1953; Pornpitakpan, 2004;
Rhoads & Cialdini, 2002). Trustworthiness refers to the degree to which the source is
motivated, based on honesty and lack of bias, to provide valid information (Hovland, et
al., 1953; Pornpitakpan, 2004; Rhoads & Cialdini, 2002). A number of scholars also
argue that there is a third dimension of credibility, goodwill, representing the degree to
which the source understands and has empathy for audience members (McCroskey &
Teven, 1999; Perloff, 2010). While I agree with this argument, I omitted the goodwill
dimension of credibility in the present project to reduce respondent burden.
Early research on credibility conceptualized the contexts in which credibility
judgments are made as those in which only one source is evaluated—an audience
member listening to a politician delivering a campaign speech, for example, or watching
a television news anchorperson report the nightly news, with the politician and the
anchorperson, respectively, being the sole source to be judged. In contrast, contemporary
research has increasingly recognized the “multilevel” nature of credibility, that is, that
111
audience members may simultaneously evaluate the credibility of the multiple “sources”
that contribute to a particular unit of information or knowledge. Scholars argue that in the
case of a newspaper op-ed, for example, a reader may judge the credibility of the
information provided in the op-ed (message credibility), the credibility of the op-ed
author (source credibility), the credibility of the newspaper that published the op-ed
(sponsor credibility), and/or the credibility of newspapers in general (medium credibility)
(Flanagin & Metzger, 2008; Wathen & Burkell, 2002). In this sort of scenario, research
suggests that the reader may or may not make distinct credibility judgments for the
different levels of sources, depending on the nature of the reader’s attribution of the
source (Flanagin & Metzger, 2007, 2008; Rieh & Danielson, 2007). Additionally, the
reader’s credibility judgment for one level of source may influence his or her credibility
judgment for another level of source. In a study of Web site credibility, Westerwick
(2012) found that judgments of Web site sponsor credibility positively affected
judgments of Web site message credibility. Others have observed a similarly positive
empirical relationship between sponsor credibility and message credibility (Flanagin &
Metzger, 2007; Hilligoss & Rieh, 2008; Metzger, Flanagin, & Medders, 2010; Sundar,
2008).
Credibility differs conceptually from legitimacy, status, and reputation not only
because it originates from different disciplines and literatures, but also because it is
fundamentally an individual evaluation, rather than a social one. Credibility is an
individual evaluation because although it may be influenced by others’ opinions, a
credibility judgment is the evaluator’s own assessment of source believability and reflects
the evaluator’s unique relationship and history with the source (Rieh & Danielson, 2007).
112
Only one evaluator is needed for a credibility judgment to be formed. Legitimacy is a
social evaluation because although the evaluators may be individuals, a legitimacy
judgment is the evaluator’s assessment of the degree to which an organization or
organizational practice adheres to collectively-held norms and values (regardless of
whether the evaluator herself believes in these norms and values) (Colyvas & Powell,
2006). More than one evaluator is needed for a legitimacy judgment to be formed. As
Suchman (1995) explains, “when one says that a certain pattern of behavior possesses
legitimacy, one asserts that some group of observers, as a whole, accepts or supports
what those observers perceive to be the behavioral pattern, as a whole—despite
reservations that any single observer might have about any single behavior” (p. 574).
Credibility is not socially conferred; it may draw upon social norms, and be influenced by
the credibility evaluations of others, but it is not fundamentally normative. A person can
make a credibility judgment in a vacuum, whereas that is not possible with legitimacy (or
with reputation or status). Therefore, it is possible for a person to judge a knowledge
source as legitimate but not credible—that is, as a source that adheres to social norms but
that the evaluator personally judges to lack expertise and be untrustworthy.
Why use credibility in addition to legitimacy in the present project, and
incorporate credibility under the umbrella of institutional theory? Why not investigate
credibility only, for example, and omit consideration of legitimacy? The rationale for
using legitimacy was that institutional theory offers a thorough theoretical account of the
diffusion and adoption of best-practice knowledge, whereas the research on credibility,
while providing strong support for credibility as an important antecedent in knowledge
source evaluation, does not provide an overarching theoretical account of knowledge
113
diffusion. On the other hand, one might ask, why not investigate legitimacy only, and
omit consideration of credibility? The rationale for using credibility was that legitimacy
is not conceptualized as being applicable to human knowledge sources. One must
consider both the legitimacy of nonhuman sources and the credibility of nonhuman and
human sources in order to conceptualize both humans and nonhumans as equivalent
actors influencing knowledge communication in a sociomaterial network.
In addition, there was an ancillary theoretical benefit to using both legitimacy and
credibility in this project. The two concepts are clearly at least anecdotally similar,
suggesting that scholarly efforts to formally, substantively relate one to the other would
help expand our understanding of both. In the institutional theory literature, scholars
often define legitimacy using the term “credibility,” depicting the latter as distinct from
but contributing to the former. For example, Aldrich and Fiol (1994) observe that novel
organizational forms typically lack legitimacy because they “lack the familiarity and
credibility that constitute the fundamental basis of [social] interaction” (p. 647). Suchman
(1995) writes that “part of the cultural congruence captured by the term legitimacy
involves the existence of a credible collective account or rationale explaining what the
organization is doing and why” (p. 575). Further, in the communication and persuasion
literatures, researchers have identified the concept of sponsor credibility—the credibility
of the organizational or institutional author of a particular message—in a way that seems
parallel to the concept of institutional legitimacy.
However, very little research has actually investigated similarities and differences
between legitimacy and credibility, and scholars examining one concept often seem
unaware of related work conducted on the other concept. In a recent investigation of
114
institutional legitimacy, for example, Huy and colleagues (2014) remark that the concept
of credibility “has received very little scholarly attention relative to legitimacy” (p. 1654).
This observation is true if one limits one’s consideration to the management literature
only, but is inaccurate if one considers the sizeable credibility scholarship in the
communication and persuasion literatures. The only two studies I found that substantively
investigated both concepts together were published by political science and public policy
scholars: Mondak (1990) observed a positive association between source credibility and
policy legitimacy in a study of U.S. Supreme Court decisions. Cash and colleagues
(2003) argued that scientific information must be simultaneously salient, credible, and
legitimate in order to influence public policy. Scholarly understanding of both legitimacy
and credibility could be enhanced, therefore, by research that makes a substantive
distinction between the two concepts.
Legitimacy and Credibility as Forms of Sociomaterial Agency
There is a small but substantive body of work on how artifacts are imbued with
legitimacy, driven in part by organizational scholars’ enthusiasm for institutional theory
over the past three decades. Scholars theorize that artifact knowledge sources “retain an
imprint” (D'Adderio, 2011, p. 199), “inscription” (D'Adderio, 2011; Latour, 1992), or
“institutional residue” (Kaghan & Lounsbury, 2006, p. 259) of the organizations and
people who created them or who are most closely associated with them. This inscription
carries a degree of legitimacy reflective of the degree to which the organization and
people who created the artifact adhere to laws, social norms, and collective values
(Ashcraft, et al., 2009; Bechky, 2003; Czarniawska, 2008; Fiol & O'Connor, 2006;
Kaghan & Lounsbury, 2006). Further, Czarniawska (2008) notes, institutions are not only
115
codified into material artifacts, but the very act of codifying them into artifacts further
legitimizes them, making them seem, as Czarniawska puts it, “impossible to question” (p.
775).
For example, in her study of the role of artifacts in the work of members of a
manufacturing company, Bechky (2003) observed that one type of artifact, design
drawings of machines created by engineers, was imbued with a high level of legitimacy
associated with the engineering profession. In comparison, a second type of artifact, the
machines themselves, built by technicians and assemblers, were imbued with a lower
level of legitimacy associated with the technician and assembler professions. Both the
drawings and the machines were representations of knowledge, but members of the
company viewed the drawings as a more legitimate representation because of the greater
legitimacy of the engineering profession within the organization. In a second example,
Fiol and O’Connor (2006) examined how U.S. physicians in the early 1900s began
wearing the white laboratory coats associated with scientists, in order to confer the
legitimacy of scientists upon themselves. Finally, in a third example, this time in the
context of health care best practices, Avorn (2013) argues that guidelines for breast
cancer screening authored by the U.S. Centers for Disease Control and Prevention
(CDC), an agency of the federal government, are imprinted with greater legitimacy than
guidelines for breast cancer screening authored by the American Medical Association, a
for-profit professional association for physicians. To Avorn, the CDC guidelines carry
greater legitimacy because he believes the CDC, a public agency, is less susceptible to
being influenced by commercial interests than the American Medical Association. For
116
him, the legitimacy of the CDC is inscribed onto the artifacts—in this case, best-practice
guidelines—that it produces.
There is also a small but substantive body of work on how artifacts are imbued
with credibility, driven in part by credibility scholars’ interest in how audiences perceive
various material features of new ICTs. Empirical research indicates that audiences
perceive the material features of knowledge sources, such as the design elements of Web
pages, as signals conveying more or less credibility (Flanagin & Metzger, 2007; Fogg et
al., 2001; Fogg et al., 2003; Wathen & Burkell, 2002). For example, in a focus group
study, Metzger and colleagues (2010) found that two common heuristics used to assess
web site credibility were professional (versus amateur) design quality and the absence (or
presence) of spelling and grammatical errors.
Given the above scholarship, it is not difficult to make the case that the legitimacy
and credibility of artifact knowledge sources have sociomaterial agency, and that
artifactual legitimacy and credibility may exert influence over how knowledge and
practices are communicated, adopted, or neglected in sociomaterial organizational
networks. It may be more difficult, however, to understand how the credibility of human
knowledge sources represents a form of sociomaterial agency—why wouldn’t a person’s
credibility be a form of solely human agency? What is material about a person’s
credibility?
Recall the argument, introduced in Chapter 1, that knowledge itself is never solely
a product of human or social agency and never solely a product of artifact or material
agency—it is sociomaterial. In the empirical context of the present project, the human
knowledge sources examined—physicians and their colleagues—possess knowledge
117
about the new cholesterol best-practice guidelines that originates in part from artifacts,
from guidelines published in a journal or summarized on a Web site. As empirical
research has shown (Whelan & Teigland, 2013), people at work who regularly monitor
the artifactual sources of knowledge associated with their professions, such as the
newsletters, journals, and Web sites of professional associations, tend to become known
among their colleagues as better sources of knowledge. In this way, their credibility as a
knowledge source is in part material, derived from their association with material
artifacts. In the same manner that an organization may derive legitimacy as a knowledge
source from various material trappings—its new office furniture, the sleek business cards
it provides its members, etc., a person may derive credibility as a knowledge source from
material trappings as well, from his or her association with various material artifacts.
D’Adderio (2011) explains: “the skills and capabilities of actors are mediated and
fundamentally transformed by the capabilities of the tools and instruments that they use
in their work” (p. 210). This argument is similar to Veblen’s (1899/2007) classic
observation that consumer artifacts carry an imprint of the social and cultural status of
their manufacturers, and are purchased not only for their functional utility but also for the
status they may confer upon consumers.
In this dissertation, therefore, I viewed the legitimacy and credibility of
knowledge sources as forms of sociomaterial agency affecting the structure of knowledge
networks. Again, recall that agency is the ability of one actor to “modify a state of affairs
by making a difference” (Latour, 2005, p. 71) in another actor’s actions, and that in an
empirical network of human and nonhuman actors, it is often difficult to disentangle
human and material agencies when they are jointly engaged in relationships that
118
contribute to a particular social phenomenon, such as the communication of knowledge.
Just as it is difficult to blame gun violence solely on guns or solely on people, it is
difficult to attribute the influences that legitimacy and credibility exert on knowledge tie
formation solely to material agency or solely to human agency.
How might knowledge source legitimacy and credibility influence—or in
Latour’s words, “make a difference” in—the previously-discussed hypothetical scenario
of the physician listening to the radio on her commute home from work? Recall that the
physician happens to hear a news report about new best-practice guidelines for the
treatment of high cholesterol. Imagine that she listens to this radio station during her
commute because she views it as a news source with a high level of legitimacy; she
believes that people in her social networks, community, and the general public perceive
this radio station as conforming to social norms, rules, and values. She also listens to the
station because she views it as having a high level of credibility. In her personal
experience, the station’s qualifications and trustworthiness as an information source are
high.
Because of the physician’s favorable perceptions of the legitimacy and credibility
of the radio station as a source of knowledge, she continues to listen to the station when
the news report about the new cholesterol treatment guidelines is broadcast. Additionally,
when she hears that the American College of Cardiology and the American Heart
Association are the organizational authors of the cholesterol guidelines, she feels further
motivated to pay attention to the news report, because she believes that the legitimacy
and credibility of these organizations are also high. The physician’s perceptions of the
legitimacy and credibility of a material artifact, the radio station news report, and that of
119
the organizational authors of the guidelines, influence her decision to listen to the news
report and to use it as a source of knowledge about the new best practice.
When the physician arrives at work the next morning, she greets the colleague
with whom she shares an office and recalls the radio news report. Because her colleague
attends professional conferences more frequently than anyone else in the office, and
maintains subscriptions to several of the leading medical journals, the physician views
him as a highly credible source of best-practice knowledge. She decides to ask him if he
has heard about the new cholesterol guidelines and formed an opinion about them. Here,
the colleague’s association with material knowledge sources influences the physician’s
perception of his credibility, and in this way the colleague’s credibility acts as a form of
sociomaterial agency influencing the physician to form a knowledge tie with him.
By conceptualizing both nonhuman and human knowledge sources as exercising
agency through a sociomaterial ontology, we can better understand how knowledge is
imbued with legitimacy and credibility. The structure of many knowledge networks likely
reflects the influences that knowledge sources’ legitimacy and credibility exert on
communication relationships, just as the structure of such networks also likely reflects the
influence that knowledge consumers’ communication load exerts on such relationships.
The Roles of Legitimacy and Credibility in Sociomaterial Knowledge Networks
In one sense, institutional theory and the empirical scenario of best-practice
knowledge communication in health care are an odd couple. In their classic article on
institutional theory, DiMaggio and Powell (1983) ask why organizations are so similar,
and use institutional theory as an explanation for this similarity. Many scholars and
practitioners interested in public health and health care ask opposite questions, that is,
120
why are health organizations and the health services they provide so different from each
other, and how can we disseminate best practices to make these organizations and their
services more similar.
In other ways, however, institutional theory is a logical tool with which to address
questions about health organizations and health care, and a number of scholars have
applied institutional theory to this empirical context (e.g., Barbour, 2010; Barbour &
Lammers, 2007; Battilana, 2011; Davidson & Chismar, 2007; Dunn & Jones, 2010;
Gondo & Amis, 2013; W. R. Scott, Reuf, Mendel, & Caronna, 2000; Westphal, et al.,
1997). Institutional theory offers a perspective with which to understand how conformity
to industry standards and best practices is emphasized and valued in health care and
public health. Barbour (2010) points out that legitimacy motivations are often at play in
health professionals’ seeking of best-practice knowledge from interorganizational
networks, and in their operationalization of best practices within their own organizations.
Legitimacy judgments are often used by audiences to evaluate health care best-practice
knowledge, as Guallar and Laine (2014) explain in their discussion of why clinical best-
practice guidelines are often controversial:
Attacks on guidelines raise several common criticisms. Critics may question the
legitimacy of the guideline committee [i.e., the group of experts who author the
guidelines]. When a primary care group, such as the USPSTF [U.S. Preventive
Services Task Force] issues a guideline, subspecialists often label the
recommendations unsound because the panel did not include subspecialty experts.
This criticism can be vicious when recommendations call for more parsimonious
use of interventions from which subspecialists profit. . . . Conversely, when
121
subspecialists issue guidelines, generalists sometimes gripe about the lack of
primary care perspective. When guidelines focus on [clinical research] trials,
critics call for recommendations to consider observational data and clinical
experience. The presence of conflicts of interest is another common source of
controversy, with claims that recommendations are designed to fill the pockets of
those who would profit from the interventions advocated. (p. 361)
The scholarship on institutional theory, legitimacy, and credibility thus offers a
useful theoretical lens with which to investigate the communication of best practices in a
health care organization. This scholarship provided the base from which I derived four
sets of hypotheses about how knowledge source legitimacy and credibility act as forms of
sociomaterial agency affecting the structure of best-practice knowledge communication
among physicians in an organizational network.
In the context of the communication of organizational knowledge, institutional
theory suggests that the greater the legitimacy of the knowledge and/or knowledge
source, the more likely it is to be communicated within and among organizations.
Organizational knowledge and practices that are viewed as legitimate and that have
become taken-for-granted ways of doing business “reduce the cognitive load associated
with decisions, as well as decrease risk by providing well-rehearsed modes of
communication and action and ready-made categories for resolving uncertainties”
(Colyvas & Powell, 2006, p. 311; for a similar discussion, see Zimmerman & Zeitz,
2002). Therefore we might expect that among best-practice knowledge seekers in
organizational networks, the more legitimate a source of knowledge is perceived to be,
the more likely the source is to be used.
122
Empirical research supports this positive association between knowledge source
legitimacy and knowledge communication. For example, in an analysis of the diffusion of
corporate governance practices among firms in the 1980s, Davis and Greve (1997)
observed that the practice that was perceived by corporate board members as more
legitimate, the poison pill, diffused much more quickly than the practice perceived as less
legitimate, the golden parachute. In a study of the use of an electronic database system
for internal knowledge-sharing in a management consulting firm, Hansen and Haas
(2001) found that the electronic knowledge source documents that were used most by
consultants were those that were perceived to hold knowledge of greater quality and
focus, and consequently, had a better reputation. Similarly, in a study of corporate
benchmarking, Still and Strang (2009) found that an alter organization’s prestige was a
key influence on the decision by the focal organization to adopt the alter’s best practice.
Accordingly, in the following two of hypotheses, I propose a positive association
between the formation of a communication tie in a sociomaterial knowledge network and
the perceived legitimacy of the knowledge source. I defined knowledge source as
encompassing both artifacts that contain best-practice knowledge, such as Web sites and
newspapers; and the organizational authors of the best-practice knowledge, in this case
the American College of Cardiology and the American Heart Association.
Hypothesis 5: Physicians will be more likely to seek or receive knowledge from
artifacts with greater legitimacy.
Hypothesis 6: Physicians who perceive the legitimacy of a best-practice author to
be greater will be more likely to seek or receive best-practice knowledge from (a)
their colleagues, (b) artifacts, and (c) both colleagues and artifacts.
123
By extension, if greater legitimacy increases the likelihood of knowledge
communication ties, greater credibility should also increase this likelihood. Indeed,
scholars have found that credibility is a primary criterion for the selection and use of
information (Flanagin & Metzger, in press; Wathen & Burkell, 2002). In her synthesis of
credibility research, Pornpitakpan (2004) found that almost all of the main-effect results
of credibility indicate that a knowledge source with high credibility is more persuasive in
changing both attitudes and behavior than a source with low credibility. High credibility
sources are particularly influential when the knowledge consumer is less involved in
knowledge search, selection, and use; and when the knowledge of interest is quantitative,
complex, novel, technical, or associated with risk (Pornpitakpan, 2004).
Credibility has been positively associated with knowledge transfer in
organizational contexts (Ko, Kirsch, & King, 2005), as has the related concept of trust. A
meta-analysis of the knowledge transfer literature concluded that based on its large effect
size relative to other variables, trust was one of the most important network antecedents
of knowledge transfer (van Wijk, et al., 2008). In their study of management consultants,
Su and Contractor (2011) found a positive association between the perceived expertise of
human knowledge sources and a consultant’s seeking of knowledge from that source;
expertise is a dimension of credibility. Borgatti and Cross (2003) similarly found that an
organizational member’s decision to seek information from a colleague was predicted by
the seeker’s positive evaluation of the source’s expertise.
Based on the literature on credibility and knowledge communication, I developed
the following two hypotheses, congruent with the previous two hypotheses regarding
legitimacy:
124
Hypothesis 7: Physicians will be more likely to seek or receive knowledge from
(a) colleagues and (b) artifacts with greater credibility.
Hypothesis 8: Physicians who perceive the credibility of a best-practice author to
be greater will be more likely to seek or receive best-practice knowledge from (a)
their colleagues, (b) artifacts, and (c) both colleagues and artifacts.
In my approach to conceptualizing the legitimacy and credibility of human and
nonhuman knowledge sources, I treated the two concepts as distinct, proposing separate
hypotheses for the legitimacy and credibility of knowledge artifacts, and for the
legitimacy and credibility of organizational best-practice authors. I did so in accordance
with the theoretical literature on legitimacy and credibility, which indicates the former is
a social evaluation and the latter is an individual evaluation, and that each concept is
composed of different dimensions. Empirically, however, whether the legitimacy and
credibility of knowledge artifacts and authors are actually distinct, separate concepts,
with the theoretical distinctions translating into empirically observable differences, is an
open question. One alternative possibility, for example, is that sponsor credibility is
empirically equivalent to legitimacy. Research strongly suggests that one of the key
cognitive heuristics used to evaluate credibility is an authority heuristic, which refers to
whether a source conveys official authority (Hilligoss & Rieh, 2008; Metzger, et al.,
2010; Sundar, 2008). This authority heuristic used for credibility judgments seems
similar to legitimacy. Because of the lack of empirical research on the distinctions
between legitimacy and credibility, I developed the following research question:
Research Question 1a: For artifact knowledge sources, is there an empirical
difference between legitimacy and credibility, as manifested in the influences of
125
artifact source legitimacy and credibility on presence of knowledge ties with
physicians?
Research Question 1b: For authors of best practices, is there an empirical
difference between legitimacy and credibility, as manifested in the influences of
author legitimacy and credibility on presence of knowledge ties with physicians?
Lastly, I developed a hypothesis considering the effects of all three forms of
sociomaterial agency—communication load, legitimacy and credibility—on the overall
knowledge network.
Hypothesis 9: The multilevel sociomaterial model will explain more variance in
knowledge network structure than the unipartite social model alone or the
bipartite material model alone.
Table 1 (see page 126) presents a summary of the hypotheses and research question
developed in this chapter and in Chapter 3.
126
Table 1. Research Hypotheses and Corresponding Model Parameters
Hypothesis/Research Question Net Parameter MPNet Name Visualization
Communication Load
H1a
Physicians with higher communication load will
be less likely to seek/receive knowledge from
their colleagues.
A
Attribute-based
sender activity
Load_SenderA
H1b
Physicians with higher communication load will
be less likely to seek/receive knowledge from
artifacts.
X
Attribute-based
sender activity
Load_XEdgeA
H1c
Physicians with higher communication load will
be less likely to seek/receive knowledge from
both colleagues and artifacts.
M
Attribute-based
sender
centralization
Load_Star2AXSender
H2a
Among physician sources, those with ties to one
or more artifacts (i.e., those who are potential
brokers to one or more artifacts) will be more
likely to be sources of knowledge for their
colleagues.
M
Affiliation-based
popularity
In2StarAX
H2b
Among physician sources, those with ties to
artifacts (i.e., those who are potential brokers to
artifacts) will be more popular sources of
knowledge for their colleagues.
M
Affiliation-based
popularity
AAinS1X
•
•
•
127
Hypothesis/Research Question Net Parameter MPNet Name Visualization
H3
Physicians who seek/receive knowledge from one
or more artifacts will tend to have ties to
colleagues who are also connected to those same
artifacts.
M
Affiliation-based
homophily/closure
TXAXarc, ATXAXarc
•
•
•
H4a
Physicians will be more likely to seek/receive
knowledge from mass media, rather than niche
media, artifacts.
X
Attribute-based
receiver popularity
Mass Media_XEdgeB
1
H4b
Physicians with higher communication load will
be even more likely to seek/receive knowledge
from mass media artifacts.
X
Attribute-based
sender-receiver
interaction
Load-Mass Media_XEdgeAB
1
Legitimacy & Credibility
H5
Physicians will be more likely to seek/receive
knowledge from artifacts with greater legitimacy.
X
Attribute-based
receiver popularity
Artifact Legitimacy_XEdgeB
128
Hypothesis/Research Question Net Parameter MPNet Name Visualization
H6a
Physicians who perceive the legitimacy of a best-
practice author to be greater will be more likely
to seek/receive best-practice knowledge from
their colleagues.
A
Attribute-based
sender activity
ACC Legitimacy_SenderA
AHA Legitimacy_SenderA
H6b
Physicians who perceive the legitimacy of a best-
practice author to be greater will be more likely
to seek/receive best-practice knowledge from
artifacts.
X
Attribute-based
sender activity
ACC Legitimacy_XEdgeA
AHA Legitimacy_XEdgeA
H6c
Physicians who perceive the legitimacy of a best-
practice author to be greater will be more likely
to seek/receive best-practice knowledge from
both colleagues and artifacts.
M
Attribute-based
sender
centralization
ACC Legitimacy_Star2AXSender
AHA Legitimacy_Star2AXSender
H7a
Physicians will be more likely to seek/receive
knowledge from colleagues with greater
credibility.
A
Attribute-based
receiver popularity
Physician Credibility_ReceiverA
H7b
Physicians will be more likely to seek/receive
knowledge from artifacts with greater credibility.
X
Attribute-based
receiver popularity
Artifact Credibility_XEdgeB
H8a
Physicians who perceive the credibility of a best-
practice author to be greater will be more likely
to seek/receive best-practice knowledge from
their colleagues.
A
Attribute-based
sender activity
ACC Credibility_SenderA
AHA Credibility_SenderA
H8b
Physicians who perceive the credibility of a best-
practice author to be greater will be more likely
to seek/receive best-practice knowledge from
artifacts.
X
Attribute-based
sender activity
ACC Credibility_XEdgeA
AHA Credibility_XEdgeA
129
Hypothesis/Research Question Net Parameter MPNet Name Visualization
H8c
Physicians who perceive the credibility of a best-
practice author to be greater will be more likely
to seek/receive best-practice knowledge from
both colleagues and artifacts.
M
Attribute-based
sender
centralization
ACC Credibility_Star2AX Sender
AHA Credibility_Star2AX Sender
RQ1a
For artifact sources, is there an empirical
difference, as manifested in likelihood to have
ties to physicians, between legitimacy and
credibility?
X — — —
RQ1b
For authors of best practices, is there an empirical
difference, as manifested in likelihood to
influence physicians’ ties to colleague and artifact
sources, between legitimacy and credibility?
A,
X,
M
— — —
Overall Sociomaterial Model
H9
The multilevel sociomaterial model will explain
more variance in knowledge network structure
than either the unipartite social model or the
bipartite material model.
A,
X,
M
— — —
Note. In the visualizations, physicians are represented by blue squares, and artifacts by red circles.
130
CHAPTER 5: RESEARCH DESIGN
Having introduced the ontology and empirical context for this project, and derived
hypotheses about how three forms of sociomaterial agency may affect the structure of an
organizational knowledge network, I turn in this chapter to the methods used to test these
hypotheses. First, I describe the specific study population and provide a more detailed
account of the best-practice knowledge of interest here. I then summarize the data
collection and data analysis procedures.
Study Population
Data were collected using a survey distributed to all of the 185 primary care
physicians employed at a non-profit multi-specialty medical group, hereafter referred to
as Alpha Medical Group. At the time of data collection, Alpha was one of six non-profit
medical groups that comprised a large organization providing health care in a
metropolitan area of the United States. Alpha offers outpatient care in 35 medical
specialty areas, of which primary care is one, and is affiliated with a number of teaching
and community hospitals in the area. It employs approximately 600 physicians and 3,700
other medical professionals and administrative and managerial employees, who together
provide care for 500,000 adults and children across 21 clinical locations in the
metropolitan area. Each of the primary care physicians in the study population worked in
one of 17 of these clinical locations (four of the organization’s clinical locations do not
offer primary care).
Primary care is defined as providing four main services: acting as the first point of
access for a person seeking medical care; providing long-term, continuous care;
providing comprehensive care focusing on the entire person, rather than a specific
131
disease; and coordinating care across the health system (Starfield, 1998). A primary care
physician for adults is generally considered, in the U.S. health care context, to be a
physician who has trained in one of two medical specialties, family medicine or general
internal medicine (Starfield, Shi, & Macinko, 2005), and who has received one of two
educational degrees, MD (doctor of medicine) or DO (doctor of osteopathy). In the
present research, primary care physicians with any of these specialties and degrees were
included in the study population. Primary care physicians were selected for the study
population for several reasons. First, in Alpha Medical Group, primary care physicians
comprise the largest category of physicians with the same professional training and
clinical job responsibilities, and therefore consisted of the largest population of potential
study participants. Second, among physicians in the United States, primary care
physicians are burdened (or blessed, depending on one’s perspective) with a particularly
large number of clinical best-practice guidelines applicable to their work, given the
generalist nature of their clinical responsibilities (Bodenheimer, 2006; Østbye, et al.,
2005; Shaneyfelt & Centor, 2009; Shaneyfelt, et al., 1999; Yarnall, et al., 2003). Third,
primary care is one of the main medical specialties responsible for monitoring a person’s
blood cholesterol levels and providing treatment for high cholesterol, the subject of the
best-practice guideline studied in the present research (and discussed further in the next
section).
Depending on the reference group to which one compares it, Alpha Medical
Group may differ from other health care provider organizations in the United States in at
least two ways of relevance to the communication of clinical best practices: its
organizational form, and the health care variations endemic to the geographic region in
132
which it is located. With regard to organizational form, there are a handful of different
types of health care organizations that provide primary care in the United States: solo
practices, small group practices, large group practices such as Alpha, federally-funded
community health centers for people with low incomes, hospitals (via outpatient or
emergency departments), and “integrated care systems” such as Kaiser Permanente that
combine health care provision with health insurance (Bodenheimer & Pham, 2010;
Hsiao, Cherry, Beatty, & Rechsteiner, 2010). The majority of primary care is provided in
solo or small group practices; only 7% of primary care physicians work in group
practices with 11 or more physicians of any specialty (Bodenheimer & Pham, 2010;
Hsiao, et al., 2010). Therefore, as members of a large multi-specialty health care provider
organization with approximately 600 physicians, primary care physicians at Alpha may
have access to different sources of best-practice knowledge, and communicate about
clinical best practices in different ways, as compared with those working in a solo
practice or a very large integrated care system.
In addition to form, Alpha Medical Group may differ from other health care
provider organizations in the United States in terms of the health care practice variations
endemic to its geographic location. Research indicates that health care practices vary
significantly by geographic location across the United States, such that two people who
live in different regions of the country, but who have the same diagnosis and similar
demographic characteristics and health histories, may receive very different care (Fisher
et al., 2003a, 2003b; Sirovich, Gottlieb, Welch, & Fisher, 2006). Alpha is located in a
region in which the volume of health care services provided to patients is among the
highest in the country, after controlling for geographic differences in health care pricing
133
and the distribution of disease (Fisher, Goodman, Skinner, & Bronner, 2009). Simply by
virtue of its location in a metropolitan area, Alpha also operates in a geographic region
with a much greater supply of primary care physicians; across the United States, the ratio
of primary care physicians to the general population in urban areas is more than double
that in rural areas (Council on Graduate Medical Education, 2007). As with the
differences in organizational form among health care providers, these geographic
variations in health care practices may mean that the primary care physicians at Alpha
Medical Group learn and communicate about guidelines for health care best practices
differently than their professional counterparts working in other regions of the United
States.
Best-Practice Knowledge
This study asked primary care physicians about a particular set of clinical best
practices, those governing the treatment of high blood cholesterol levels in adults in order
to reduce the risk of cardiovascular disease caused by atherosclerosis. Clinical best-
practice guidelines on this topic were published in November 2013 (Stone et al., 2013,
2014) by two organizations that shared authorship responsibilities: the American College
of Cardiology (ACC), the primary professional organization for cardiologists in the
United States, and the American Heart Association (AHA), a U.S. charitable organization
providing a range of services to both patients and health care professionals.
The ACC/AHA cholesterol guidelines were selected for this study for several
reasons. First, hypercholesterolemia and its potential deleterious effects are common and
serious health concerns addressed by primary care physicians, and thus best-practice
guidelines regarding their prevention and treatment are highly relevant to the work of
134
virtually all primary care physicians. High cholesterol is a cause of atherosclerotic
cardiovascular disease, which is the leading cause of death and largest contributor to
medical expenses in the United States (Go et al., 2013). Cholesterol levels can be lowered
through diet, exercise, and medication. Medications to lower cholesterol levels, called
statins, are the second most common class of drugs given or prescribed by physicians
during ambulatory care office appointments (behind analgesics such as aspirin and
ibuprofen) (Hsiao, et al., 2010). Use of such medications for the prevention and treatment
of atherosclerosis was the focus of the 2013 ACC/AHA cholesterol guidelines.
Second, the ACC/AHA cholesterol guidelines were selected as the best-practice
knowledge of interest in this study because of their recency. The guidelines were
published in November 2013, seven months before data collection commenced. Research
suggests that for survey respondents, recall of recently experienced events, behaviors, and
attitudes is easier and more accurate than for those in the more distant past (Groves et al.,
2009).
Third, the cholesterol guidelines recommended significant changes in physicians’
practices and were controversial, and thus likely generated (and required) more attention
from physicians than best-practice guidelines recommending more incremental and less
controversial changes in practice. Several aspects of the guidelines generated controversy
and criticism. The 2013 guidelines represented one in a series of regularly updated and
published best-practice recommendations on cholesterol, and the publication of the 2013
update was significantly delayed. This delay occurred in part because the National Heart,
Lung, and Blood Institute, a division of the U.S. National Institutes of Health, decided
during the middle of the guideline authorship process to transfer its institutional
135
management and control of the process to the ACC and AHA (Downs & Good, 2014;
Mitka, 2013). The transfer of authorship caused some uncertainty, as stakeholders
wondered how the transfer may have changed the content of the guidelines, and the
length of the delay frustrated those who believed new guidelines were overdue.
When they were finally published, the 2013 guidelines recommended several
significant changes in practice, described by observers as “a marked departure”
(Krumholz, 2013) and “a paradigm shift with tremendous implications” (Downs & Good,
2014, p. 354). The previous 2001 version of the guidelines recommended that physicians
base treatment decisions primarily on the level of a particular type of cholesterol, low-
density lipoprotein (LDL) or “bad” cholesterol, with the goal of reaching a targeted
healthy level of LDL cholesterol through statin drug therapy and lifestyle changes. In
contrast, the 2013 guidelines recommended that physicians base treatment decisions
primarily on a detailed calculation of a person’s risk of atherosclerotic cardiovascular
disease, with statin drug therapy to be initiated if the person exceeded a certain level of
risk. Adherence to the new guidelines meant elimination of regular testing of LDL
cholesterol levels (Keaney, Curfman, & Jarcho, 2014). It also meant implementation of a
new risk calculator that experts both internal and external to the guidelines authorship
process reported as overestimating the risk of disease by 70% to 150% (Kolata, 2013;
Martin & Blumenthal, 2014; Ridker & Cook, 2013), prompting concern that following
the new guidelines could lead to unnecessary, expensive, and potentially harmful
overtreatment with statin medications (Downs & Good, 2014; Ridker & Cook, 2013).
Stakeholders worried that the guidelines’ increased emphasis on prescribing statin
136
medications might reflect the inappropriate influence of the pharmaceutical companies
that sell statins on guideline authors (Abramson & Redberg, 2013).
Additionally, the previous iteration of the guidelines was supported primarily on
the basis of expert consensus, whereas the 2013 guidelines were supported primarily on
the basis of research evidence from clinical trials that had been conducted after the
publication of the previous guidelines in 2001 (Downs & Good, 2014). This might seem
like a natural and reasonable shift in the evidentiary basis for the best practices. However,
as Guallar and Lane (2014) point out, the type of evidence and expertise supporting
clinical best-practice guidelines is often contested, with some stakeholders strongly
believing that one type is superior to another.
During design of the present research project, it was hoped that selecting best-
practice guidelines that attracted a lot of attention from physicians, like the 2013
cholesterol guidelines, would increase the likelihood that survey respondents would (a)
easily and accurately remember their attitudes and behaviors surrounding the guidelines,
and (b) be motivated to participate in the survey. In addition to making the guidelines
potentially more memorable and interesting to survey respondents, the controversy
surrounding them also addressed two of the independent variables of interest in this
research, source legitimacy and credibility. In an article regarding the questions raised
about the new cardiovascular disease risk calculator, New York Times (Kolata, 2013)
reporting highlighted the questions of source legitimacy and credibility at play in the
dissemination of the cholesterol guidelines:
Asked to comment on the situation [i.e., controversy over the risk calculator] on
Sunday, some doctors said they worried that, with many people already leery of
137
statins, the public would lose its trust in the guidelines or the heart associations
[the ACC and AHA]. ‘We’re surrounded by a real disaster in terms of credibility,’
said Dr. Peter Libby, the chairman of the department of cardiovascular medicine
at Brigham and Women’s Hospital.
A fourth reason for the selection of the 2013 cholesterol guidelines for this
research was that because they were controversial, the guidelines attracted substantial
attention in the popular press and from various medical publications and professional
groups. The New York Times, for example, published an editorial recommending that
people in “good cardiovascular health” ignore the guidelines until the new risk calculator
was reevaluated and adjusted to address concerns (New York Times Editorial Board,
2013). The substantial media attention the guidelines attracted was advantageous in that it
had the potential to improve survey respondents’ recall and also interest in participating
in the study. It also meant that the number and range of information sources available
about the guidelines were likely at their largest. Of course, the high level of attention also
made the 2013 cholesterol guidelines somewhat unique; few physician best-practice
guidelines prompt an editorial in The New York Times. At the same time, controversy
over clinical best-practice guidelines for physicians and other health care practitioners is
not unusual. As observers have pointed out, a number of other physician best-practice
guidelines have been met with considerable controversy, including 2009
recommendations for breast cancer screening and 2012 guidelines for prostate cancer
screening (Guallar & Laine, 2014). Clinical best-practice guidelines that attract
controversy may simply reflect an amplified but perennial set of questions and issues that
surround almost all “best” practices (Guallar & Laine, 2014; Orlikowski, 2002; Rycroft-
138
Malone, et al., 2004; Wellstein & Kieser, 2011). Who gets to decide which practices are
best? Whom are they best for? What levels of legitimacy and credibility do the authors of
best practices have? What are their motives? What is the appropriate evidentiary basis for
judgments about what is best?
Data Collection
A survey instrument was used to collect the study data over an approximately
three-month period from June 2014 through August 2014. The survey was created and
administered using Qualtrics, an online survey data collection tool. It was composed of
three main sections: network questions identifying physician and artifact sources of best-
practice knowledge, questions about the perceived legitimacy and credibility of those
sources, and questions about communication load. In its online format, the survey
consisted of 20 questions and 22 web pages, and was designed to take approximately 10
to 15 minutes to complete. Pre-programmed drop-down menus and skip logic were
included where possible to reduce respondent burden. The survey was formatted so that
respondents were able to skip any question in the survey, to exit the survey at any time,
and to complete the survey in multiple separate sessions. A preliminary version of the
survey was pilot-tested by five subject matter experts and three Alpha Medical Group
physicians, and revised based on their feedback. Appendix A presents a text version of
the survey without the online formatting, informed consent information, and thank-you
page.
The survey was distributed to all of the 185 primary care physicians employed by
Alpha Medical Group at the time of data collection. As discussed previously, each of
these physicians worked in a primary care department in one of 17 Alpha ambulatory
139
care clinics. Each primary care department was led by a department chief who was a
primary care physician or physician assistant. Invitations to voluntarily participate in the
survey were sent to the study population via Alpha’s interdepartmental mail system and
then, two days later, via e-mail. The interdepartmental mail invitation contained a $20.00
gift certificate for Amazon.com, the online retail store, as a token of appreciation for
completing the survey. The e-mail invitation provided an individualized URL link to the
online survey. A maximum of five reminders to participate in the survey were sent to
non-respondents, and thank-you letters were also sent to respondents, both via
interdepartmental mail and e-mail.
In addition to this basic protocol, a primary care physician at Alpha Medical
Group was recruited to serve as a study co-principal investigator for the data collection
stage of the research project. The co-principal investigator presented information about
the survey at a meeting of the primary care department chiefs prior to the commencement
of data collection, and sent personalized reminders about the survey to chiefs whose
clinics had a high number of non-respondents. The co-principal investigator also co-
signed all correspondence sent to potential respondents, including recruitment letters,
reminders, thank-you letters, and the survey itself.
In addition to the survey instrument, demographic data about the primary care
physicians in the study population was collected from publicly-available records. These
demographic data consisted of each physician’s gender and tenure in the organization,
available on Alpha Medical Group’s public Web site; and each physician’s tenure in the
profession, available on the state medical board of registration Web site. Two additional
140
physician demographic variables, department chief status and clinic site, were obtained
from Alpha Medical Group’s administrative records.
Measures
Network structure
The research design required measurement of the structure of two networks, the
unipartite physician-to-physician knowledge communication network, and the bipartite
physician-to-artifact knowledge communication network. (The third network under
investigation, the multilevel network, was simply the combination of the first two
networks, and therefore required no additional measures.) Researchers use two main
approaches to elicit network structure from survey respondents: the roster or recognition
approach, in which the respondent is provided with a list of names of all network
members and asked to identify those with whom he or she has network ties; and the recall
approach, in which the respondent is asked to identify from memory those with whom he
or she has ties (Marsden, 1990). In empirical settings in which researchers have access to
the names of all network members, evidence strongly supports the use of a roster rather
than recall design to aid respondents in identifying ties in whole-network surveys, in
order to reduce recall biases and cognitive load (Brewer, 2000; Marsden, 1990, 2005,
2011; Pustejovsky & Spillane, 2009)—except when network size exceeds approximately
100 nodes and using a roster of that length would place a substantial burden on
respondents (Marsden, 2011; Valente, 2010). Because the number of physicians in the
unipartite physician-to-physician network was 185, the present project used a roster
format featuring drop-down menus with drill-down display logic that filtered physician
141
names according to the clinic to which they belonged, making the roster length less
burdensome. The prompt for the unipartite network question read:
“Please list in the spaces below the names of all of your [Alpha Medical Group]
PCP [primary care physician] colleagues from whom you have learned about the
2013 ACC/AHA cholesterol guidelines. Select the person’s clinic location and
name from the alphabetized drop-down menus provided. List as many or as few
names as appropriate, then scroll to the bottom of the page to move to the next
question. If you wish to list more names, there is an option to do so at the bottom
of the page.”
In contrast to the unipartite network question, the network question for the
bipartite physician-to-artifact network used a recall format, given that there was no way
to construct a roster of all possible media artifacts that a respondent might wish to
identify. The survey asked respondents to identify from memory the artifacts from which
they had learned about the cholesterol guidelines. The prompt for the bipartite network
question read:
“Please list in the spaces below all of those sources other than people, from which
you have learned about the 2013 ACC/AHA cholesterol guidelines. These may
include UpToDate; the official guidelines articles published in the journals JACC
[Journal of the American College of Cardiology] and Circulation; mass media
such as newspapers, TV, and radio; websites; or other sources. The medium
(website, e-mail, paper, etc.) by which you accessed this source does not matter.
Please provide the specific name of the source (for example, “JACC” or “Boston
Globe”) rather than a category (for example, “journal” or “newspaper”). List as
142
many or as few sources as appropriate, then scroll to the bottom of the page to
move to the next question.”
In addition to choosing between roster and recall approaches, the implementation
of network “name generator” questions in survey instruments also requires the researcher
to decide on what sorts of limits, if any, should be placed on the number of names the
respondent may identify in response to the question. Substantial research supports placing
no limits on the number of names that may be listed, in order to avoid biasing
respondents’ out-degree scores and the density of the whole network (Marsden, 1990,
2005, 2011; Vehovar, Manfreda, Koren, & Hlebec, 2008). In many empirical contexts,
however, such a design imposes a significant burden on respondents and/or proves
impractical for the researcher to implement given the mode or format of the survey
instrument. For the present project, the Qualtrics online survey platform was able to
support the programming required for no more than about ten names to be listed in roster
format for the unipartite and bipartite network questions, and therefore respondents were
able to list up to ten names for each of those questions.
Following data collection, the raw data describing the structure of the two
networks were processed in several steps. First, for the artifacts nominated by survey
respondents, I combined duplicate names and clarified ambiguous names using Google
searches, in consultation with an Alpha Medical Group physician who did not participate
in the survey.
Second, for both networks, all physicians and artifact sources were assigned ID
numbers, and the quantity and nature of missing network data were assessed. Physicians
in the study population who were both survey non-respondents and isolates (i.e., they
143
were not nominated as knowledge sources by their colleagues) were excluded from the
analysis.
In the final step of data processing, adjacency matrices for the two networks were
created. An adjacency matrix for all possible physician dyads in the unipartite knowledge
communication network was created where “1” indicated that a tie existed between two
physicians, and “0” indicated the absence of a tie. The matrix was constructed such that
the ego in the dyad (i.e., the “sender” of the tie or “nominator” of the knowledge source)
was the physician who reported either seeking or receiving best-practice knowledge, and
the alter (i.e., the “receiver” of the tie) was the physician who was the source of the
knowledge. An adjacency matrix for all possible physician-artifact dyads in the bipartite
knowledge communication network was also created, where “1” indicated that a tie
existed between a physician and artifact, and “0” indicated the absence of a tie. The two
adjacency matrices were then used as data files in the network analysis.
Legitimacy
As discussed in Chapter 4, the present study focused on sociopolitical legitimacy,
which was defined as a social evaluation of an organization’s desirability and adherence
to laws, social norms and collective values. In the present empirical setting, the
organizations of interest were the ACC and AHA, the institutional authors of the
cholesterol treatment guidelines, as well as all of the organizations that produced the
media artifacts that were identified by study participants as being sources of knowledge
about the guidelines. Scholars have articulated two dimensions of sociopolitical
legitimacy, moral and regulatory, although as Deephouse and Suchman note, these two
dimensions may not always be “fully separable empirical phenomena” (2008, p. 67).
144
Moral legitimacy refers to an organization’s general benefit to society. Receiving
favorable media coverage and forming interorganizational alliances are two means by
which organizations may acquire moral legitimacy. Regulatory legitimacy refers to an
organization’s conformity to laws, regulations, and standards created by governments,
credentialing associations, and professional bodies.
The perceived sociopolitical legitimacy of the artifact knowledge sources was
measured with six items, three for the moral dimension, and three for the regulatory
dimension. The items were adapted from items in the extant literature when possible. For
moral legitimacy, the items were: “People think this source is valuable to our society”
(adapted from Certo & Hodge, 2007; Shoemaker, 1982); “People in the general public
would approve of this source if asked their opinion” (adapted from Elsbach, 1994); and
“This source is viewed by industry experts as one of the top organizations in its field”
(adapted from Certo & Hodge, 2007; Elsbach, 1994). For regulatory legitimacy, the items
were: “This source always follows laws and regulations” (adapted from J. Y. Chung,
Berger, & DeCoster, 2011; Elsbach, 1994); “People would be shocked to hear that this
source violated any professional codes of conduct” (created for the present study); and
“This source’s organizational leaders believe in ‘playing by the rules’ and following
accepted operating guidelines” (adapted from Elsbach, 1994).
In the survey instrument, the legitimacy scale was formatted as a “name
interpreter” question for the bipartite network question. In social network survey
research, name interpreter questions follow network, or “name generator,” questions, and
elicit information about the attributes of the actors with whom a respondent has network
relationships. Name interpreter questions may be formatted in alter-wise blocks, such that
145
a respondent is asked to answer them immediately after providing each alter’s name (e.g.,
identification of Alter #1, name interpreter questions about Alter #1; identification of
Alter #2, name interpreter questions about Alter #2; and so forth). The alternative to the
alter-wise approach is question-wise formatting, where a respondent is asked to answer
the name interpreter questions only after identifying all of the alters’ names first (e.g.,
identification of Alters #1 and # 2, name interpreter questions about Alters #1 and #2).
Research suggests that the question-wise format is associated with lower rates of item
non-response and respondent dropout, and more reliability in the responses elicited,
especially in the case of surveys administered online (Alwin, 2010; Coromina &
Coenders, 2006; Vehovar, et al., 2008). Therefore question-wise formatting was used for
the legitimacy questions in the present survey, as well as for the credibility questions
described below. Survey respondents were asked to evaluate the legitimacy of each
artifact they nominated, as well as that of the ACC and AHA, on a seven-point scale,
with one anchor of the scale being “Strongly Agree” and the other being “Strongly
Disagree.” Following data collection, the legitimacy scores were coded so that “7”
equaled the former and “1” the latter.
The raw data assessing perceived legitimacy were processed in several steps.
First, the nature and quantity of missing data were assessed. Legitimacy data were
defined as missing if a respondent identified one or more artifacts in response to the
artifact name generator question, but did not answer one or more items in the legitimacy
name interpreter scale for each artifact identified (i.e., item non-response). Missing data
were imputed as described in the Treatment of Missing Data section, below.
146
Second, a confirmatory factor analysis was conducted to assess how well
respondents’ answers to the six legitimacy items were explained by the theorized
structure of one underlying factor, sociopolitical legitimacy, with two dimensions, moral
and regulatory. The confirmatory factor analysis was conducted following the procedures
described in the Preliminary Analyses section, also below.
Third, total legitimacy scores were calculated for each physician-artifact dyad by
averaging, for each artifact, the scores of the six individual items in the scale. Subsequent
treatment of these total legitimacy scores differed according to whether they were for the
ACC and AHA versus for the artifacts. In the case of the physician-ACC and physician-
AHA dyads, the two dyadic total legitimacy scores were converted, without any
additional computation, into two monadic scores describing each physician’s perception
of the ACC and AHA. So, for example, Physician #1’s total legitimacy score for the ACC
was treated as an attribute of Physician #1 called “ACC legitimacy,” instead of being
treated as an attribute of the Physician #1-ACC dyadic relationship.
In the case of the physician-artifact dyads, the total legitimacy scores were
converted into monadic scores describing the collectively perceived legitimacy of each
artifact. This was accomplished by averaging the total legitimacy scores an artifact
received from all physicians. So, for example, if Artifact A was identified as being used
by two physician respondents, and Physician #1’s total score for the artifact was 5 and
Physician #2’s total score was 6, then Artifact A’s final monadic legitimacy score became
5.5.
The dyadic legitimacy scores were converted to monadic scores for several
reasons. First, in the case of ACC legitimacy and AHA legitimacy, I was interested,
147
conceptually, in legitimacy as a property of the physician, that is, legitimacy as perceived
in the “eye of the beholder.” I wanted to know if physicians who had better perceptions
of the ACC’s (and AHA’s) legitimacy were more likely to access media artifacts
containing knowledge about the ACC/AHA guidelines, and thus wanted to operationalize
legitimacy as an attribute of the physician. Second, in the case of artifact legitimacy, I
was interested, conceptually, in legitimacy as property of the artifact, one that is
collectively conferred upon it by members of an organizational or professional
community. Here I wanted to know if artifacts with greater legitimacy were more likely
to be used by physicians as sources of best-practice knowledge, and thus wanted to
operationalize legitimacy as an attribute of the artifact. Third, from a practical perspective
that applies to both cases, the MPNet analytical software used in this project does not
accommodate analysis of dyadic versions of attributes in multilevel models, and thus use
of the dyadic legitimacy scores for the ACC, the AHA, and the artifacts was not
technically possible.
10
Fourth, and also from a practical perspective, even if MPNet
accommodated analysis of dyadic covariates in its multilevel models, such analysis
would not be possible for the artifact legitimacy data. This is because use of dyadic
scores for the artifact legitimacy variable would have required that dyadic legitimacy data
be available for all possible physician-artifact dyads in the network, not just those
physician-artifact dyads that shared a knowledge communication tie.
The fourth and final data processing step was to prepare the legitimacy data for
network analysis. MPNet requires that for each network being analyzed, attribute data
files are prepared for each type (e.g., binary, continuous, categorical) of attribute variable.
10
It is likely that future versions of MPNet, or future versions of an alternative multilevel network
modeling platform, will allow for analysis of dyadic covariates in multilevel models.
148
In each attribute data file, the data were separated such that each row corresponds to a
node in the network and each column corresponds to an attribute that applies to the node.
For the unipartite network, the ACC legitimacy and AHA legitimacy scores for each
physician were entered into the unipartite continuous attribute file. For the bipartite and
multilevel networks, the ACC legitimacy and AHA legitimacy scores for each physician
were similarly entered into the bipartite/multilevel continuous attribute file, with zeros
inserted for each artifact. The legitimacy scores for each artifact were also entered into
the bipartite/multilevel continuous attribute file, with zeros inserted for each physician.
Credibility
As discussed in Chapter 4, source credibility was defined as an individual
evaluation of the “believability” (Flanagin & Metzger, 2008, p. 8) of a source. In contrast
to legitimacy, which is a social evaluation, credibility is an individual evaluation, because
although it may be influenced by others’ opinions, a credibility judgment is the
evaluator’s own assessment of source believability and reflects the evaluator’s unique
relationship and history with the source. In the present empirical setting, source
credibility was measured for both the physician and artifact sources of best-practice
knowledge, as well as for the ACC and AHA. Scholarly consensus identifies two
dimensions of credibility, expertise and trustworthiness. Expertise refers to the degree to
which the source is qualified, based on knowledge and experience, to provide valid
information. Trustworthiness refers to the degree to which the source is motivated, based
on honesty and lack of bias, to provide valid information.
The credibility of the knowledge sources was measured with six items, three for
the expertise dimension, and three for the trustworthiness dimension. For the expertise
149
dimension, the items, all adapted from McCroskey and Teven (1999), were: “This source
provides expert information”; “This source is well-informed”; and “This source conveys
competence.” For the trustworthiness dimension, the items were: “This source is
trustworthy” (adapted from Flanagin & Metzger, 2000, 2007; McCroskey & Teven,
1999); “This source is unbiased” (adapted from Flanagin & Metzger, 2000; Flanagin &
Metzger, 2007); and “This source is reliable” (adapted from Flanagin & Metzger, 2007).
In the survey instrument, the credibility questions were treated as name interpreter
questions for the network name generator questions. As was the case for the legitimacy
questions, question-wise formatting was used. Survey respondents were asked to evaluate
the credibility of each physician and artifact they nominated as knowledge sources, as
well as that of the ACC and AHA. They were asked to make the evaluation on a seven-
point scale, with one anchor of the scale being “Strongly Agree” and the other being
“Strongly Disagree.” Following data collection, the credibility scores were coded so that
“7” equaled the former and “1” the latter.
The imputation of missing credibility data and the confirmatory factor analysis of
the credibility construct were conducted in the same manner as for the legitimacy data
and construct. These procedures are described in more detail in the Treatment of Missing
Data and Preliminary Analyses sections, below.
Total credibility scores were calculated for each physician-knowledge source
dyad by averaging, for each knowledge source, the scores of the six individual items in
the credibility scale. Subsequent treatment of the total credibility scores differed
according to whether they were for the ACC or AHA, the artifact knowledge sources, or
the physician knowledge sources. The treatment of credibility scores for the ACC, AHA,
150
and artifacts was the same as that for the legitimacy scores. Thus in the case of the
physician-ACC and physician-AHA dyads, the two dyadic total credibility scores were
converted, without any additional computation, into two monadic scores describing each
physician’s perception of the ACC and AHA. These monadic credibility scores were
attributes of the physician. In the case of the physician-artifact dyads, the total credibility
scores were converted into monadic scores by averaging the scores an artifact received
from all physicians. These monadic credibility scores were attributes of the artifacts.
In the case of the physician-physician dyads, the total credibility scores were also
converted into monadic scores that were the attributes of each physician, but the
procedures to accomplish this differed. A key challenge to converting credibility into a
monadic score was the lack of credibility data for physicians who were not nominated as
knowledge sources. Because it was not feasible to ask survey respondents to evaluate the
credibility of all 185 physicians in the organization, credibility scores were collected only
for those physicians whom respondents identified as knowledge sources. Physician
evaluations of the credibility and legitimacy of artifact knowledge sources were solicited
in the same way—by asking respondents to evaluate only those sources they actually
used, rather than the entire universe of artifacts containing information about the
cholesterol guidelines. However, doing so did not complicate the legitimacy and
credibility data for the artifact nodes, because artifacts not used by physicians were not
included in the bipartite network, meaning there were no attribute data missing for the
artifacts. Because physicians who were seekers or recipients of knowledge, but who were
not knowledge sources, were included in all three networks, the credibility data for those
“non-source” physicians were “missing.”
151
There were no ideal approaches for handling the missing credibility data for non-
source physicians. For example, imputation of the missing data would have likely
involved an unappealingly substantial proportion of the physician credibility data, subject
to the number of physicians identified as sources, and further, it would have involved
imputing data that were not missing at random, a practice that is not recommended
(Allison, 2010). A second option was to not convert the credibility scores from dyadic to
monadic form; this option was rejected due to the disadvantages of dyadic scores that
were discussed previously, in the section on legitimacy measurement. A third option was
to use two modes to represent the physicians, one for physicians identified as sources and
one for non-source physicians. This option did not account for physicians who were
simultaneously both seekers and sources of knowledge, nor was it methodologically
possible, given that MPNet currently does not accommodate the three-mode networks
that such a research design would require, with two physician modes and one artifact
mode.
Given the limitations of the above three options, a fourth approach was
implemented. The standard deviation of the physician credibility scores was subtracted
from the lowest physician credibility score, and the amount of the resulting difference
was used as a baseline credibility score for all non-source physicians. The objective of
this approach was to focus on the credibility of the physicians who were nominated as
knowledge sources, and to investigate if the physician sources who were perceived as
more credible were also more popular among their colleagues. This approach was
adapted from procedures described by Preciado and colleagues (2012, p. 22) in a study
assessing the effect of spatial distance on friendship networks. In their study, some
152
members of the friendship dyads lived in the same neighborhoods, next door to one
another, and so there were spatial distance measurements that were zero, which can cause
instability in model fitting. To address this problem, the authors transformed zero
distances into the smallest empirically observed distance, greater than zero, in the data
set. One problem with adapting Preciado et al.’s approach to the present data set is that in
the case of the friendship study, only a small proportion of the data in question had zero
values, whereas in the present study, a potentially large proportion of the physicians
could have missing credibility scores, skewing the distribution of these data. Because it
was not an optimal approach, two sets of models were run for unipartite and multilevel
networks: one with the physician credibility attribute, and one without.
A final note regarding the measurement of credibility is warranted. Theoretical
consensus, as discussed in Chapter 4, suggests that whereas legitimacy is a collective
evaluation that may be possible in many contexts to appropriately measure monadically,
credibility is an individual evaluation. One could therefore argue that physician and
artifact source credibility would be more appropriately measured using a dyadic attribute,
which considers only the credibility judgment of the focal actor in the dyad, rather than
the aggregate credibility judgment of all actors in the network. Because MPNet does not
accommodate analysis of dyadic versions of attributes in multilevel models, however—
and because it was not feasible to collect physicians’ credibility judgments about all their
colleagues and all of the artifacts they and their colleagues used—I chose instead to
measure physician and artifact credibility using monadic attributes.
After monadic credibility scores were computed for the ACC, AHA, artifacts, and
physicians, the credibility data were input into files formatted for network analysis. For
153
the unipartite network, the ACC credibility, AHA credibility, and physician credibility
scores for each physician were entered into the unipartite continuous attribute file. For the
bipartite and multilevel networks, the ACC credibility and AHA credibility scores for
each physician were entered into the bipartite/multilevel continuous attribute file, with
zeros inserted for each artifact. Likewise, the artifact credibility scores were also entered
into the bipartite/multilevel continuous attribute file, with zeros inserted for each
physician.
Communication load
Communication overload was defined as occurring when the high quantity and/or
low quality of communication a person receives is (1) different from what he or she
desires or (2) hinders his or her ability to process what is being communicated. The
theoretical literature suggests that there are two dimensions of communication load:
quantity, referring to the overall volume and number of different sources of
communication; and quality, referring to the relevance and ambiguity of the
communication. In the empirical literature, however, reported exploratory and
confirmatory factor analyses of items measuring communication and information load
consistently indicate one dimension only (Ballard & Seibold, 2006; Chen & Lee, 2013;
Cho, et al., 2011; Karr-Wisniewski & Lu, 2010; O'Reilly, 1980; Stephens & Davis,
2009). Communication load was thus conceptualized as having only one dimension in the
present study.
In the survey instrument, the physician respondents were prompted to “think
about all of the information (best practices as well as all other types of information) you
need to do your clinical work.” They were then asked to assess their level of
154
communication load “in a typical week in the previous month,” with seven items adapted
from two previously published scales (Ballard & Seibold, 2006; C. J. Chung &
Goldhaber, 1991). Each respondent was asked to estimate how often (1) “you feel you
have to engage in too much communication (for example, too many phone calls,
meetings, memos, letters, face-to-face conversations, emails, text messages, etc.)”; (2)
“you receive more information than you can process”; (3) “you receive more information
than you need in order to do your job effectively”; (4) “you have more discussion than
you wish to about confusing or ambiguous information”; (5) “your communicating with
others involves too many decisions”; (6) “you receive information that requires you to
make too many decisions”; and (7) “you receive information that needs too many
explanations in order for it to be useful to you.” The answer choices for these questions
were presented on a five-point scale, with one anchor of the scale being “Always” and the
other being “Never.” Following data collection, the communication load scores were
coded so that “5” equaled the former and “1” the latter.
Following data collection, missing communication load data were imputed and a
confirmatory factor analysis of the construct was performed as discussed in the Treatment
of Missing Data and Preliminary Analyses sections, below. Total communication load
scores were then calculated for each physician by averaging the scores of the seven
individual items in the scale.
For the research hypothesis regarding the interaction of physician communication
load and artifact mass media status (H4b), an additional data computation step was
required. In order to create an MPNet parameter that expressed the interaction of these
two attributes, both variables had to use the same form of measurement, requiring
155
dichotomization of the communication load variable to match the binary mass media
variable. Although researchers typically try to avoid dichotomizing variables because the
practice results in data loss, no alternative means to express the interaction hypothesis
were available. Therefore for this hypothesis only, the communication load scores were
split into two categories, with the split set at the sample median. Scores above the median
were coded as “1,” scores below the median were coded as “0,” and scores equaling the
median were also coded as “0,” because I wanted the “high” category to be more
restrictive than the “low” category.
For the attribute data files used for network analysis, the continuous
communication load scores for each physician were entered into the unipartite continuous
attribute file. The continuous and binary versions of the communication load scores were
entered into the bipartite continuous and binary attribute files, respectively, with zeros
inserted for each artifact.
Qualitative data
One open-ended question was included at the end of the survey to elicit
qualitative data regarding the research hypotheses and the general subject of how
physicians learn about best-practice knowledge. The question prompt stated:
“In the space below, please provide any additional information you feel is
important for us to better understand how you communicate with your colleagues
or use other sources to learn about clinical best practices. For example, you may
wish to elaborate on one of your above answers, or suggest important questions
that we omitted but should have asked about this topic.”
156
Following data collection, responses to this question were examined to identify
data that complemented, contradicted, or helped explain results derived from the
quantitative data, and these qualitative insights are discussed in Chapters 6 and 7.
Systematic analysis of these qualitative data, however, was beyond the scope of the
present project.
Mass media status
Following data collection, all of the artifacts identified as sources of best-practice
knowledge by the physician respondents were coded as being either mass media artifacts
or niche media artifacts. I defined mass media artifacts as those targeting a mass, general
audience, and coded them with a score of “1.” I defined niche media artifacts as those
targeting physicians, and coded them with a score of “0”. For the attribute data files used
for network analysis, the mass media status scores were entered into the bipartite binary
attribute file, with zeros inserted for each physician.
Demographic characteristics
Data were collected on six demographic characteristics of the primary care
physicians in the study population: gender, department chief status, clinic site, number of
clinical sessions worked per week, professional tenure, and organizational tenure.
Gender data were collected from Alpha Medical Group’s public Web site. Gender
was treated as a binary variable, where “1” equaled woman. The gender attribute variable
was included in the analysis to control for any differences in knowledge communication
behavior between men and women, and to test for the possibility of gender homophily or
heterophily effects, all of which have been reported previously in research on
157
organizational knowledge networks (Borgatti & Cross, 2003; Brass, 1985; Ibarra, 1992,
1997; Lazega & van Duijn, 1997; Lomi, et al., 2014; McPherson, et al., 2001).
Data on which physicians were chiefs of the primary care departments were
obtained from Alpha’s administrative records. Department chief status was also treated as
a binary variable, with “1” indicating the physician was a chief. This attribute variable
was included in the analysis to test for any differences in knowledge communication
behavior between chiefs and other physicians. Many scholars have argued for the
importance of considering both formal structure and informal relationships in the study of
organizational networks (e.g., Adler & Kwon, 2002; McEvily, Soda, & Tortoriello,
2014). It was possible that in the case of Alpha, chiefs were more likely to be seen by
their colleagues as sources of best-practice knowledge, given their higher status and more
visible position in the formal organizational hierarchy. A number of previous studies of
organizational knowledge networks have observed that those with a supervisory role or
higher position in the formal hierarchy are more likely to be identified as sources of
work-related expertise and advice (e.g., Agneessens & Wittek, 2012; Lazega & van
Duijn, 1997; Roberts & O'Reilly, 1979; Yousefi-Nooraie, Dobbins, & Marin, 2014). It
was also possible that chiefs were more likely to proactively act as both sources and
seekers of best-practice knowledge, given their formal responsibility within the
organization for promoting departmental adoption of and adherence to clinical best
practices.
Clinic site was treated as a categorical variable; there were 17 Alpha Medical
Group clinic locations with primary care departments. Clinic site data for each physician
in the study population were obtained from Alpha’s administrative records. The clinic
158
site attribute variable was included in the analysis to control for the possibility that
physicians in the same clinic may be more likely to use the same knowledge artifacts due
to social influence (Asch, 1951), groupthink (Janis, 1972), or other, similar mechanisms
based on the principle of the contagion of knowledge, attitudes, and behaviors within a
social network (Burt, 1987; P. R. Monge & Contractor, 2003; Su & Contractor, 2011).
Data on the number of clinical sessions worked per week were collected via
survey instrument. A question was in included in the survey regarding the average
number of hours the respondent spent in clinic (as opposed to time spent on
administrative or research responsibilities). Like many health care provider organizations,
Alpha Medical Group measures clinical time spent caring for patients in terms of the
number of half-day clinical sessions a physician works per week, with a maximum of 10
sessions being possible in a 5-day work week. The clinical sessions attribute variable was
included in the analysis to test for any differences in knowledge communication behavior
between physicians who spent more versus less time conducting patient care. It was
possible that physicians who spent more time seeing patients were more likely to be
identified as knowledge sources, either because their greater experience made them
preferred as sources by their colleagues, or because they were simply more accessible or
more active as knowledge sources/providers given the greater amount of time they spent
in the clinic, physically co-located with their colleagues. Keating and colleagues (2007),
for example, observed that physicians who spent more time in their practices caring for
women patients, and who spent more time overall in weekly clinical sessions, were more
likely to be identified by their colleagues as sources of influential information about
women’s health. Borgatti and Cross (2003) similarly found that having timely,
159
convenient access to a colleague acted as a significant predictor of seeking information
from that colleague. It was also possible that in the Alpha knowledge network, physicians
who saw more patients were more likely to seek out knowledge from colleagues and
artifacts because their greater clinical responsibilities meant they had a greater need for
best-practice knowledge, and/or they had more opportunities to passively receive best-
practice knowledge from colleagues.
Note that of the six demographic variables, clinical sessions was the only one that
was collected via survey instrument, and thus it was the only variable for which any data
were expected to be missing. Missing data were imputed as described in the Treatment of
Missing Data section, below.
Professional tenure (i.e., tenure as a physician) and organizational tenure (i.e.,
tenure at Alpha Medical Group) were measured in years, as continuous variables. The
professional tenure data were collected from the state medical board of registration Web
site, which provides the year of medical school graduation for each physician registered
in the state. The organizational tenure data were collected from Alpha’s public Web site.
These two variables were included in the analysis because extant literature suggests that
there are several mechanisms by which professional and organizational tenure may affect
the structure of organizational knowledge networks. First, tie formation in the network
may have been influenced by a preference among physicians for knowledge sources with
greater seniority. Colleagues with professional and/or organizational seniority have more
experience in performing day-to-day responsibilities and conforming to organizational
norms and practices, which may increase their desirability as knowledge sources.
Research on transactive memory systems indicates that greater professional tenure and
160
greater organizational tenure may act as cues or heuristics among knowledge seekers
trying to identify sources of expertise among their colleagues (Ren & Argote, 2011).
Roberts and O’Reilly (1979) found that in the U.S. Navy, those with greater professional
tenures were more likely to be identified as sources of expertise by their colleagues, and
Lazega and van Duijn (1997) observed that among attorneys at a law firm, there was a
preference for seeking advice from those with organizational tenures greater than one’s
own. Second, scholars have observed that, conversely, knowledge seekers are more likely
to be those with lesser professional (Zappa, 2011) and organizational tenures (Lazega &
van Duijn, 1997). A final mechanism by which professional and organizational tenure
may affect tie formation in knowledge networks is homophily; physicians may have a
preference for communicating about best practices with colleagues with similar levels of
seniority. In empirical research, scholars investigating knowledge networks have found
homophily to work in tandem with a preference for seniority. In addition to observing
seniority effects in advice networks, for example, both Zappa (2011) and Lazega and van
Duijn (1997) also observed strong homophily effects.
Treatment of Missing Data
The quantity and nature of the missing data were assessed for the network
structure data, for the primary attribute variables of interest—communication load,
legitimacy, and credibility—and for the clinical sessions demographic attribute variable.
As described above, for the network structure data, physicians in the study
population who did not respond to the survey and who were not nominated as knowledge
sources by any of their colleagues (i.e., non-respondent isolates) were excluded from the
data analysis. Physicians in the study population who did not respond to the survey but
161
who were nominated as knowledge sources by one or more colleagues (i.e., non-
respondent non-isolates) were included in the analysis.
This treatment of the missing network structure data meant that there were two
types of missing attribute data: unit-missing attribute data for the non-respondent non-
isolates, and item-missing attribute data for the respondents. Both types of missing
attribute data were imputed in Lisrel (version 9.1) using the full-information (a.k.a.
“direct” or “raw”) maximum likelihood (FIML) method, as recommended by Allison
(2010). This imputation method was chosen because it carries no assumption of
multivariate normality, it does not require the imputation and analysis models to be
identical, and it is relatively simple to execute as compared with other methods, such as
multiple imputation (Allison, 2010; Tabachnick & Fidell, 2007). A data set was
constructed for the imputation procedure that included all of the variables with missing
data, plus the demographic attribute variables without missing data (i.e., the gender, chief
status, clinic site, and tenure variables), also as recommended by Allison (2010). This
data set used the legitimacy and credibility data in their dyadic format, prior to
conversion to monadic format, and was constructed in “wide” form, such that there was
only one record (or row) per case (rather than multiple rows per case if the study
participant sent or received multiple network nominations) (Allison, 2010). Following the
FIML imputation procedure, I examined all of the imputed values and found that seven
were unrealistically, nonsensically small. For these seven values I therefore imputed
using the mean substitution method (Groves, et al., 2009).
162
Preliminary Analyses
Preliminary analyses of the data included analysis of non-response bias as well as
Cronbach’s coefficient alpha calculation and confirmatory factor analysis of the
measurement models for the three scaled variables, communication load, credibility, and
legitimacy.
Analysis of non-response bias
The potential for non-response bias was assessed in SPSS (version 22) using a
multivariate analysis of variance that compared the demographic characteristics of
respondents to those of non-respondents.
Scale reliability and confirmatory factor analysis
The Cronbach’s alpha calculations and confirmatory factor analyses were
conducted on the imputed data set using SPSS and Lisrel, respectively. A Cronbach’s
alpha value of 0.70 or greater was considered indicative of satisfactory scale reliability
(DeVellis, 2003). All items in the three scaled variables were ordinal measures, and
several of the credibility items displayed moderate to substantial amounts of skewness
(i.e., −1.5 to −1.8) and kurtosis (i.e., 3.5 to 4.0), indicating non-normal distributions.
Consequently, for the confirmatory factor analyses, polychoric correlation and
asymptotic covariance matrices were calculated in Prelis 2 and diagonally weighted least
squares estimation was used for modeling. In each model, the metric of the factor(s) was
defined by designating the first indicator variable of each factor as the marker.
Global goodness of fit of the measurement models was evaluated using statistics
in each the three main categories of fit indices: absolute fit, comparative fit, and
parsimony-adjusted fit (T. A. Brown, 2006). Fit statistics were chosen based on their
163
appropriateness for non-normally distributed data modeled using diagonally weighted
least squares estimation, and based on the recommendations of Brown (2006) and Kline
(2011), authors of two structural equation modeling guidebooks that summarize recent
research and expert consensus on goodness of fit assessment.
11
Among absolute fit indices, the Satorra-Bentler scaled chi-square, recommended
for the various forms of weighted least squares estimation, was used, where a non-
significant chi-square value suggests satisfactory fit. Chi-square statistics are sensitive to
sample size, and chi-square statistics calculated using diagonally-weighted least squares
estimation have been shown to be particularly sensitive to small sample sizes of 250 or
less (Nye & Drasgow, 2011). As a result many experts have recommended that models
not be rejected solely on the basis of significant chi-square values (e.g., T. A. Brown,
2006; Kline, 2011). A second absolute fit index, the standardized root mean square
residual (SRMSR), was also used, where values range from 0.0 to 1.0, a value of 0.0
indicates a perfect fit, and values of 0.08 or less are considered indicative of satisfactory
fit (Hu & Bentler, 1999).
Among comparative fit indices, Bentler’s comparative fit index (CFI) was used,
with values greater than 0.95 suggestive of satisfactory fit (Hu & Bentler, 1999).
Among parsimony-adjusted fit indices, the root mean square error of
approximation (RMSEA) was used, with values of 0.06 or less, narrow 90% confidence
intervals, and p values greater than 0.05 suggestive of satisfactory fit (Browne & Cudeck,
11
Note, however, that currently there is some uncertainty about whether commonly-used fit indices and
their generally accepted cutoff criteria, such as those used here, apply to models estimated using diagonally
weighted least squares in the same way that they apply to models estimated using maximum likelihood
(Jöreskog, 2004; Nye & Drasgow, 2011). Relatedly, to the best of my knowledge, there is no agreed-upon
set of equations for Satorra-Bentler scaled chi-square difference tests when diagonally weighted least
squares is used for model estimation; equations are available only for Satorra-Bentler scaled chi-square
difference tests when maximum likelihood is used (F. B. Bryant & Satorra, 2012, 2013). For this reason, I
did not calculate chi-square difference tests for any of the confirmatory factor analyses reported here.
164
1993; Hu & Bentler, 1999). RMSEA, however, like chi-square statistics, is sensitive to
sample size and has been shown to “over-reject” otherwise adequately-fitting models
with sample sizes of 250 or less (Hu & Bentler, 1999).
Localized goodness of fit in the three measurement models was assessed by
examining the parameter estimates, the standardized residuals, and the modification
indices. Parameter estimates were inspected for statistical significance, small standard
errors, the absence of negative values, and in the case of the standardized estimates, the
absence of values above 1.0 (T. A. Brown, 2006). Standardized residuals and
modification indices were inspected for potential sources of poor model fit, with any
absolute values above 1.96 or above 3.84, respectively, indicating a potentially poor fit
(T. A. Brown, 2006).
Respecification of the hypothesized model was attempted when theoretically
justifiable and when suggested by multiple global and local indicators of poor fit.
Network Analysis
To conduct the network analysis, the research hypotheses proposed in Chapters 3
and 4 were translated into network model parameters by identifying the network
mechanisms underlying each hypothesis. In the sociomaterial network conceptualized for
this study, there were two main types (i.e., modes) of actors (i.e., nodes): (1) physicians
who were both knowledge consumers and knowledge sources, and (2) various artifacts
that also acted as knowledge sources. Knowledge communication relations were possible
between physicians (e.g., Physician A obtained knowledge from Physician B, or
Physician B shared knowledge with Physician A) and between physicians and artifacts
(e.g., Physician A seeks knowledge from Artifact A). To accommodate a two-mode
165
network model with relations existing between actors of the same mode, and between
actors of different modes, multilevel exponential random graph modeling was used to
conduct the analysis. In the remainder of this chapter, multilevel exponential random
graph modeling and the network models used in the study are described in more detail.
The exponential random graph model
Exponential random graph models (ERGMs)
12
have been used extensively in
recent years to analyze cross-sectional network data. An ERGM is a type of stochastic
model for the structure of social networks. The objective of ERGM is to identify the
underlying social processes responsible for the formation of a particular empirically-
observed social network. In ERGM, the presence of each tie in the observed network is
treated as dependent on the presence of these underlying social processes, which are
manifested by various network tie formations (such as reciprocity or transitivity) and by
actor attributes (such demographic characteristics, behaviors, or attitudes, in the case of
human nodes). This assumption of dependence stands in contrast to the assumption of
independent observations in a regression model, and is the defining feature not just of
ERGM but of the field of network theory and analysis writ broadly. In ERGM, as in other
forms of network modeling, global network structure is thus presented as the collective
result of the structural formations and actor attributes that occur locally in the network
(Wang, Robins, Pattison, & Lazega, 2013).
Currently there are two main software platforms for ERGM: the PNet program
and the statnet package used in the R statistical computing environment. PNet was
selected for this analysis because at the time of the study, it was the only platform that
allowed for the modeling of multilevel networks with cross-sectional data.
12
I use ERGM in this document as an abbreviation for both exponential random graph model and modeling.
166
Model estimation in ERGM proceeds as follows.
13
The researcher first specifies,
according to theory and empirical evidence, the structural and attribute parameters to be
included in the network model that may significantly influence the global structure of the
observed network. The modeling routine then generates Markov chain Monte Carlo
simulations of hypothetical networks, each having the same number of nodes, modes, and
types of ties as the observed network. It compares the probability sample of simulated
networks to the observed network in terms of the structural and attribute parameters of
interest, in order to produce maximum likelihood estimates for each parameter specified
in the model. If the model is appropriately specified and successfully converges, the
resulting maximum likelihood estimates assess whether a parameter is significantly more
or less likely to occur in the observed network, as compared to how often it would occur
in a similar network (such as one in the sample of simulated networks) simply by chance.
In other words, if the model converges, the maximum likelihood estimates indicate the
values of the parameters that are most likely to have generated the observed network
(Koskinen & Snijders, 2013).
For each parameter specified in the model, a standard error, convergence t-ratio,
and sample autocorrelation function are calculated in PNet to accompany the maximum
likelihood estimate. The standard error is used in determining the significance of a
parameter; the parameter estimate is statistically significant if its absolute value is more
than twice the standard error (Koskinen & Snijders, 2013).
14
The convergence t-ratio
13
For more detailed information about ERGM, particularly as implemented in the suite of PNet software,
please refer to reviews by Robins and colleagues (Lusher, Koskinen, & Robins, 2013; Robins, Pattison,
Kalish, & Lusher, 2007; Robins, Snijders, Wang, Handcock, & Pattison, 2007), Shumate and Palazzolo
(2010), and Valente (2010, pp. 151-171).
14
The quotient of the parameter estimate divided by the standard error is sometimes referred to as the “t-
statistic,” which should be distinguished from the convergence and goodness of fit t-ratios.
167
considers the difference between the value of the parameter estimate in the observed
network and the mean value of the parameter estimates in the sample of simulated
networks, and compares this difference to the standard deviation of the sample mean
estimate (Robins, Pattison, & Wang, 2009). The convergence t-ratio indicates how well
the parameter has converged in the model, and should be less than 0.1 (Koskinen &
Snijders, 2013). The sample autocorrelation function of each parameter estimate indicates
how similar the simulated networks are to each other in terms of that parameter
(Koskinen & Snijders, 2013; Ripley, Snijders, Boda, Vörös, & Preciado, 2013; Robins &
Lusher, 2013). If the networks are too similar, autocorrelation in the statistics generated
by the estimation algorithm will be high, indicating a less reliable parameter estimate.
The sample autocorrelation function should be less than 0.4.
Following parameter estimation and model convergence, goodness of fit statistics
are generated to assess how well the overall model fits the observed network, both in
terms of the specified structural and attribute parameters and also in terms of parameters
that were not hypothesized by the researcher to be significant in the observed network
and that were not explicitly specified in the model. Typically, the latter includes all
possible parameters that are available for estimation in the particular ERGM software
suite (PNet or statnet) used by the researcher. The goodness of fit statistics evaluate how
accurately the specified model describes all of the features of the network, not just those
that are explicitly modeled and found to significantly influence the observed network
structure. In PNet, goodness of fit statistics for each parameter include the count of the
instances of the parameter in the observed network; the mean of the instances of the
parameter in the sample of simulated networks; and the goodness of fit t-ratio, which is
168
the same statistic as the convergence t-ratio. For good model fit, the parameter count and
mean values should be close to each other. For parameters specified in the model, the
goodness of fit t-ratio should be approximately less than 0.1, as with the convergence t-
ratio (Koskinen & Snijders, 2013). For parameters not explicitly specified in the model,
the goodness of fit t-ratio should be approximately less than 2.0. PNet also includes a
Mahalanobis distance statistic for the overall model as a global goodness of fit heuristic.
The Mahalanobis distance assesses how far away the empirical network is from the
distribution of networks simulated from the model; smaller values indicate that the center
of the distribution of simulated networks is close to the observed network (Wang, Sharpe,
Robins, & Pattison, 2009). Wang and colleagues (2009) suggest that Mahalanobis
distance be used to identify the best-fitting model when a researcher is comparing several
converged models with similar individual goodness of fit t-ratios.
Multilevel exponential random graph modeling
Basic ERGMs, as implemented in PNet or statnet, can be used to analyze cross-
sectional 1-mode uniplex networks. ERGMs have also been developed for 2-mode
uniplex networks and 1-mode multiplex networks. Wang and colleagues (Wang, Robins,
Pattison, & Lazega, 2013; Wang, Robins, Pattison, Lazega, & Jourda, 2013) recently
developed a multilevel network extension to ERGMs in PNet called MPNet, which
allows for modeling of 2-mode multiplex networks. MPNet is designed for the analysis of
nodes of two modes, with each mode representing a different level (Wang, Robins,
Pattison, & Lazega, 2013). Each level represents a 1-mode network, and a 2-mode
network exists between the nodes from two adjacent levels. For example, a multilevel
ERGM might consist of a macro-level, 1-mode, uniplex network A; a micro-level, 1-
169
mode, uniplex network B; a meso-level, 2-mode, uniplex network X; and a combined
multilevel, 2-mode, multiplex network M that considers the three macro-, micro-, and
meso-level networks simultaneously (Wang, Robins, Pattison, & Lazega, 2013). Figure 1
(see page 171) illustrates this generic multilevel network model for ERGM in MPNet.
MPNet is designed to investigate dependencies among the micro-, meso-, and
macro-level networks (Wang, Robins, Pattison, & Lazega, 2013). The interpretation of a
network at one level is conditional on the other networks in the model. MPNet allows a
researcher to estimate and compare models with and without multilevel effects, in order
to assess the importance of multilevel effects to overall model goodness of fit (Wang,
Robins, Pattison, & Lazega, 2013). It includes specifications for both undirected and
directed networks, and for social selection models (Wang, Robins, Pattison, Lazega, et
al., 2013). Network directionality indicates which node in a dyad sends or initiates a
relationship, and which node receives or accepts the tie; in undirected networks, the
directionality is not specified or is not relevant to the relationship. A social selection
network model conceptualizes the causal relationship between structural and actor
attribute variables such that attributes are assumed to influence structure, rather than
structure affecting attributes. Social selection models are discussed in further detail in the
General approach to model specification section, below.
MPNet was developed to accommodate hierarchically nested network data, such
as data on organizations and the departments within the organizations, or on schools and
classrooms. However, it is flexible enough to also accommodate a network
conceptualization that is not inherently hierarchical—that is, a network that would not
usually be thought of as having multiple “levels”—but that has (a) two modes and (b) ties
170
that exist both between nodes of different modes and nodes of the same mode. In such a
“non-hierarchical” multilevel network model, the definition of the “levels” is not
determined by the hierarchical nature of the nodes (e.g., individuals and groups), but by
specific attributes of the nodes (e.g., male and female). Such a network model, with two
modes and ties within and between the modes, but without hierarchical levels (Iacobucci
& Wasserman, 1990; Wasserman & Iacobucci, 1991), assumes “that the network effects
are heterogeneous among nodes with different attributes” (Wang, Robins, Pattison, &
Lazega, 2013, p. 111).
For this dissertation, I analyzed data in MPNet (version 1.04) using this non-
hierarchical type of multilevel network model, in which there were three networks, two
modes, and ties within and between the modes. I conceptualized the first two networks as
(a) a unipartite uniplex Network A comprised of the physician consumers and/or
physician sources of knowledge (the first mode); and (b) a bipartite uniplex Network X
representing an affiliation network between the physicians and artifact sources of
knowledge (the second mode). When these two “subordinate” networks are considered
together and modeled simultaneously, the result is a third, multilevel, bipartite, multiplex
Network M representing the physicians’ communication ties to the two types of
knowledge sources available to them, other physicians and media artifacts. Note that all
of the ties throughout the multilevel network are conceptualized as knowledge
communication ties, although the nature of the communication differs according to the
type of actors involved. In Network A, the physicians have directed knowledge
communication ties with each other, and each physician is both a potential consumer and
a potential source of knowledge. In Network X, the physicians have undirected
171
knowledge communication ties (or “affiliations”) with the artifact sources from which
they have acquired knowledge. In Network M, both types of ties are considered
simultaneously in order to investigate the dependences between them. Figure 2 (see page
172) illustrates the multilevel model of the sociomaterial knowledge network that I
estimated in MPNet. Note that this model differs from the generic model illustrated in
Figure 1, in that no ties are specified between the artifact nodes in “Network B.”
Macro‐level
Network (A)
Micro‐level
Network (B)
Meso‐level
Network (X)
Figure 1. Generic multilevel network model in MPNet.
172
Network A:
Physician‐
Physician
Knowledge
Ties
Artifact
Knowledge
Sources
Network X:
Physician‐
Artifact
Knowledge
Ties
Figure 2. Multilevel model of the sociomaterial knowledge network.
General approach to model specification
To test the hypotheses proposed in Chapters 3 and 4 using multilevel ERGM in
MPNet, model specification proceeded in three steps. First, the combination of structural
parameters representing the best-fitting models for the structure of the unipartite
physician-physician network and bipartite physician-artifact networks were identified.
Second, actor attribute parameters corresponding to the demographic variables and
research hypotheses were added to each network structural model. Third, the two
networks were estimated simultaneously, and cross-level structural and attribute
parameters were added to the combined multilevel model. During each step, when the
estimation process yielded poorly-fitting and degenerate models, I made changes to
parameter specifications based on theory and the goodness-of-fit statistics. Additionally,
for the second and third steps involving attribute parameters, I estimated two separate sets
173
of models for the unipartite, bipartite, and multilevel networks: one with the physician
credibility attribute variable, and one without it, given the suboptimal measurement of
this variable discussed previously.
Before providing a more detailed explanation of model specification, it is
important to discuss how it was based on the logic of the network social selection model.
In network modeling, a basic network dependence model examines the effect of local
structural variables (the independent variables) on the global network structure (the
dependent variable). With this basic model, the focus is restricted to endogenous
influences on network structure—what network analysts refer to as the processes of
“network self-organization” (Lusher & Robins, 2013b). The social selection model
complicates this picture by conceptualizing actor attributes as exogenous influences that
may affect the global network structure in addition to local structural variables. An
alternative network model incorporating actor attributes, the social influence model, is
also used in network modeling, and comparison of the two logics is useful in
understanding why social selection was the more appropriate model for the present
analysis. The social influence model features a conceptualization of network causality
that is opposite that of the social selection model: network structural variables are the
independent variables that predict actor attributes, which are the dependent variables.
In network modeling of cross-sectional data, the choice between a social selection
model and a social influence model should be based on relevant theory (Robins &
Daraganova, 2013). Importantly, however, the results of ERGM with cross-sectional data
cannot actually disentangle the direction of causality between variables and confirm that
the theory or theories in question are correct; that is, cross-sectional ERGM cannot
174
confirm whether the network processes at work are those of social selection or social
influence. In order to answer that latter question, longitudinal data are required. As with
the results of cross-sectional regression analysis, the results of cross-sectional network
modeling should be interpreted with the proviso that significant findings indicate an
association between actor attributes and network structure, but not a causal relationship.
I used a social selection model for this research because, in comparison to a social
influence model, it better represented my research questions, theory, and hypotheses. I
was interested in how the sociomaterial characteristics of knowledge consumers and
knowledge sources affected how consumers connected to those sources to learn about
best-practice knowledge, rather than in how consumers’ ties to knowledge sources
affected the consumers’ or sources’ sociomaterial characteristics. Applying the logic of
the social selection model to the hypotheses in the present research, I hypothesized that
communication load, source legitimacy, and source credibility were actor attribute
variables that exert sociomaterial agency in affecting the global structure of the
organizational knowledge network. In the case of communication load, I conceptualized
the construct as a subjective perception or attitude about the amount of communication in
which one engaged, and I was interested in how this attribute affected the degree to
which people obtained best-practice knowledge. I hypothesized that physicians with
greater self-perceived communication load would obtain less knowledge about best-
practice guidelines from colleagues or artifacts because such physicians will feel they
have less time, energy, and motivation to do so. This follows the logic of social selection:
the actor attribute, communication load, was hypothesized to influence the formation of
network ties, that is, communication ties between the focal physician and sources of best-
175
practice knowledge. Note, however, that if I had instead been interested in how the
number of communication ties people hold affects their self-perceptions of
communication load, I would have used a social influence model. Following the logic of
social influence, I would have hypothesized that physicians with a greater number of
communication ties would have greater self-perceived communication load—that is, that
network structure affects actor attributes.
The same points apply in the cases of the other two actor attributes of interest in
this dissertation, source credibility and source legitimacy. My hypotheses about these
variables used the logic of social selection, proposing that a source’s credibility and
legitimacy, as perceived by members of an organization, would positively influence the
number of communication ties the source receives from those organizational members. If
I had been interested in, conversely, how network structure influences a source’s
credibility and legitimacy, I would have used the logic of social influence to hypothesize
that the number of communication ties a source holds will positively influence that
source’s perceived credibility and legitimacy.
Having outlined the logic of social selection that informed my research
hypotheses and guided their translation into structural and attribute parameters for the
unipartite, bipartite, and multilevel network models, I turn now to a more detailed
description of model specification.
Parameters for network structure
In network social selection models such as the present one, in which the focus is
on how nodal attributes affect the global network structure, most endogenous local
structural processes are viewed as control variables (Robins & Daraganova, 2013). They
176
can be thought of as analogous to the nodal attribute control variables that represent the
physicians’ demographic characteristics in the model, and that are discussed in the next
section. For the unipartite physician-to-physician network and bipartite physician-to-
artifact network, I had no explicit hypotheses about network structure. For this reason,
initial selection of structural parameters for these two networks was guided by general
recommendations for structural parameter selection published in the network modeling
literature (Lusher, et al., 2013; Robins, et al., 2009; Robins, Snijders, et al., 2007;
Snijders, Pattison, Robins, & Handcock, 2006; Wang, Pattison, & Robins, 2013; Wang,
Robins, Pattison, & Lazega, 2013; Wang, et al., 2009). For example, it is standard
practice to include the Arc parameter in models of directed unipartite networks, because
this parameter represents the basic propensity for tie formation. Other local network
structures expected to play a role in determining the global structure of most unipartite
directed networks include dyadic reciprocity, nodal activity, nodal popularity, and triadic
closure, represented in MPNet by the Reciprocity, Out-Star or Alternating Out-Star, In-
Star or Alternating In-Star, and Triangle or Alternating Triangle parameters, respectively
(Wang, Robins, Pattison, & Koskinen, 2014). Similarly, local network structures
expected to play a role in determining the global structure of most bipartite networks
include the basic propensity for tie formation, nodal activity, and bipartite closure,
represented in MPNet by the Edge, Star or Alternating-Star, and 4-Cycle parameters,
respectively (Wang, et al., 2014). In the process of identifying the best-fitting structural
model for the unipartite and bipartite networks, I tested all of the above parameters, as
well as a number of others.
177
For the multilevel network, specification of network structural parameters
proceeded slightly differently. Given the recent development of software for multilevel
ERGM (Wang, Robins, Pattison, & Lazega, 2013), there are no consensus
recommendations for the selection of structural parameters for multilevel networks, as
there are for unipartite and, to a lesser extent, for bipartite networks. Therefore, I
specified the structure of the multilevel network by testing all applicable multilevel
structural parameters available in MPNet, including those that were explicitly
hypothesized as well as those that were not hypothesized. These included parameters for
affiliation-based activity and popularity; affiliation-based closure; and affiliation-based
degree assortativity, which describes the tendency for popular or active nodes to connect
to each other (Wang, Robins, Pattison, & Lazega, 2013). In MPNet, these multilevel local
network structures are represented by the In- and Out-2StarAX parameters and their
alternating star counterparts; the TXAX-Arc and TXAX-Reciprocity and their alternating
triangle counterparts; and the L3XAX and L3XAX-Reciprocity parameters, respectively
(Wang, et al., 2014). The hypotheses regarding multilevel structure, and their
corresponding model parameters, are described in more detail in the Parameters for
research hypotheses section, below.
Once converged structural models of each of the three networks were obtained,
model goodness-of-fit statistics were used to guide further refinement of network
structure. For each of the three structural models, all possible structural parameters (not
just those specified in the model estimation) were used to generate goodness-of-fit
statistics. For goodness-of-fit tests for the unipartite directed physician network, this
meant 40 possible structural parameters; for the bipartite physician-artifact network, 21
178
possible structural parameters; and for the multilevel network, 12 additional possible
structural parameters. When a parameter not specified in the model estimation had a
goodness-of-fit t-ratio of 2.0 or greater, I added it to the current model specification and
re-estimated the model. I retained the newly added parameter in the model if after
successful model estimation and convergence it met the following conditions: its
maximum likelihood estimate was statistically significant and/or its addition improved
the overall model goodness-of-fit; and if its sample autocorrelation function was less than
0.4, and its addition to the model did not cause the sample autocorrelation functions of
other parameters to increase to 0.4 or greater. For each network, the final structural model
used for hypothesis testing reflected the one with the best goodness-of-fit statistics. More
detailed descriptions of the best-fitting structural parameters in each of the final models
are provided in Chapter 6 with the network analysis results.
Parameters for demographic characteristics
After the best-fitting structural parameters were identified for each of the three
models, parameters for the physicians’ demographic characteristics were added as
attribute control variables. As elaborated above, the demographic characteristics were
gender, department chief status, clinic site, number of clinical sessions worked per week,
professional tenure, and organizational tenure. For each of the attributes except clinic
site, the same seven parameters were tested. In the unipartite physician-physician
network (Network A), sender and receiver effects were tested to determine if physicians
with the attribute in question were more or less likely to seek or receive best-practice
knowledge from their colleagues, and to be a source of information for their colleagues,
respectively. A homophily effect was also tested in the unipartite network to determine if
179
two physicians with the same attribute or similar levels of the attribute were more or less
likely to exchange knowledge. For the bipartite physician-artifact network (Network X),
a sender effect was tested to determine if physicians with the attribute of interest were
more or less likely to seek or receive knowledge from artifacts, and a receiver effect was
tested to determine if knowledge sources with the attribute of interest were more or less
likely to be used by physicians. For the multilevel network (Network M), multilevel
versions of the sender, receiver, and homophily effects were again tested.
Note that unipartite and multilevel homophily effects were not tested for the
department chief attribute, because no department chiefs reported seeking or receiving
best-practice knowledge from another chief—a point discussed further in Chapter 6.
Additionally, for the clinic site variable, only one effect was tested, a homophily-based
affiliation effect for Network X, which examined if physicians working at the same clinic
site were more likely to seek or receive knowledge from the same artifacts. Table 2 (see
page 180) presents detailed descriptions and visualizations of the attribute parameters that
were used to represent the demographic characteristics in the three network models.
180
Table 2. Model Parameters for Demographic Characteristics
Net Parameter MPNet Name Description Visualization
Gender
A
Attribute-based
sender activity
Gender_SenderA
Are men/women more/less likely to seek/receive knowledge
from colleagues?
1
A
Attribute-based
receiver popularity
Gender_ReceiverA
Are men/women more/less likely to be sources of knowledge for
colleagues?
1
A Homophily Gender_InteractionA
Is there a gender homophily effect among knowledge
seekers/recipients and sources?
1 1
X
Attribute-based
sender activity
Gender_XEdgeA
Are men/women more/less likely to seek/receive knowledge
from artifacts?
1
M
Attribute-based
sender
centralization
Gender_
Star2AXSender
Gender_SenderA and Gender_XEdgeA, combined. Are
men/women more/less likely to seek/receive knowledge from
both colleagues and artifacts?
1
M
Attribute-based
receiver
centralization
Gender_
Star2AXReceiver
Gender_ReceiverA and GenderXEdgeA, combined. Are
men/women more/less likely to both be sources of knowledge for
their colleagues and to seek/receive knowledge from artifacts?
1
181
Net Parameter MPNet Name Description Visualization
M
Homophily-based
triadic closure
Gender_
TXAXInteractionArc
Are physicians of the same gender more likely to seek/ receive
knowledge from each other and from the same artifact?
1 1
Department Chief
1
A
Attribute-based
sender activity
Chief_SenderA
Are chiefs more likely to seek/receive knowledge from
colleagues?
1
A
Attribute-based
receiver popularity
Chief_ReceiverA
Are chiefs more likely to be sources of knowledge for
colleagues?
1
X
Attribute-based
sender activity
Chief_XEdgeA Are chiefs more likely to seek/receive knowledge from artifacts?
1
M
Attribute-based
sender
centralization
Chief_Star2AXSender
Chief_SenderA and Chief_XEdgeA, combined. Are chiefs more
likely to seek/receive knowledge from both colleagues and
artifacts?
1
182
M
Attribute-based
receiver
centralization
Chief_
Star2AXReceiver
Chief_ReceiverA and Chief_XEdgeA, combined. Are chiefs
more likely to both be sources of knowledge for colleagues and
to seek/receive knowledge from artifacts?
1
Clinic Site
X
Homophily-based
affiliation
Clinic_X2StarAMatch
Are physicians who work at the same clinic site more likely to
seek/receive knowledge from the same artifacts?
Clinical Sessions
A
Attribute-based
sender activity
Clinical Sessions_
SenderA
Are physicians who work more clinical sessions more likely to
seek/receive knowledge from colleagues?
A
Attribute-based
receiver popularity
Clinical Sessions_
ReceiverA
Are physicians who work more clinical sessions more likely to be
sources of knowledge for colleagues?
A Homophily
Clinical Sessions_
DifferenceA
Is there a clinical sessions homophily effect among knowledge
seekers/recipients and sources?
X
Attribute-based
sender activity
Clinical Sessions_
XEdgeA
Are physicians who work more clinical sessions more likely to
seek/receive knowledge from artifacts?
183
Net Parameter MPNet Name Description Visualization
M
Attribute-based
sender
centralization
Clinical Sessions_
Star2AXSender
ClinicalSessions_SenderA and ClinicalSessions_XEdgeA,
combined. Are physicians who work more clinical sessions more
likely to seek/receive knowledge from both colleagues and
artifacts?
M
Attribute-based
receiver
centralization
Clinical Sessions_
Star2AXReceiver
ClinicalSessions_ReceiverA and ClinicalSessions_ XEdgeA,
combined. Are physicians who work more clinical sessions more
likely to both be sources of knowledge for their colleagues and to
seek/receive knowledge from artifacts?
M
Homophily-based
triadic closure
Clinical Sessions_
TXAXDiffArc
Are physicians who work more clinical sessions more likely to
seek/receive knowledge from each other and from the same
artifact?
−
Professional Tenure
A
Attribute-based
sender activity
ProfessionalTenure_
SenderA
Are physicians with greater professional tenure more likely to
seek/receive knowledge from colleagues?
A
Attribute-based
receiver popularity
ProfessionalTenure_
ReceiverA
Are physicians with greater professional tenure more likely to be
sources of knowledge for colleagues?
184
Net Parameter MPNet Name Description Visualization
A Homophily
ProfessionalTenure_
DifferenceA
Is there a professional tenure homophily effect among knowledge
seekers/recipients and sources?
X
Attribute-based
sender activity
ProfessionalTenure_
XEdgeA
Are physicians with greater professional tenure more likely to
seek/receive knowledge from artifacts?
M
Attribute-based
sender
centralization
ProfessionalTenure_
Star2AXSender
ProfessionalTenure_SenderA and Professional Tenure_XEdgeA,
combined. Are physicians with greater professional tenure more
likely to seek/receive knowledge from both colleagues and
artifacts?
M
Attribute-based
receiver
centralization
ProfessionalTenure_
Star2AXReceiver
ProfessionalTenure_ReceiverA and Professional
Tenure_XEdgeA, combined. Are physicians with greater
professional tenure more likely to both be sources of knowledge
for their colleagues and to seek/receive knowledge from
artifacts?
M
Homophily-based
triadic closure
ProfessionalTenure_
TXAXDiffArc
Are physicians with similar professional tenures more likely to
seek/receive knowledge from each other and from the same
artifact?
−
185
Net Parameter MPNet Name Description Visualization
Organizational Tenure
A
Attribute-based
sender activity
OrganizationalTenure_
SenderA
Are physicians with greater organizational tenure more likely to
seek/receive knowledge from colleagues?
A
Attribute-based
receiver popularity
Organizational Tenure_
Receiver A
Are physicians with greater organizational tenure more likely to
be sources of knowledge for colleagues?
A Homophily
Organizational Tenure_
DifferenceA
Is there an organizational tenure homophily effect among
knowledge seekers/recipients and sources?
X
Attribute-based
sender activity
Organizational Tenure_
XEdgeA
Are physicians with greater organizational tenure more likely to
seek/receive knowledge from artifacts?
M
Attribute-based
sender
centralization
Organizational Tenure_
Star2AXSender
ClinicalSessions_SenderA and ClinicalSessions_ XEdgeA,
combined. Are physicians with greater organizational tenure
more likely to seek/receive knowledge from both colleagues and
artifacts?
M
Attribute-based
receiver
centralization
Organizational Tenure_
Star2AXReceiver
OrganizationalTenure_ReceiverA and Organizational
Tenure_XEdgeA, combined. Are physicians with greater
organizational tenure more likely to both be sources of
knowledge for colleagues and to seek/receive knowledge from
artifacts?
186
Net Parameter MPNet Name Description Visualization
M
Homophily-based
triadic closure
Organizational Tenure_
TXAXDiffArc
Are physicians with similar organizational tenures more likely to
seek/receive knowledge from each other and from the same
artifact?
−
Note. In the visualizations, physicians are represented by blue squares, and artifacts by red circles.
1
Unipartite and multilevel homophily effects were not tested for the department chief variable because no department chiefs reported seeking or
receiving best-practice knowledge from another chief.
187
Parameters for research hypotheses
Following the addition of parameters for the demographic characteristics,
parameters representing the research hypotheses were added to the models. There were
three sets of hypotheses: the first focused on communication load, the second on
legitimacy and credibility, and the third on the value of the overall sociomaterial model.
Table 1 (see page 126) presents a summary of the hypotheses, the network(s) to which
each hypothesis applies, and the names and visualizations of the corresponding MPNet
parameters. In the remainder of this section, the translation of research hypotheses to
ERGM parameters is explained in more detail.
The first set of hypotheses focused on the effects of physicians’ perceptions of
communication load on the knowledge network. Hypothesis 1 proposed that physicians
with higher levels of perceived communication load would be less likely to seek or
receive best-practice knowledge from their colleagues and from artifacts. For the
unipartite physician-to-physician and bipartite physician-to-artifact networks, this
hypothesis was tested with parameters (Load_SenderA and Load_XEdgeA, respectively)
assessing attribute-based sender activity; that is, whether actors with a particular attribute
are more or less likely to “send” a tie to another actor. For the multilevel network, this
hypothesis was tested with an attribute-based sender centralization parameter
(Load_Star2AXSender) assessing cross-level attribute-based sender activity.
Hypothesis 2 proposed that physicians would be more likely to seek or receive
best-practice knowledge from colleagues with ties to one or more artifact knowledge
sources—that is, colleagues who were potential brokers to one or more artifacts. This
hypothesis was based on the premise that people tend to prefer human sources of
188
knowledge over artifact sources, especially in situations where they perceive themselves
to be at risk for communication overload. This hypothesis was tested in the multilevel
network with two structural parameters for affiliation-based popularity. The first,
In2StarAX, tested the basic proposition that physicians with ties to one or more artifacts
were more likely to also be identified as a knowledge source by a colleague. The second
parameter, AAinS1X, tested the proposition that such physicians were more likely to be
identified as a source by multiple colleagues; that is, that they would be significantly
more popular knowledge sources than physicians without ties to artifacts.
Hypothesis 3 proposed that physicians who seek or receive knowledge from one
or more artifacts were more likely to have ties to colleagues who were also connected to
the same artifact(s). This hypothesis was based on the old premise that a friend of a friend
is a friend, translated to the knowledge network context. It proposed that people tend to
prefer potential knowledge sources to which one of their existing knowledge sources is
already connected (i.e., they prefer knowledge sources with third-party certification)—
especially in situations where they perceive themselves to be at risk for communication
overload. This hypothesis was tested in the multilevel network with structural parameters
for affiliation-based closure (which could also be viewed as affiliation-based homophily):
TXAXarc, for the case in which two physicians have triadic closure with a single artifact,
and its alternating counterpart, ATXAXarc, for the closure-with-multiple-artifacts case.
Note that this hypothesis and its corresponding parameter make no assumptions about the
timing of tie formation in these triadic structures: in the cross-sectional ERGM context,
we cannot know if the ego physician first sought best-practice knowledge from an
artifact, then discovered a colleague was also using that same artifact, and so
189
subsequently sought knowledge from that colleague; or alternatively, if the ego physician
first sought knowledge from a colleague, then discovered that the colleague was using an
artifact, and so subsequently sought knowledge from that same artifact. Such time
sequencing in triadic structures can only be disentangled through longitudinal network
data and modeling.
Hypothesis 4 proposed that physicians would be more likely to receive best-
practice knowledge from mass media, rather than niche media, artifacts, and that
physicians with higher perceived communication load would be particularly more likely
to do so. Hypothesis 4 was tested in the bipartite network with an attribute-based receiver
popularity parameter, Mass Media_XEdgeB, and an attribute-based sender-receiver
interaction parameter, Load-Mass Media_XEdgeAB. The former tests the simpler
proposition that all physicians will tend to prefer mass media artifacts over niche media
artifacts, and the latter tests the proposition that overloaded physicians will be
particularly likely to exhibit this preference.
The second set of hypotheses focused on how knowledge sources’ perceived
legitimacy and credibility affected the structure of the overall network. Hypothesis 5
proposed that physicians would be more likely to seek or receive knowledge from
artifacts with greater legitimacy. It was tested in the bipartite network with a parameter,
Artifact Legitimacy_XEdgeB, representing attribute-based receiver popularity; that is,
whether actors with a particular attribute—in this case, greater legitimacy—are more or
less likely to “receive” a tie from another actor.
Hypothesis 6 proposed that physicians would be more likely to seek or receive
best-practice knowledge from their colleagues and from artifacts if they perceived the
190
legitimacy of the “author” of the best-practice guidelines to be greater. In other words,
the more legitimate a physician perceives the author of the guidelines to be, the more
likely he or she is to seek or be receptive to receiving knowledge about those guidelines.
“Author” here refers to the original authors of the best practices, rather than secondary
sources of knowledge about them. In the case of the cholesterol guidelines, the original
authors were organizations: the ACC and AHA. This hypothesis was tested in the
unipartite and bipartite networks with attribute-based sender activity parameters, ACC
Legitimacy_SenderA and AHA Legitimacy_SenderA. In the multilevel network it was
tested with parameters for attribute-based sender centralization, ACC
Legitimacy_Star2AXSender and AHA Legitimacy_Star2AXSender.
Hypotheses 7 and 8 repeat, for credibility, Hypotheses 5 and 6. Hypothesis 7
proposed that physicians would be more likely to seek or receive knowledge from
colleagues and artifacts with greater credibility. It was tested with attribute-based receiver
popularity parameters: a Physician Credibility_ReceiverA parameter in the unipartite
network and, as in Hypothesis 5 for legitimacy, an Artifact Credibility_XEdgeB
parameter in the bipartite network. Hypothesis 8 proposed that the more credible a
physician perceives the author of best-practice guidelines to be, the more likely that
physician is to seek, or be receptive to receiving, knowledge about those guidelines. It
was tested with parameters for attribute-based sender activity in the unipartite and
bipartite networks: ACC Credibility_SenderA and AHA Credibility_SenderA, and ACC
Credibility_XEdgeA and AHA Credibility_XEdgeA, respectively. In the multilevel
network, Hypothesis 8 was tested with parameters for attribute-based sender
centralization: ACC Credibility_Star2AXSender and AHA Credibility_Star2AXSender.
191
After all structural and attribute parameters were added for both the demographic
characteristics and for the research hypotheses, model goodness-of-fit statistics were
again used to guide further refinement of each of the network models. As with the
structure-only models, all plausible structural and attribute parameters (not just those
specified in the model estimation) were used to generate goodness-of-fit statistics for
each of the three networks. For goodness-of-fit tests for the unipartite physician-
physician network, this meant 40 structural parameters and 102 attribute parameters; for
the bipartite physician-artifact network, 21 structural parameters and 83 attribute
parameters; and for the multilevel network, 12 additional structural parameters and 88
additional attribute parameters. For the overall multilevel model, then, 346 parameters
were included in the goodness-of-fit testing. (For the versions of the three models that
included the physician credibility variable, an additional 11, 6, and 10 attribute
parameters, respectively, were included in the goodness-of-fit tests.) When a parameter
not specified in the model estimation had a goodness-of-fit t-ratio of 2.0 or greater, I
added it to the current model specification and re-estimated the model. As before, I
retained the newly added parameter in the model if after successful model estimation and
convergence it met the following conditions: its maximum likelihood estimate was
statistically significant and/or its addition improved the overall model goodness-of-fit;
and if its sample autocorrelation function was less than 0.4, and its addition to the model
did not cause the sample autocorrelation functions of other parameters to increase to 0.4
or greater.
Finally, unlike the preceding hypotheses, Research Question 1 and Hypothesis 9
were not tested directly by specific model parameters estimated in the multilevel ERGM,
192
but rather by comparing differences in the significance of individual parameter estimates
and in the goodness of fit statistics of the three network models following completion of
model estimation. Research Question 1 asked if there was evidence of an empirical
difference between legitimacy and credibility, as manifested in the ERGM results in two
ways. The first way (RQ1a) involved comparing the significance of the parameter
estimate for legitimacy-based artifact source popularity (Artifact Legitimacy_XEdgeB,
H5) with the significance of the parameter estimates for credibility-based physician and
artifact source popularity (Physician Credibility_ReceiverA and Artifact
Credibility_XEdgeB, H7). The second way (RQ1b) involved comparing the parameter
estimates for sender activity based on physician perception of the legitimacy of the best-
practice authors (ACC/AHA Legitimacy_SenderA, ACC/AHA Legitimacy_XEdgeA, and
ACC/AHA Legitimacy_Star2AXSender; H6) with the parameter estimates for sender
activity based on physician perception of the credibility of the best-practice authors
(ACC/AHA Credibility_SenderA, ACC/AHA Credibility_XEdgeA, and ACC/AHA
Credibility_Star2aXSender; H8). In both cases, if the legitimacy parameters were
significant and the credibility estimates were not significant, or vice versa, these results
would offer evidence of an empirical difference between the two constructs. If the
legitimacy and credibility parameters were both significant, or both not significant, there
would be no evidence of an empirical difference.
Hypothesis 9 proposed that the multilevel sociomaterial network model would
explain more of the variance in the structure of the knowledge network than either the
unipartite “social” network model or the bipartite “material” network model alone.
Following Wang and colleagues (2013; 2013), this hypothesis was tested in two steps.
193
First, differences in the significance of the individual parameter estimates were compared
across the three models. If the addition of significant cross-level parameters to the overall
multilevel model caused previously-significant unipartite and bipartite parameters to lose
their statistical significance, these results might suggest that the multilevel model
explained more variance than the other models. In Wang and colleagues’ (2013; 2013)
empirical study, for example, inclusion of cross-level parameters in the overall model
greatly simplified the previous specifications of two unipartite models estimated
separately. When cross-level parameters were included in their overall model, a number
of unipartite parameters that were significant when the networks were modeled separately
became non-significant.
Second, the goodness-of-fit statistics were compared across the three models.
Better goodness of fit for the multilevel model than for the other models would suggest
that the former explains more variance than the latter. For Wang and colleagues (2013;
2013), model fit statistics were best for the overall multilevel network model, as
compared to those for the two unipartite network models, leading the authors to conclude
the multilevel model offered the best explanation for the structure of the network.
194
CHAPTER 6: RESULTS
In the first half of this chapter I review the descriptive results and the results of
the preliminary analyses. I begin by providing the descriptive statistics for the study
population, and the results of the analysis of non-response bias for the collected data. I
then offer the results of the scale reliability and confirmatory factor analyses for the three
scaled actor attribute variables, physician communication load, knowledge source
legitimacy, and knowledge source credibility. I also provide descriptive statistics for the
structure of the knowledge network and for the actor attributes featured in the research
hypotheses.
In the second half of the chapter, I review the results of the network analysis. I
recount the results for the unipartite physician-physician network and bipartite physician-
artifact network, considered separately; and then review the results of the combined
multilevel network.
Descriptive Statistics for Study Population
The study population consisted of 185 primary care physicians working in 17
clinics. The majority of these physicians were trained in internal medicine (97%) rather
than family medicine, and had MD degrees (95%) rather than DO degrees. In each clinic,
the number of primary care physicians ranged from 3 to 21, with a mean of 11. One
primary care physician in each clinic served as the primary care department chief, with
two exceptions: in one clinic, the department chief was a physician assistant, rather than a
physician, and so was not included in the study population; and in one clinic the
department chief responsibilities were shared by two primary care physicians. Therefore
the total number of department chiefs in the study population was 17.
195
Of the 185 primary care physicians, 131 returned the survey instrument for an
overall response rate of 71%. None of the returned surveys were excluded from the data
analysis, and data on 12 non-respondents were included in the analysis because these
non-respondent physicians were nominated by respondents as knowledge sources.
Therefore the total number of primary care physicians included in the analysis was 143,
or 77% of the original study population.
Across the 17 clinics, the intra-clinic response rate ranged from 38% to 100%,
with mean and median intra-clinic response rates of 70%. Of the 17 primary care
physician department chiefs, 15 completed the survey for a department chief response
rate of 88%.
Table 3. Descriptive Statistics for Demographic Characteristics
Characteristic Measure MinMax MSD Missing
1
Gender (1 = woman) Binary 0 1 0.550 0.499 0
Women = 79 (55%)
Men = 64 (45%)
Department chief (n = 15; 10%) Binary 0 1 0.120 0.325 0
Professional tenure (years) Continuous 5 55 25.780 10.515 0
Organizational tenure (years) Continuous 0
2
3912.95010.233 0
Clinical sessions/week Continuous 1 10 6.840 1.830 7 (5%)
Note. For these data, n = 143, the number of physicians included in the data analysis.
1
The Missing statistic reflects the number and percentage of item-missing data from respondents
only (n = 131), not the total study population (n = 185) or those included in the network for data
analysis (n = 143). The only demographic characteristic for which there were data missing was
the Clinical sessions variable. Descriptive statistics for this variable were calculated after the
imputation of missing values.
2
An organizational tenure of 0 indicates the physician had worked at Alpha Medical Group for
less than 1 year.
Table 3 (page 195) presents a summary of the demographic characteristics of the
143 physicians included in the data analysis. Fifty-five percent were women. Average
196
professional tenure and average organizational tenure were both approximately 10 years,
although there was substantial range in both tenure variables, indicating a relatively
heterogeneous population in terms of tenure, and by implication, physician age. Average
number of clinical sessions worked per week—a measure of the amount of time the
physicians spent on patient care—was approximately 7, which translates to 3.5 days per
week.
Preliminary Analyses
Analysis of non-response bias
Survey non-response bias was assessed by a multivariate analysis of variance
(MANOVA) comparing the clinic affiliation and demographic characteristics of
respondents—gender, department chief status, organizational tenure, and professional
tenure—to those of non-respondents. Results indicated that the differences between the
two groups were just beyond significance (Wilks’ Λ = .94, F(5, 179) = 2.25, p = .051),
with 6% of the multivariate variance in the demographic characteristics being associated
with whether a physician in the study population was a respondent or non-respondent
(multivariate η
2
= .06). Because of the borderline non-significance in the differences
between respondents and non-respondents in the MANOVA, individual analyses of
variance (ANOVAs) were also conducted on each variable to determine if there were
significant differences between the groups on one of more of the variables when
considered individually. Using the Bonferroni method, each ANOVA was tested at the
.01 level. These univariate ANOVAs indicated no significant differences between
respondents and non-respondents. Taken together, the results of the MANOVA and
197
univariate ANOVAs suggested negligible differences in the demographic characteristics
of respondents and non-respondents.
Scale reliability and confirmatory factor analysis
The factor structure of the three scaled independent variables—communication
load, legitimacy, and credibility—were assessed by Cronbach’s coefficient alpha
calculations and confirmatory factor analyses. Following imputation of the missing data,
the number of cases for the factor analyses was 143 for communication load (i.e., the
number of physicians included in the network analysis), 642 for credibility (i.e., the
number of physician-physician dyads plus physician-artifact dyads in the network
analysis), and 540 for legitimacy (i.e., the number of physician-artifact dyads in the
network analysis).
As discussed in Chapter 5, communication load was measured using 7 items, with
previous empirical research indicating that those items share one common dimension.
The Cronbach’s alpha for this 7-item scale was 0.89, indicating satisfactory reliability,
with no projected improvement in alpha if any of the items were deleted. This
Cronbach’s alpha value compares favorably to other alpha values reported in the
literature for very similar or identical scales, which range from 0.72 to 0.93 (Ballard &
Seibold, 2006; Chen & Lee, 2013; Cho, et al., 2011; Karr-Wisniewski & Lu, 2010;
Stephens & Davis, 2009). The 7-indicator, 1-factor model specified for the confirmatory
factor analysis was over-identified with 14 degrees of freedom. Global goodness of fit
statistics suggested that the hypothesized model produced a poor fit of the data (see
Appendix B, Table B2). Local indicators of sources of poor fit were inspected to identify
any theoretically justifiable changes to the model. All parameter estimates had
198
statistically significant factor loadings, with small standard errors and appropriate values.
The highest standardized residual (2.762) and modification index value (45.166) was for
the relationship between Item 1 and Item 2, and for theta-delta (2,1), respectively.
Additionally, Item 1 displayed the lowest standardized parameter estimate and the lowest
bivariate correlations with the other items in the scale, and the Cronbach’s alpha analysis
indicated no change in reliability of the scale if this item was deleted. Item 1, “How often
do you feel you have to engage in too much communication…?” is different from the
other scale items in that it refers only to communication quantity, without making any
qualifications about quality, whereas the other six items refer to both quantity and quality
(see Appendix B, Table B1), which the theoretical literature suggests are equally
important aspects of communication load. Based on these statistical indicators and
theoretical considerations, it seemed plausible that Item 1 was a poor indicator of
communication load, and therefore Item 1 was removed from the scale.
The respecified 6-item, 1-factor model, over-identified with 9 degrees of freedom,
was subsequently estimated. Global goodness of fit statistics for Model 2 suggested an
improved but still marginal fit of the data (see Appendix B, Table B2). Local indicators
of sources of poor fit were inspected to identify any theoretically-justifiable changes to
the model. All parameter estimates had statistically significant factor loadings, with small
standard errors and appropriate values. The highest standardized residual (2.181) and
modification index value (11.793) was for the relationship between Item 2 and Item 3,
and for theta-delta (2,1) (referring to the error covariance between Items 3 and 2),
respectively. Item 2 asks, “How often do you receive more information than you can
process?”, and Item 3 asks, “How often do you receive more information than you need
199
in order to do your job effectively?” Given the similarity in the wording of these two
items, it is possible that survey respondents experienced difficulty distinguishing between
them, leading to a correlation between the items’ measurement errors. When there is a
plausible empirical reason for measurement error correlation, such as this, a model
modification is recommended (Bollen, 1989; T. A. Brown, 2006). Based on these
statistical and empirical considerations, the model was thus respecified with a correlation
between the measurement errors of Item 2 and Item 3.
The respecified 6-item, 1-factor model, Model 3, over-identified with 8 degrees of
freedom, was subsequently estimated. Global goodness of fit statistics for Model 3
suggested an improved and satisfactory fit of the data. Although the Satorra-Bentler
scaled chi-square value remained non-significant (16.649, p = 0.034) and the RMSEA
value (0.080) was above the 0.06 cut-off criterion, the remaining fit indices suggested
satisfactory fit (SRSMR = 0.038; CFA = 0.990; RMSEA 90% CI = 0.000 – 0.140, p =
0.134). Additionally, given that the chi-square and RMSEA indices are particularly likely
to erroneously reject good-fitting models with sample sizes of 250 or less (for the
communication load model, n = 143), and given that the RMSEA p value for close fit
was significant, the global fit of the model was judged satisfactory. Further, all parameter
estimates had statistically significant factor loadings, with small standard errors and
appropriate values, and inspection of the standardized residuals and modification indices
suggested no additional theoretically or empirically justifiable changes to the model. The
factor loading estimates suggested strong communalities between the indictors and their
respective factors, with R
2
values ranging from 0.451 to 0.853. Appendix B presents the
polychoric correlation matrix, standard deviations, and means of the communication load
200
indicators used in the third and final model (Table B1); a comparison of the global
goodness of fit statistics for the three models that were estimated (Table B2); and the
unstandardized and standardized parameter estimates of the final model (Figure B1).
Legitimacy was measured with six items underlying two dimensions, moral
legitimacy and regulatory legitimacy. The Cronbach’s alpha for this 6-item scale was
0.81, indicating satisfactory reliability, with no projected improvement in alpha if any of
the items were deleted. Although this alpha value is lower than that of the other two
scales, it falls within the minimum accepted rule of thumb of at least 0.80. This
Cronbach’s alpha value falls between other alpha values for legitimacy reported in the
extant literature (0.89 and 0.78, respectively, for Certo & Hodge, 2007; J. Y. Chung, et
al., 2011). The 6-indicator, 2-factor model specified for the confirmatory factor analysis
was over-identified with 8 degrees of freedom. Global goodness of fit statistics suggested
that the hypothesized model produced a satisfactory fit of the data (see Appendix C,
Table C2). Local indicators of sources of poor fit were inspected to identify any
theoretically-justifiable changes to the model. All parameter estimates had statistically
significant factor loadings, with small standard errors and appropriate values. The highest
standardized residual (2.268) was for the relationship between Item 1 and Item 2, and the
two highest modification index values (20.171 and 21.039) were for theta-delta (2,1) and
theta-delta (6,4), respectively. Item 1 states, “People think this organization is valuable to
our society,” and Item 2 states, “The general public would approve of this organization if
asked their opinion.” Given the similarity in the meaning of these two items, intended as
indicators of moral legitimacy, it is possible that survey respondents experienced
difficulty distinguishing between them, leading to a correlation between the items’
201
measurement errors. With regard to Items 4 and 6, Item 4 states, “This organization
always follows laws and regulations,” and Item 6 states, “This organization’s leaders
believe in ‘playing by the rules’ and following accepted operating guidelines.” Again,
given the similarity in the meaning of these two items, intended as indicators of
regulatory legitimacy, it is possible that survey respondents had trouble distinguishing
between them, leading to a correlation between the items’ measurement errors. As in the
case for the communication load scale, the model was thus respecified with correlations
between the measurement errors of Items 1 and 2, and Items 4 and 6, added one at a time
to respecified Models 2 and 3, respectively.
Models 2 and 3 were over-identified with 7 and 6 degrees of freedom,
respectively. Global goodness of fit statistics suggested that both models produced an
improved and satisfactory fit of the data, with the indices for Model 3 suggesting a
slightly better fit than those for Model 2 (see Appendix C, Table C2). For Model 3, the
goodness of fit statistics were as follows: Satorra-Bentler scaled chi-square = 13.632, p =
0.034; SRSMR = 0.021; CFA = 0.996; RMSEA = 0.052, 90% CI = 0.0178 – 0.086, p =
0.473. All parameter estimates for Model 3 had statistically significant factor loadings,
with small standard errors and appropriate values, and inspection of the standardized
residuals and modification indices suggested no additional theoretically or empirically
justifiable changes to the model. The factor loading estimates suggested strong
communalities between the indictors and their respective factors, with R
2
values ranging
from 0.553 to 0.697. The standardized parameter estimate of the covariance between the
two factors, 0.738, suggested a strong relationship between the two hypothesized
dimensions of legitimacy, moral legitimacy and regulatory legitimacy. Appendix C
202
presents the polychoric correlation matrix, standard deviations, and means of the
legitimacy indicators used in the final model (Table C1); a comparison of the global
goodness of fit statistics across the three models (Table C2); and the unstandardized and
standardized parameter estimates for the final model (Figure C1).
Credibility was measured with six items underlying two dimensions, expertise
and trustworthiness. The Cronbach’s alpha for this 6-item scale was 0.90, indicating
satisfactory reliability, with a projected improvement to 0.91 if Item 5 for the
trustworthiness dimension, “This source is unbiased” was deleted. The Cronbach’s alpha
value of 0.90 compares favorably to other alpha values reported in the literature for very
similar or identical scales, which range from 0.77 to 0.85 (Flanagin & Metzger, 2000,
2007; McCroskey & Teven, 1999; Westerwick, 2012). For this reason, and because of the
theoretical and empirical support for Item 5 in the extent literature, Item 5 was not
deleted. The 6-indicator, 2-factor model specified for the confirmatory factor analysis
was over-identified with 8 degrees of freedom. Global goodness of fit statistics suggested
that the hypothesized model produced a satisfactory fit of the data: Satorra-Bentler scaled
chi-square = 29.281, p = 0.000; SRSMR = 0.028; CFA = 0.996; RMSEA = 0.065, 90%
CI = 0.041 – 0.090, p = 0.150. All parameter estimates had statistically significant factor
loadings, with small standard errors and appropriate values, and inspection of the
standardized residuals and modification indices suggested no theoretically or empirically
justifiable changes to the model. The factor loading estimates suggested strong
communalities between the indictors and their respective factors, with R
2
values ranging
from 0.531 to 0.955. The standardized parameter estimate of the covariance between the
two factors, 0.864, suggested a strong relationship between the two hypothesized
203
dimensions of credibility, expertise and trustworthiness. Appendix D presents the
polychoric correlation matrix, standard deviations, and means of the communication load
indicators used in the model (Table D1); a summary of the global goodness of fit
statistics (Table D2); and the unstandardized and standardized parameter estimates
(Figure D1).
Given the results of the factor analyses, Item 1 was removed from the
communication load variable used in the network analysis, whereas no changes were
made to the composition of the credibility and legitimacy variables.
Descriptive Statistics for Actor Attributes
Descriptive statistics for the actor attributes that were the focus of the research
hypotheses—physician communication load, knowledge source legitimacy, and
knowledge source credibility—are presented in Table 4 (see page 205).
With regard to communication load, on a scale of 1 to 5, with 1 representing
“never” and 5 representing “always,” the average frequency with which physicians
reported experiencing overload was about 3.5. This translates to physicians experiencing
overload somewhere between “sometimes” and “frequently” in a typical work week.
Recall from Chapter 5 that Hypothesis 4b, regarding the interaction of communication
load and artifact mass media status, required a dichotomized version of the
communication load variable. The communication load scores were split into two
categories, with the split set at the sample median. The median was 3.33, so for the binary
version of communication load, scores above or equaling 3.33 were coded as “1,” and
scores below 3.33 were coded as “0.”
204
With regard to knowledge source legitimacy and credibility, the mean scores for
all forms of legitimacy and credibility ranged from approximately 5.7 to 6.3 on a 7-point
scale. Artifact legitimacy received the lowest mean score and ACC credibility received
the highest. The standard deviations were greatest for artifact legitimacy, as well as for
artifact credibility (which was also the variable that had the lowest minimum score) and
physician credibility. The scores for ACC and AHA legitimacy and credibility were
relatively uniform, with ACC credibility receiving the highest average rating of those
four variables. Overall, respondents perceived their colleagues and artifacts to possess, on
average, relatively equivalent degrees of credibility as knowledge sources. They
perceived the legitimacy of specific artifacts containing best-practice knowledge to be, on
average, slightly less than that of the ACC and AHA, the organizational authors of those
best practices.
Recall from Chapter 5 that it was not possible to collect credibility ratings on all
185 physicians in the study population, and that as a result, the credibility ratings were
missing for physicians who were seekers or recipients of knowledge, but who were not
nominated as knowledge sources. To calculate the “missing” credibility scores for these
physicians, the standard deviation of the physician credibility scores (prior to imputation)
was subtracted from the lowest physician credibility score, and the amount of the
resulting difference was used as the credibility score for all non-source physicians. Prior
to imputation, the lowest score was 4.00 and the standard deviation was 0.72, so the value
used to replace all of the “missing” credibility ratings was 3.28.
For the actor attribute data collected via survey instrument, descriptive statistics
on the quantity of missing data are also presented in Table 4. The amount of data missing
205
for each of the credibility variables was higher (15% - 18%) than that for the other
variables (2% - 10%). Because of a technical malfunction, Qualtrics, the online survey
tool used to collect the data, lost much of the credibility data for the first 25 survey
respondents, resulting in this greater amount of missing data for the credibility variables.
Table 4. Descriptive Statistics for Actor Attributes
Attribute Measure MinMax M SD Missing
1
All physicians
Communication load Continuous 2 5 3.536 0.675 51 (6%)
ACC legitimacy Continuous 4.333 7.000 6.003 0.690 75 (10%)
ACC credibility Continuous 4.000 7.000 6.312 0.617 138 (18%)
AHA legitimacy Continuous 4.333 7.000 6.057 0.655 81 (10%)
AHA credibility Continuous 4.000 7.000 6.280 0.632 141 (18%)
Physician sources
Credibility Continuous 3.8337.0006.067 0.819 91 (15%)
Artifact sources
Mass media Binary 0 1 0.268 0.447 0
Mass media = 15 (27%)
Niche media = 41 (73%)
Legitimacy Continuous 2.0007.0005.696 0.934 36 (2%)
Credibility Continuous 1.8337.0006.144 0.996 227 (15%)
Note. For these data, n = 143, the number of physicians included in the data analysis. Descriptive
statistics for all variables were calculated after the imputation of missing values (with the
exception of the Mass media variable, for which there were no missing values).
1
The Missing statistic reflects the number and percentage of item-missing data from respondents
only (n = 131), not the total study population (n = 185) or those included in the network for
ERGM analysis (n = 143). For communication load, ACC legitimacy and credibility, and AHA
legitimacy and credibility, the denominator was 131 × 6 (the number of items used to measure the
variable) = 786. For physician credibility, the denominator was 102 (the number of physician
nominations made by respondents) × 6 = 612, and for artifact credibility and legitimacy, the
denominator was 254 (the number of artifact nominations made by respondents) × 6 = 1,524.
Also included in Table 4 are descriptive statistics for the dichotomization of the
knowledge artifacts into mass and niche media sources. Survey respondents identified 56
206
unique artifacts from which they had learned about the 2013 cholesterol best-practice
guidelines. Of these, the majority (73%) were niche media artifacts. Appendix E provides
a complete list of all of the artifacts that were nominated, the number of nominations each
received, and their classification as mass or niche media sources.
Several explanatory notes regarding the artifact nominations presented in
Appendix E are merited here. First, the most popular artifact among the physician
respondents, with 41 nominations, was UpToDate, a commercial encyclopedia of best-
practice knowledge to which Alpha Medical Group subscribes and to which all
physicians employed by Alpha have access. At Alpha, the UpToDate database is
integrated into its electronic medical record system, such that physicians can access
UpToDate while entering notes regarding patient visits, answering patient e-mail
messages, and ordering prescriptions, diagnostic tests, and other procedures. In the health
care industry, UpToDate is referred to as an electronic decision support system, and such
systems have been advocated as useful tools for the communication of best-practice
knowledge. For example, the Institute of Medicine (2011) concluded that “incorporation
of reminders and clinical care algorithms into electronic decision support systems holds
great promise to promote use of CPGs [clinical practice guidelines]” (p. 149).
Second, in addition to UpToDate, seven other artifacts received more than 10
nominations. These additional very popular artifacts were either niche medical journals or
publications (n = 5), or mass media newspapers (n = 2). The ACC/AHA cholesterol
guidelines document itself received 6 nominations; the two medical journals that jointly
published the guidelines, the Journal of the ACC and Circulation, received 14 and 6
207
nominations, respectively; and the ACC, the AHA, their organizational Web sites, and
their jointly-produced mobile app for the guidelines received a total of 9 nominations.
Third, in cleaning the artifact nomination data, I chose to include all nominations
made by respondents, even if their classification as an artifact was debatable. For
example, educational conferences and courses received eight nominations from physician
respondents and were retained for analysis as artifacts, despite their somewhat ambiguous
ontology—that is, they could be considered human or artifactual sources of knowledge,
depending on one’s perspective.
Network Analysis
In the following paragraphs, structural descriptive statistics and ERGM results for
the one-mode and two-mode networks are first reported separately, and then the
descriptives and results for the combined multilevel model are reported. For each of the
three networks, I first provide the structural descriptive statistics and then describe the
ERGM results.
Unipartite physician-physician network
Descriptive statistics. Descriptive statistics and a visualization of the unipartite
physician-physician network are presented in Table 5 (page 208) and Figure 3 (page
209), respectively. Forty-three percent of survey respondents reported seeking or
receiving best-practice knowledge from 65 different colleagues (35% of the total
population of primary care physicians). In-degree scores, indicating the popularity of
physician knowledge sources, ranged from 0 to 6. Out-degree scores, indicating the
activity of physician knowledge seekers or receivers, ranged from 0 to 9. Both degree
distributions were relatively skewed.
208
Table 5. Descriptive Statistics for Network Structure
Network A Network X Network M
Nodes 143199 199
Ties 102254 356
Knowledge sources nominated 65 (35%)
1
56 121
Physicians who nominated sources 56 (43%)
2
111 (85%)
2
119 (91%)
2
Isolates 44 (31%) 32 (16%) 9 (5%)
Components (excl. isolates)
3
18 6 3
Density 0.0050.032 0.013
Degree statistics In-degree Degree, physicians Out-degree
4
Range 67 11
M 0.713 1.776 2.490
SD 1.025 1.494 2.123
Skewness 2.0690.874 1.377
Out-degreeDegree, artifacts
Range 940
M 0.713 4.536
SD 1.271 7.405
Skewness 3.2242.940
In/out degree correlation
5
− 0.42
1
Denominator for percentage calculation is 185, the total population of PCPs.
2
Denominator for percentage calculation is 131, the total number of survey respondents.
3
“Component” refers to a disconnected group of nodes that each have ties within the group, but
that have no ties outside of the group.
4
Refers both to physicians’ out-degree ties to physician sources and to their undirected ties to
artifact sources.
5
The correlation was calculated using Spearman’s rho because the data were not normally
distributed; it was not significant.
209
Figure 3. Visualization of the unipartite physician-physician network.
210
The unipartite network was sparse, with a density of 0.005. Thirty-one percent of
physicians in the total population were isolates. The most striking descriptive result for
the unipartite network was that there were no interclinic ties between physicians, with 18
separate components in the network, excluding isolates. The absence of interclinic ties
was particularly surprising in the subset of physicians that served as primary care
department chiefs for each clinic site, because these chiefs attended monthly meetings at
Alpha Medical Group headquarters and thus interacted regularly. Also notable was that
there was substantial variation in intraclinic structure. In some clinics, for example,
highly active and/or highly popular physicians drove the network structure, creating star-
like configurations; others clinics featured triads or nodes serving in brokerage roles. In
two of the clinics, none of the physicians reported seeking or receiving best-practice
knowledge from their colleagues.
Because of the absence of interclinic ties in the unipartite network, I created a
structural zero file for those ties and included it in the model. A structural zero file is used
in network modeling when a researcher wishes to fix part of a network as exogenous for
empirical or theoretical reasons. For the structural zero file, I created an adjacency matrix
where for the intraclinic ties, “1” indicated that the tie was not fixed, and for the
interclinic ties, “0” indicated that the tie was fixed.
Additionally, after conducting an analysis of the in- and out-degree distributions, I
fixed the ties of two physicians with outlier degree statistics. The first physician had the
highest out-degree count, 9, in the network, and the second physician had the highest in-
degree count, 6. For the first physician I fixed the intraclinic out-degree ties only, and for
the second physician I fixed the intraclinic in-degree ties only. These two physicians
211
represented the only two nodes in the network with a degree statistic that was more than
five standard deviations from the mean. I fixed the in- and out-degree ties of these nodes
only after running numerous iterations of the network structural models with these ties
not fixed—including (as recommended by Robins & Lusher, 2013; Wang, Robins,
Pattison, & Lazega, 2013; Wang, et al., 2009) models with both Markov star parameters
and social circuit alternating star parameters, alternating star parameters with lambda
values greater than 2.0,
15
and alternating two-path parameters—and obtaining non-
convergent models or models with very poor goodness of fit statistics. Robins and Lusher
(2013) explain that when a network’s degree distributions are skewed, fixing the ties of
nodes with outlier degree statistics may be necessary in order to obtain convergent, good-
fitting models:
Degree distributions can often be hard to model. . . . If there are a few very high-
degree nodes, it may be necessary to treat them as exogenous—that is, the ties to
those nodes are treated as predictors of other network ties. This approach then
models the rest of the network conditional on these high-degree hubs. The
underlying justification is that the hubs are so different from the rest of the
network that they can be treated as special, influencing other network ties but not
themselves much affected by the rest of the structure. (p. 184)
15
For alternating social circuit model parameters in ERGM, such as alternating star parameters, the lambda
value determines the amount of centralization based on high-degree nodes. In PNet, the default lambda
value is 2.0. Higher lambda values encourage higher-degree nodes and may be more appropriate for
modeling networks with relatively skewed degree distributions. PNet allows researchers to adjust the
lambda value to accommodate the modeling of such networks. For more detailed explanations of the
lambda value, see Koskinen and Daraganova (2013) and Snijders and colleagues (2006).
212
Table 6. ERGM Results for Unipartite Physician-Physician Network
Version 1 Version 2
H Parameter Estimate SE Estimate SE
ArcA -5.7044 (1.809)* -12.2933 (2.171)*
ReciprocityA 2.3702 (0.614)* 3.2374 (0.669)*
TwoPathA -0.2327 (0.094)* -0.1962 (0.116)
SourceA 1.0861 (0.431)*
SinkA 2.2275 (0.416)*
AoutSA 1.3909 (0.259)* 0.5475 (0.259)*
Chief_SenderA 1.0127 (0.323)* 1.1239 (0.348)*
Chief_ReceiverA 1.9997 (0.324)* 1.6954 (0.384)*
Gender_SenderA -0.3018 (0.295) -0.3710 (0.317)
Gender_ReceiverA -1.4042 (0.423)* -1.1943 (0.510)*
Gender_InteractionA 0.9856 (0.511) 1.2710 (0.522)*
Clinical Sessions_SenderA -0.0292 (0.059) -0.0027 (0.064)
Clinical Sessions_ReceiverA -0.0720 (0.067) -0.0943 (0.084)
Clinical Sessions_DifferenceA 0.0381 (0.075) 0.0241 (0.083)
Organizational Tenure_SenderA 0.0112 (0.015) 0.0193 (0.015)
Organizational Tenure_ReceiverA 0.0134 (0.015) 0.0115 (0.019)
Organizational Tenure_DifferenceA 0.0011 (0.014) -0.0083 (0.016)
Professional Tenure_SenderA -0.0173 (0.014) -0.0191 (0.014)
Professional Tenure_ReceiverA -0.0107 (0.015) -0.0125 (0.019)
Professional Tenure_DifferenceA -0.0277 (0.016) -0.0174 (0.017)
1a Load_SenderA 0.2034 (0.135) 0.1835 (0.174)
6a ACC Legitimacy_SenderA 0.9878 (0.570) 1.1758 (0.688)
ACC Legitimacy_DifferenceA 0.5063 (0.205)*
6a AHA Legitimacy_SenderA -0.5786 (0.591) -0.6687 (0.708)
7a Physician Credibility_ReceiverA 1.2318 (0.161) *
8a ACC Credibility_SenderA 0.0443 (0.487) -0.2290 (0.540)
8a AHA Credibility_SenderA -0.0184 (0.501) 0.1340 (0.552)
Note. Version 1 omits the physician credibility parameter; Version 2 includes it.
* Indicates that the effect was significant (i.e., that the parameter estimate is greater than two
times the standard error in absolute value).
213
As discussed in Chapter 5, for the unipartite network, as well as the multilevel
network, two versions of the network were modeled: Version 1 excluded the physician
credibility variable, and Version 2 included it. The results of the best-fitting models for
both versions of the unipartite network are presented in Table 6 (page 212). For each of
the parameters, estimates and standard errors are provided. Recall from Chapter 5 that a
significant and positive (or negative) parameter estimate indicates that the parameter
occurs significantly more (or less) than would be expected by chance, given the other
parameters included in the model. The remainder of this section summarizes the
unipartite network modeling results, reviewing first the parameters for structure, then for
the control variables, and then for the research hypotheses.
ERGM results for structural parameters. In Version 1 of the unipartite network
model, there were five significant structural parameters: ArcA, ReciprocityA, TwoPathA,
SourceA, and AoutSA. The Arc parameter in MPNet is roughly analogous to the intercept
in a linear regression model, and indicates the baseline propensity for tie formation
(Lusher & Robins, 2013a). The Arc parameter typically becomes more negative as
parameters with positive values are added to a model. In this model, ArcA was negative
and significant, indicating there was a tendency not to form ties (i.e., there was tendency
for physicians not to seek or receive best practice knowledge from their colleagues),
unless other contervailing effects were present (e.g., except under conditions of
reciprocity). The ReciprocityA parameter was positive and significant, indicating a
tendency for the knowledge communication ties between physicians to be reciprocated.
This result is consistent with the tendency for reciprocal ties that has been observed in
numerous other knowledge networks reported in the research literature. The TwoPathA
214
parameter was negative and significant, indicating that a correlation between physician
popularity and physician activity was less likely to occur in the network. TwoPath
parameters are typically negative in unipartite directed networks (Robins & Lusher,
2013). The SourceA parameter was positive and significant, indicating a tendency in this
network for physicians to be seekers or receivers of best-practice knowledge (i.e., to have
positive out-degree scores), but not to be sources (i.e., to have an in-degree score of 0).
Lastly, the AoutSA parameter, or alternating out-star parameter, was positive and
significant, indicating network centralization based on highly active physicians with high
out-degree scores. This result echoed the descriptive results, supporting the picture the
descriptives painted of the out-degree distribution, driven by physician knowledge-
seeking and -receiving activity, being a dominant structural determinant in the unipartite
network. Although there were popular physician knowledge sources with in-degree
scores that were high relative to those of their colleagues, in-degree was not a significant
feature of the network structure. This is not a surprising result given that there were no
interclinic ties, limiting the scope of any particular physician’s popularity to that of his or
her own clinic. Triads were also not a significant driver of the network structure.
In Version 2 of the unipartite network model, which included the physician
credibility parameter, there were four significant structural parameters: ArcA,
ReciprocityA, AoutSA, and SinkA. As in Version 1 of the model, ArcA was negative and
significant, ReciprocityA was positive and significant, and AoutSA was also positive and
significant. As was true in Version 1, these three effects indicated a tendency not to form
ties unless other countervailing effects were present; a tendency for knowledge ties
between physicians to be reciprocated; and the presence of network centralization based
215
on highly active physicians with high out-degree scores. Unlike Version 1 of the model,
however, which included a positive and significant SourceA parameter, in Version 2
SourceA was not significant. In its place was a positive and significant SinkA effect.
Whereas the Source parameter indicates the tendency for nodes to have positive out-
degree scores and an in-degree score of 0, the Sink parameter suggests the opposite,
indicating the tendency for nodes to have positive in-degree scores (i.e., to be popular
sources) and an out-degree score of 0 (i.e., to have no outgoing knowledge-seeking or -
receiving ties). The presence of SourceA in Version 1 of the model and SinkA in Version
2 should not necessarily be interpreted as contradictory results. Rather, the fact that both
were positive and significant effects, together with the positive and significant AoutSA
parameters in both models, highlights that in two very similar versions of the unipartite
network, activity as a knowledge seeker or receiver and popularity as a knowledge source
were not correlated. That is, physicians tended to be either active knowledge seekers or
popular sources, but not to “specialize” in both roles. Lastly, in addition to the four
significant structural parameters in Version 2 of the model, a TwoPathA effect was also
included in the model although it was not significant. Whereas TwoPathA was a
significant effect in Version 1, here it was included because it proved critical for model
convergence and for keeping the SACFs of the other model parameters at acceptably low
values.
ERGM results for demographic characteristics. After identifying the structural
effects that best represented the empirical knowledge network, I added the parameters for
the physician demographic characteristics proposed as control variables in the model.
Five control variables were considered in the network: department chief status, gender,
216
number of clinical sessions a physician worked, organizational tenure, and professional
tenure. In Version 1 of the unipartite model, three parameters for these demographic
characteristics were significant: Chief_SenderA, Chief_ReceiverA, and GenderReceiverA.
The Chief_SenderA and Chief_ReceiverA parameters were both positive and significant,
indicating that physicians who were department chiefs were both more likely to seek or
receive best-practice knowledge from colleagues, and to act as sources of knowledge for
their colleagues. These results echo numerous reports in the knowledge network literature
that formal hierarchy and roles in organizations influence employees’ informal
knowledge communication behaviors. The GenderReceiverA parameter was also
significant, but negative, indicating that physician knowledge sources were more likely to
be men. (To create the binary gender variable, I coded men as 0 and women as 1). This
result was somewhat surprising, given that 55% of the primary care physicians at Alpha
Medical Group were women, and department chief status and male gender were not
significantly correlated. The other three demographic characteristics considered in the
network models—number of clinical sessions, organizational tenure, and professional
tenure—exerted no significant effects on Version 1 of the unipartite network.
In Version 2 of the unipartite network model, like Version 1, the Chief_SenderA
and Chief_ReceiverA parameters were positive and significant, and the GenderReceiverA
parameter was negative and significant. In Version 2 of the model, a fourth parameter for
the demographic characteristics was also significant. Gender_InteractionA, a homophily
effect, was positive and significant, indicating that female physicians were more likely to
seek or receive best-practice knowledge from other female colleagues. Taken together,
the two significant gender effects in Version 2 of the model suggested that men tended to
217
be more popular knowledge sources, but that when women sought or received best-
practice knowledge, they tended to prefer their female colleagues. As in Version 1 of the
model, none of the other demographic characteristics—the clinical sessions and tenure
variables—exerted a significant effect on Version 2.
ERGM results for research hypotheses. Following estimation of the structural
effects and physician demographic variables in the unipartite model, I tested the
parameters representing the research hypotheses. The first set of hypotheses focused on
the sociomaterial agency of communication load, and one of these hypotheses applied to
the unipartite network. Hypothesis 1a proposed that physicians with higher
communication load would be less likely to seek or receive best-practice knowledge from
their colleagues. It was tested with the Load_SenderA parameter, and was not significant
in both Versions 1 and 2 of the model.
The second set of hypotheses focused on the sociomaterial agencies of legitimacy
and credibility, and three of these hypotheses applied to the unipartite network.
Hypothesis 6a and Hypothesis 8a proposed that physicians who perceived the legitimacy
and credibility, respectively, of the best-practice authors (the ACC and the AHA) to be
greater would be more likely to seek or receive knowledge from their colleagues. These
hypotheses were tested with four parameters: ACC Legitimacy_SenderA, AHA
Legitimacy_SenderA, ACC Credibility_SenderA, and AHA Credibility_SenderA. Of these
four parameters, none were significant in either Version 1 or Version 2 of the model.
However, goodness-of-fit testing suggested inclusion of a fifth parameter regarding the
legitimacy of the ACC, ACC Legitimacy_DifferenceA, and this parameter was positive
and significant in Version 2 of the model. The Difference parameter in MPNet tests for a
218
homophily effect, and a positive and significant result indicates heterophily (that is, a
large difference in the attribute scores between the two nodes). Therefore, the positive
and significant ACC Legitimacy_DifferenceA parameter in Version 2 of the unipartite
model indicated the tendency for physician knowledge seekers or receivers to have a
different perception of ACC legitimacy than physician knowledge sources. This result did
not indicate, however, whether it was the seekers or the sources who had the higher
opinion of ACC legitimacy, so in isolation, without other significant effects for the
legitimacy of the best-practice authors, it was difficult to interpret.
Lastly, Hypothesis 7a proposed that physicians would be more likely to seek or
receive knowledge from colleagues with greater credibility. This hypothesis was tested
with the Physician Credibility_ReceiverA parameter, and was applied only to Version 2
of the unipartite model. The parameter was positive and significant, suggesting that
physicians were more likely to obtain best-practice knowledge from colleagues whom
were collectively regarded as more credible sources. Because of the non-optimal
measurement of the physician credibility variable, however, this result should be
interpreted with caution and viewed as offering only qualified support for the hypothesis.
Appendix F presents the results for the goodness of fit tests for all versions of the
unipartite, bipartite, and combined multilevel models, focusing for reasons of space and
clarity on the reporting of poorly-fitting parameters only. In goodness of fit evaluation in
ERGM, poorly-fitting parameters not explicitly specified in the model are defined as
those with a goodness of fit t-ratio of more than 2.0. In Versions 1 and 2 of the unipartite
models, there were no such poorly-fitting parameters.
219
Figure 4. Visualization of the bipartite physician-artifact network.
220
Bipartite physician-artifact network
Descriptive statistics. Descriptive statistics and a visualization of the bipartite
physician-artifact network are presented in Table 5 (page 208) and Figure 4 (page 219),
respectively. Eighty-five percent of survey respondents reported seeking or receiving
best-practice knowledge from artifacts, a much higher proportion as compared to those
who reported their colleagues as knowledge sources. Degree scores for physicians,
indicating the activity of physician knowledge seekers or receivers, ranged from 0 to 7.
Degree scores for artifacts, indicating the popularity of artifact knowledge sources,
ranged from 1 to 41, with a mean degree score for artifacts of approximately 4.5. The
artifact degree distribution, like the degree distributions for the unipartite network, was
relatively skewed, with eight artifacts—represented by the small number of red circles at
the center of large star-like configurations in Figure 3—receiving 10 or more
nominations. Consistent with the degree statistics, the bipartite network was more
densely populated than the unipartite network, with 16% of physician nodes as isolates.
ERGM results for structural parameters. The results of the best-fitting model for
the bipartite network are presented in Table 7 (page 221). Note that because the physician
credibility hypothesis was not applicable to the bipartite network, only one version of the
network was modeled. There were three significant structural parameters in the bipartite
model: XEdge, XStar2B, and XASB. The XEdge parameter was negative and significant,
indicating, like the negative ArcA parameter in the unipartite model, a baseline propensity
against tie formation in the network. The XStar2B parameter was positive and significant,
indicating the presence of some very popular artifacts in the network and echoing the
descriptive results. The XASB parameter, or alternating star parameter for the artifact
221
nodes, was also positive and significant, further indicating that the network was
centralized around a few highly popular artifact nodes. The tendency for closure was not
a significant structural feature in this bipartite network.
Table 7. ERGM Results for Bipartite Physician-Artifact Network
H Parameter Estimate SE
XEdge -6.0631 (1.015)*
XStar2B 0.0287 (0.011)*
XASB ( λ = 10.0)
1
0.3011 (0.030)*
Chief_XEdgeA 0.3389 (0.206)
Gender_XEdgeA 0.1511 (0.157)
Clinic_X2StarAMatch -0.1318 (0.113)
Clinical Sessions_XEdgeA -0.0026 (0.038)
Organizational Tenure_XEdgeA 0.0073 (0.008)
Professional Tenure_XEdgeA -0.0007 (0.009)
1b Load_XEdgeA 0.1317 (0.106)
4a Mass Media_XEdgeB -0.1698 (0.173)
4b Load-Mass Media_XEdgeAB 0.3904 (0.302)
5 Artifact Legitimacy_XEdgeB -0.0002 (0.030)
6b ACC Legitimacy_XEdgeA 0.5517 (0.339)
6b AHA Legitimacy_XEdgeA -0.5116 (0.369)
7b Artifact Credibility_XEdgeB 0.0309 (0.035)
8b ACC Credibility_XEdgeA 0.0882 (0.330)
8b AHA Credibility_XEdgeA -0.0654 (0.332)
Note. Because the physician credibility parameter is not applicable to the bipartite network, only
one version of the model was estimated.
* Indicates that the effect was significant (i.e., that the parameter estimate is greater than two
times the standard error in absolute value).
1
For alternating social circuit model parameters in ERGM, such as alternating star parameters, a
lambda value is specified that determines the amount of centralization based on high-degree
nodes. In PNet, the default lambda value is 2.0. Higher lambda values encourage higher-degree
nodes and are used for modeling networks with relatively skewed degree distributions.
222
ERGM results for demographic characteristics. The demographic characteristics
exerted no significant effects in the bipartite network.
ERGM results for research hypotheses. None of the research hypotheses proved
significant in this network, either. For the first set of hypotheses focused on
communication load, three applied to the bipartite network. Hypothesis 1b proposed that
physicians with higher communication load would be less likely to seek or receive best-
practice knowledge from artifacts. This hypothesis was represented in the model by the
Load_XEdgeA parameter, and was not significant. Hypothesis 4a proposed that
physicians would be more likely to seek or receive knowledge from mass media, rather
than niche media artifacts, based on the rationale that knowledge communication with
mass media artifacts would contribute less to communication load than that with niche
artifacts. This hypothesis was represented in the model by the Mass Media_XEdgeB
parameter, and was not significant. Hypothesis 4b proposed, similarly, that physicians
with higher communication load, specifically, would be even more likely to seek or
receive knowledge from mass media as opposed to niche media artifacts. This hypothesis
was represented by the Load-Mass Media_XEdgeAB parameter, and was not significant.
For the second set of hypotheses, regarding source legitimacy and credibility, four
applied to the bipartite network. Hypothesis 5 and Hypothesis 7b proposed that
physicians would be more likely to seek or receive best-practice knowledge from artifacts
with greater legitimacy and greater credibility, respectively. These hypotheses were
tested in the model with the Artifact Legitimacy_XEdgeB and Artifact
Credibility_XEdgeB parameters, respectively, and were not significant. Hypothesis 6b
and Hypothesis 8b proposed that physicians who perceive the legitimacy and credibility,
223
respectively, of the best-practice authors to be greater would be more likely to seek or
receive knowledge from artifacts. These hypotheses were tested with four parameters:
ACC Legitimacy_XEdgeA, AHA Legitimacy_XEdgeA, ACC Credibility_XEdgeA, and
AHA Credibility_XEdgeA. None of these four parameters were significant.
Therefore, in summary, none of physician demographic characteristics or research
hypotheses regarding communication load, legitimacy, or credibility exerted a significant
influence on the bipartite network. Table 1 in Appendix F presents the results for the
goodness of fit tests for the bipartite model. Although globally, the model offered a good
fit for the observed network, two parameters not explicitly estimated in the model had a
poor fit: IsolatesXB (t-ratio = -3.170) and SD DegreeXA (t-ratio = 2.011). The IsolatesXB
parameter refers to the occurrence of isolates among the “B” nodes, which in this case
were the artifacts. The poor fit of this parameter likely reflects that it was difficult for the
ERGM to accurately represent the fact that there were no artifact isolates in the observed
network, but that physician isolates were present. SD DegreeXA is not a parameter, but
instead indicates the standard deviation of the degree statistics for the “A” nodes in the
bipartite model, which in this case were the physicians. The very slightly elevated t-ratio
for this goodness of fit statistic suggested that the ERGM was not able to perfectly
represent the physicians’ degree distribution in the observed network. Robins and Lusher
(2013) note that for goodness of fit standard deviation and skewness statistics, t-ratios
above 2.0 are not unusual, and suggest that there are some highly active or highly popular
actors in the observed network that are not completely represented in the model.
224
Figure 5. Visualization of the multilevel network.
225
Multilevel network
Descriptive statistics. Descriptive statistics and a visualization of the combined
multilevel network are presented in Table 5 (page 208) and Figure 5 (page 224),
respectively. Overall, 91% of survey respondents reported seeking or receiving best-
practice knowledge from a total of 121 different colleagues and artifacts. Only 5% of
physicians were isolates in the combined network. Degree scores for physicians,
indicating their total knowledge-seeking and -receiving activity, ranged from 0 to 11,
with a mean score of about 2.5.
ERGM results for structural parameters. The results of the best-fitting models
for Version 1 and Version 2 of the multilevel network are presented in Table 8 (page
226). Note that unlike for the unipartite and bipartite network models, only significant
parameters are included in the presentation of results for the multilevel models. A number
of parameters for the control variables and research hypotheses were not significant in the
multilevel models and were not estimated in the final models presented in Table 8.
Because of the computing power and time required to estimate models for the multilevel
network in MPNet, it was not possible to retain non-significant effects in the estimation
runs.
In Version 1 of the multilevel model, the significant unipartite and bipartite
structural effects were identical in sign (positive or negative) and significance to the
unipartite and bipartite networks estimated alone. For the combined multilevel network,
there was one positive and significant structural parameter, Out2StarAX. This effect
indicated a tendency for physicians who were active in the unipartite network to also be
active in the bipartite network, and vice versa. That is, physicians who sought or received
226
best-practice knowledge from their colleagues also tended to seek or receive knowledge
from artifacts.
Table 8. ERGM Results for Multilevel Network
Version 1 Version 2
H Parameter Estimate SE Estimate SE
Network A
ArcA -7.2791 (1.000)* -14.1716 (1.583)*
ReciprocityA 2.5228 (0.627)* 3.5373 (0.684)*
TwoPathA -0.2520 (0.094)* -0.2356 (0.108)*
SourceA 1.0367 (0.391)*
SinkA 2.2492 (0.441)*
AoutSA 1.3801 (0.263)* 0.5649 (0.247)*
Chief_SenderA 0.9593 (0.305)* 1.1083 (0.319)*
Chief_ReceiverA 1.9362 (0.302)* 1.6516 (0.391)*
Gender_ReceiverA -1.3965 (0.368)* -1.0522 (0.423)*
Gender_InteractionA 0.8524 (0.377)* 1.0462 (0.425)*
6a ACC Legitimacy_SenderA 0.4988 (0.152) * 0.5178 (0.165) *
ACC Legitimacy_DifferenceA 0.4756 (0.184)*
7a Physician Credibility_ReceiverA 1.2515 (0.165) *
Network X
XEdge -4.9422 (0.120)* -4.9476 (0.121)*
XStar2B 0.0206 (0.008)* 0.0207 (0.007)*
XASB ( λ = 10.0)
1
0.3017(0.031)*0.3015 (0.029)*
Network M
Out2StarAX 0.1028 (0.043)* 0.1099 (0.046)*
Note. Version 1 omits the physician credibility parameter; Version 2 includes it. For the
combined multilevel model, only significant parameters are presented.
* Indicates that the effect was significant (i.e., that the parameter estimate is greater than two
times the standard error in absolute value).
1
For alternating social circuit model parameters in ERGM, such as alternating star parameters, a
lambda value is specified that determines the amount of centralization based on high-degree
nodes. In PNet, the default lambda value is 2.0. Higher lambda values encourage higher-degree
nodes and are used for modeling networks with relatively skewed degree distributions.
227
Three multilevel structural effects were proposed in Hypotheses 2a, 2b, and 3, but
were not significant in the model estimation. Hypothesis 2a stated that physicians with
ties to one or more artifacts (i.e., those who acted as potential brokers to one or more
artifacts) would be more likely to be sources of knowledge for their colleagues.
Hypothesis 2b stated that these same physician-brokers would also be more popular
sources of knowledge among their colleagues. These hypotheses were represented in the
model with the In2StarAX and AAinS1X parameters. The lack of significance of these two
parameters indicated that a brokerage effect did not occur for physicians who could
connect to and interpret artifact knowledge sources for their potentially overloaded
colleagues. Hypothesis 3 stated that physicians who seek or receive knowledge from one
or more artifacts would tend to have ties to colleagues who were also connected to those
same artifacts. It was represented in the model with the TXAXarc parameter and its
alternating, social circuit analog, the ATXAXarc parameter. The lack of significance of
these two parameters indicated that a closure effect did not occur whereby a focal
physician exhibited a preference for artifacts that a colleague knowledge source also
used. Finally, the Version 1 structural results indicated no tendencies in the multilevel
network for other significant star-based or triad-based structural patterns.
In Version 2 of the multilevel model, the significant structural effects were
identical in sign and significance for the unipartite and bipartite networks, with one subtle
exception. In the unipartite model, TwoPathA was not significant but necessary for good
model fit, whereas in the multilevel model, TwoPathA was negative and significant (as
was also the case in the unipartite and multilevel models of Version 1). As in Version 1,
there was one positive and significant structural parameter, Out2StarAX, indicating a
228
tendency for physicians who were active in the unipartite network to also be active in the
bipartite network, and vice versa. Also as in Version 1, the brokerage and closure
hypotheses (H2a, H2b, and H3) in Version 2 were not significant structural effects.
ERGM results for demographic characteristics. The results for the tests of the
demographic control variables in the combined multilevel network largely echoed those
in the unipartite network. In both Versions 1 and 2 of the multilevel model, the
Chief_SenderA and Chief_ReceiverA parameters were positive and significant, and the
GenderReceiverA parameter was negative and significant. As in Version 2 of the
unipartite model, the Gender_InteractionA parameter was also positive and significant in
both versions of the multilevel model. To summarize, then, the demographic results for
the multilevel model confirmed the significant roles that department chief status and
gender preferences played in structuring knowledge communication among the primary
care physicians.
ERGM results for research hypotheses. Following estimation of the structural
effects and physician demographic variables in the combined multilevel model, I tested
the parameters representing the research hypotheses. For the first set of hypotheses
focusing on communication load, there were four hypotheses applicable to the multilevel
network. Hypothesis 1c proposed that physicians with higher communication load would
be less likely to seek or receive best-practice knowledge from both their colleagues and
artifacts. It was tested with the Load_Star2AXSender parameter, and was not significant
in both Versions 1 and 2 of the model. Hypotheses 2a, 2b, and 3 were represented by
purely structural parameters, and as discussed above, were also not significant effects in
Version 1 or Version 2 of the multilevel model. Their lack of significance meant that
229
there was no evidence that physicians, motivated by concerns about communication
overload, exhibited a preference for ties to human knowledge sources by choosing to
access artifacts (a) indirectly when brokerage via a colleague was possible, or (b) directly
when closure via a colleague also connected to the same artifact was possible.
For the second set of hypotheses focusing on legitimacy and credibility, there
were two hypotheses applicable to the combined multilevel network. Hypothesis 6c and
Hypothesis 8c proposed that physicians who perceived the legitimacy and credibility,
respectively, of the best-practice authors to be greater would be more likely to seek or
receive knowledge from both colleagues and artifacts. These hypotheses were tested with
four parameters: ACC Legitimacy_Star2AXSender, AHA Legitimacy_Star2AXSender,
ACC Credibility_Star2AXSender, and AHA Credibility_Star2AXSender. None of these
four parameters were significant in Version 1 or Version 2 of the multilevel model.
However, one of the unipartite-level legitimacy parameters that was non-
significant in Version 1 and Version 2 of the unipartite model became significant in both
versions of the combined multilevel model. The ACC Legitimacy_SenderA parameter
(H6a) was positive and significant, indicating that in the combined multilevel network,
taking into account effects driven by both human and artifact knowledge sources,
physicians who perceived the legitimacy of the ACC to be greater were more likely to
seek or receive best-practice knowledge from their colleagues. Additionally, the two
other legitimacy and credibility parameters that were significant in Version 2 of the
unipartite model remained significant in Version 2 of the multilevel model. These
parameters, Physician Credibility_ReceiverA (H7a) and ACC Legitimacy_DifferenceA,
were both positive and significant, confirming in the multilevel network the results
230
observed in Version 2 of the unipartite network: that physicians were more likely to seek
or receive best-practice knowledge from colleagues whom were collectively regarded as
more credible sources; and that there was a tendency for physician knowledge seekers or
receivers to have a different perception of ACC legitimacy than the physician knowledge
sources to whom they connected.
Table 2 in Appendix F presents the results for the goodness of fit tests for the
combined multilevel model. Although globally the model offered a good fit for the
observed network, several parameters not explicitly estimated in the model had a poor fit
in Version 1 and in Version 2. All of the poorly-fitting parameters across both versions of
the model were those applicable to the bipartite portion of the multilevel model. In
Version 1, IsolatesXA (t-ratio = 2.428), Mass Media_X4CycleB2 (t-ratio = 2.600), and SD
DegreeXA (t-ratio = 2.015) exhibited poor fit. In Version 2, IsolatesXA (t-ratio = 2.232),
IsolatesXB (t-ratio = -4.343), Mass Media_XEdgeB (t-ratio = -2.212), and SD DegreeXA
(t-ratio = 2.062) exhibited poor fit. The IsolatesXA parameter refers to the occurrence of
isolates among the “A” nodes in the model, which in this case were the physicians, and
the IsolatesXB parameter refers to the occurrence of isolates among the “B” or artifact
nodes in the model. As in the case of the poorly-fitting IsolatesXB parameter in the
goodness of fit results for the bipartite model, the poor fit of IsolatesXA and IsolatesXB in
the multilevel model likely reflects that it was difficult for the ERGM to accurately
represent the fact that there were no artifact isolates in the observed network, but that
physician isolates were present. Networks in which isolates are present or “possible” in
one mode, but not in the other, are difficult to model (Niekamp, Mercken, Hoebe, &
Dukers-Muijrers, 2013; Wang, 2013). Based on the goodness of fit results for the two
231
mass media parameters, I tested them in both versions of the multilevel model, but they
were not significant and so were omitted from the final versions of the models. Again, as
in the case of the bipartite model, the poor fit of SD DegreeXA—indicating the standard
deviation of the physicians’ degree statistics in the bipartite portion of the model—
suggested that the ERGM was not able to perfectly represent the physicians’ degree
distributions in the observed networks.
Research Question 1 and Hypothesis 9
In addition to Hypotheses 1 through 8 regarding the specific effects that
communication load, legitimacy, and credibility may exert on the structure of the
knowledge network, I proposed a final research question and hypothesis that were
intended to shed light on these proposed forms of sociomaterial agency in a more general
way. Considering the two-part research question first, Research Question 1a asked for
artifact knowledge sources, there was an empirical difference between the concepts of if,
legitimacy and credibility, as manifested in the influences of artifact source legitimacy
(H5) and artifact source credibility (H7b) on the presence of knowledge ties with
physicians. In the bipartite model, neither artifact legitimacy nor artifact credibility
exerted a significant effect on the physician-artifact network. Therefore, the ERGM
analysis offered no evidence of an empirical difference between the legitimacy and
credibility concepts.
Research Question 1b similarly asked if, for authors of best practices, there was
an empirical difference between the two concepts, as manifested in the influences of
author legitimacy (H6a, b, and c) and author credibility (H8a, b, and c) on the presence of
knowledge ties with physicians. In the unipartite and bipartite models, neither author
232
legitimacy nor author credibility exerted significant effects. In the combined multilevel
model, however, ACC legitimacy was positively associated with physicians seeking or
receiving knowledge from their colleagues, whereas ACC credibility was not. This result
offered some empirical support for the theoretical distinctions between legitimacy and
credibility suggested in the research literature.
Lastly, Hypothesis 9 stated that the multilevel, sociomaterial model would explain
more variance in knowledge network structure than the unipartite model alone or the
bipartite model alone. Following Wang and colleagues (2013; 2013), I evaluated this
hypothesis in two steps. First, I compared differences in the significance of the individual
parameter estimates across the three models. Table 9 (page 235) compares these
differences across Version 1 of the models, and Table 10 (page 236) compares them
across Version 2 of the models. Comparing parameters across Version 1 of the models,
there were two unipartite parameters that were not significant in the unipartite model—
Gender_InteractionA and ACC Legitimacy_SenderA—that became significant in the
multilevel model. Similarly, comparing effects across Version 2 of the models, there was
one unipartite parameter that was not significant in the unipartite model—again, ACC
Legitimacy_SenderA—that became significant in the multilevel model. In Wang and
colleagues’ (2013; 2013) MPNet analyses, they observed that the inclusion of multilevel
parameters in the overall multilevel model greatly simplified the previous specifications
of the two unipartite models they estimated separately. They interpreted this observation
as evidence that their combined multilevel model explained more variance than their two
unipartite models estimated separately explained. Their comparison of the parameter
estimates across models yielded a slightly different conclusion than mine—they observed
233
that the multilevel parameters in the multilevel model allowed them to remove a number
of unipartite parameters that became non-significant, thereby simplifying the final overall
model, whereas I observed that parameters that were non-significant in the unipartite
model became significant in the multilevel model. However, Wang et al.’s observation
and my observation are similar, in that in both studies, the multilevel model explained
more variance in network structure than the unipartite or bipartite models explained
alone. Therefore, the first step in the evaluation of Hypothesis 9 offered support for the
hypothesis.
For the second step in the evaluation of Hypothesis 9, I compared the goodness of
fit statistics across the three models. For both versions of the unipartite model, there were
no poorly-fitting parameters in the goodness of fit results. The Mahalanobis distance
statistic, which is a global goodness of fit heuristic, was 353 in Version 1, and 1,113 in
Version 2. For the bipartite model, there were two poorly-fitting parameters, and the
Mahalanobis distance statistic was 833. For Version 1 of the multilevel model, there were
three poorly-fitting parameters and the Mahalanobis distance was −2,984,823, whereas
for Version 2, there were four poorly-fitting parameters and the Mahalanobis distance
was −526,663. In the case of Wang and colleagues’ (2013; 2013) research, they observed
that the goodness of fit statistics were best for the overall multilevel network model, as
compared to those for the two unipartite network models, leading the authors to conclude
that the multilevel model offered the best explanation for the structure of the network.
16
In the case of the present study, however, the goodness of fit statistics across the three
models were best for the unipartite models and worst for the multilevel models.
16
Note that Wang and colleagues compared only the individual goodness of fit statistics of each parameter
included in the three models, and did not compare the global Mahalanobis distance statistics of the three
models as I did.
234
Therefore, the second step in the evaluation of Hypothesis 9 did not support the
hypothesis. Taken together, the results for Hypothesis 9 were mixed, offering partial
support for the proposition that the multilevel, sociomaterial model would explain more
variance in knowledge network structure than the unipartite model alone or the bipartite
model alone.
235
Table 9. Comparison of Results for Version 1 Models Without Physician Credibility Parameter
Network A Network X Network M
H Parameter Estimate SE Estimate SE Estimate SE
ArcA -5.7044 1.809* -7.2791 (1.000)*
ReciprocityA 2.3702 0.614* 2.5228 (0.627)*
TwoPathA -0.2327 0.094* -0.2520 (0.094)*
SourceA 1.0861 0.431* 1.0367 (0.391)*
AoutSA 1.3909 0.259* 1.3801 (0.263)*
Chief_SenderA 1.0127 0.323* 0.9593 (0.305)*
Chief_ReceiverA 1.9997 0.324* 1.9362 (0.302)*
Gender_ReceiverA -1.4042 0.423* -1.3965 (0.368)*
Gender_InteractionA 0.8524 (0.377)*
6a ACC Legitimacy_SenderA 0.4988 (0.152) *
XEdge -6.0631 (1.015)* -4.9422 (0.120)*
XStar2B 0.0287 (0.011)* 0.0206 (0.008)*
XASB ( λ = 10.0)
1
0.3011 (0.030)* 0.3017(0.031)*
Out2StarAX 0.1028 (0.043)*
Note. Only significant parameters are presented.
* Indicates that the effect was significant (i.e., that the parameter estimate is greater than two times the standard error in absolute value).
1
For alternating social circuit model parameters in ERGM, such as alternating star parameters, a lambda value is specified that determines the
amount of centralization based on high-degree nodes. In PNet, the default lambda value is 2.0. Higher lambda values encourage higher-degree
nodes and are used for modeling networks with relatively skewed degree distributions.
236
Table 10. Comparison of Results for Version 2 Models With Physician Credibility Parameter
Network A Network X Network M
H Parameter Estimate SE Estimate SE Estimate SE
ArcA -12.2933 (2.171)* -14.1716 (1.583)*
ReciprocityA 3.2374 (0.669)* 3.5373 (0.684)*
TwoPathA -0.1962 (0.116) -0.2356 (0.108)*
SinkA 2.2275 (0.416)* 2.2492 (0.441)*
AoutSA 0.5475 (0.259)* 0.5649 (0.247)*
Chief_SenderA 1.1239 (0.348)* 1.1083 (0.319)*
Chief_ReceiverA 1.6954 (0.384)* 1.6516 (0.391)*
Gender_ReceiverA -1.1943 (0.510)* -1.0522 (0.423)*
Gender_InteractionA 1.2710 (0.522)* 1.0462 (0.425)*
6a ACC Legitimacy_SenderA 0.5178 (0.165) *
ACC Legitimacy_DifferenceA 0.5063 (0.205)* 0.4756 (0.184)*
7a Physician Credibility_ReceiverA 1.2318 (0.161)* 1.2515 (0.165)*
XEdge -6.0631 (1.015)* -4.9476 (0.121)*
XStar2B 0.0287 (0.011)* 0.0207 (0.007)*
XASB ( λ = 10.0)
1
0.3011 (0.030)* 0.3015(0.029)*
Out2StarAX 0.1099 (0.046)*
Note. Only significant parameters, or non-significant parameters that improve model fit, are presented.
* Indicates that the effect was significant (i.e., that the parameter estimate is greater than two times the standard error in absolute value).
1
For alternating social circuit model parameters in ERGM, such as alternating star parameters, a lambda value is specified that determines the
amount of centralization based on high-degree nodes. In PNet, the default lambda value is 2.0. Higher lambda values encourage higher-degree
nodes and are used for modeling networks with relatively skewed degree distributions.
237
Qualitative Results
Thirty-nine physicians (27% of survey respondents) provided responses to the
open-ended survey question eliciting qualitative data regarding how they learned about
best-practice knowledge. These responses can be grouped into three general categories:
those that discussed the physician’s knowledge source preferences and the structure of his
or her best-practice knowledge network, those that discussed the influence of
communication load, and those that discussed the influences of source legitimacy and
credibility. Selected responses that complemented, contradicted, or helped explain the
results derived from the quantitative analyses are presented and discussed in Chapter 7.
238
CHAPTER 7: DISCUSSION AND CONCLUSION
In this final chapter I begin by summarizing and interpreting the study results. I
then identify a number of limitations and contributions of the study, all of which point to
directions for future research, and offer some conclusions.
Interpretation of Results
The results described in Chapter 6 paint a picture of how best-practice
knowledge—in this case, new clinical guidelines for the treatment of high cholesterol—
was communicated in a sociomaterial organizational network of primary care physicians
and knowledge artifacts. Chapter 6 outlined the effects of three sets of variables on the
global structure of the sociomaterial knowledge network: local structural variables; actor
demographic characteristics; and actor communication load, legitimacy, and credibility.
In the following paragraphs, I offer a summary and interpretation of the effects that these
three sets of variables were observed to exert on global network structure.
Network structure
The significant structural parameters in the network models, together with the
descriptive statistics, suggested that knowledge network structure varied widely among
the 17 clinics in the study population, but overall was shaped by three dominant features:
dyadic reciprocity, the absence of interclinic ties, and centralization based on both active
physicians and very popular artifact knowledge sources. The significance of reciprocity
replicates numerous previous findings in the knowledge network literature (e.g.,
Agneessens & Wittek, 2012; Keating, et al., 2007; Lomi, et al., 2014; Rank, et al., 2010;
Su, et al., 2010; Zappa, 2011).
239
The lack of interclinic ties was surprising. On one hand, this result may be partly
due to the answer choice format I used for the name generator question for the physician-
physician network. I used a roster format that featured two drop-down menus with drill-
down display logic. These drop-down menus filtered physician names according to the
clinic to which they belonged (see Appendix A). To provide the name of the first
colleague to whom he or she went for best-practice knowledge, a respondent selected a
clinic from the first drop-down menu, and then selected the name of a colleague
belonging to that particular clinic from the second drop-down menu. In other words, the
choices in the second drop-down menu were limited to only physicians who worked in
the clinic selected in the first drop-down menu. The respondent was then asked to repeat
this process of first selecting the clinic, and then the physician within that clinic, for each
additional colleague. This format was used in order to reduce respondent recall bias and
cognitive load, as well as to minimize the tedium of scrolling through a long list of 185
physician names. It is possible, however, that the use of the first drop-down menu for the
clinic selection may have primed respondents to repeatedly and only select their home
clinic for each physician they nominated as a knowledge source.
On the other hand, the absence of interclinic ties in this study may not have been
influenced by survey design factors. This result echoes numerous previous reports in the
network literature suggesting that interdepartmental or interdivisional knowledge
communication within organizations can be infrequent and fraught with obstacles
(Hansen, Mors, & Løvås, 2005; Lazega & van Duijn, 1997; Lomi, et al., 2014; McEvily,
et al., 2014; Szulanski, 1996; Tortoriello, Reagans, & McEvily, 2012; Tsai, 2002).
240
The presence of network centralization driven particularly by very popular
artifacts highlights the importance of considering both humans and artifacts when
studying organizational knowledge networks. Indeed, among knowledge sources,
physicians utilized artifacts more frequently than colleagues, an observation that would
have been impossible had this study examined knowledge communication among people
only. Of the popular artifacts in the knowledge network, UpToDate was by far the most
influential in driving network structure. The popularity of UpToDate was likely due in
part to its unique materiality as an electronic database that was not simply a website
accessible online, but one that was accessible within, and integrated into, the electronic
medical record system that Alpha physicians used to record all patient interactions and to
care plans, such as ordering diagnostic tests, prescribing medications, and making
referrals to specialists. As Physician 134
17
reported, “[UpToDate] decision support has
made new guidelines much more understandable/‘do-able.’” UpToDate’s unique
accessibility gave it a powerful advantage over other human and nonhuman knowledge
sources and likely reduced the need for physicians to look to other sources.
Despite the dominance of UpToDate and other popular artifacts in the overall
network, however, the qualitative data suggested that many physicians valued and/or
desired more knowledge communication with their colleagues, as the following
comments illustrate:
“Talking with my colleagues is a good way for me to compare my
thoughts/reactions to guidelines/new data.” –Physician 27
“I find group discussion most helpful. We had a department meeting during which
we discussed the guidelines. It was not the topic of greatest focus, but we spent
17
Identities of physician respondents were anonymized with randomly-assigned identification numbers.
241
some time discussing the implications and our personal interpretations of the data.
We also discussed whether providers had already or were planning on changing
their practices to adhere to the guidelines….It was a positive exchange. –
Physician 122
“I really like to know how people around me are practicing, i.e., the [Alpha
Medical Group] “consensus.”—Physician 48
“Colleagues need to be more comfortable in sharing information with their peers.”
–Physician 69
Demographic characteristics
Inclusion of physicians’ demographic characteristics as controls in the network
analysis suggested that two such characteristics, supervisory position and gender, exerted
significant influence on the structure of best-practice knowledge communication.
Physicians who were department chiefs were more likely to act both as knowledge
seekers or receivers and as knowledge sources, replicating numerous reports in the
knowledge network literature that formal hierarchy and roles in organizations influence
employees’ informal knowledge communication behaviors (Agneessens & Wittek, 2012;
Lazega & van Duijn, 1997; McEvily, et al., 2014; Roberts & O'Reilly, 1979; Yousefi-
Nooraie, et al., 2014). Although 55% of the physicians in the study were women, men
tended to be more popular knowledge sources. When women sought or received
knowledge, however, they were more likely to prefer their female colleagues. The
popularity of male knowledge sources over female knowledge sources was surprising, but
previous research supports the gender homophily effect (Keating, et al., 2007;
McPherson, et al., 2001).
242
Research hypotheses
The research hypotheses in this study drew upon propositions from organizational
learning theory and institutional theory to investigate how three forms of sociomaterial
agency—the communication load of physicians, and the legitimacy and credibility of
knowledge sources—affected knowledge network structure. Hypothesis testing yielded
four notable findings.
First, the multilevel ERGM analysis indicated that communication load was not a
significant influence on the knowledge network. The qualitative data, however, suggested
that many physicians in the study population did experience communication overload.
Physician qualitative responses, for example, included the following sentiments regarding
best-practice knowledge and communication load:
“I find dealing with these new guidelines extremely confusing. I have no idea
how to incorporate them into my clinical practice. They are not easy to find. [In
order to implement them] you need the information from when patients were not
on the meds for blood pressure and cholesterol, which is not easy to find either.” –
Physician 162
“There is a lag between studies, best practice guidelines and actually doing the
work, the implementation. Because so many guidelines are changing, e.g., the
new recommendation to screen with CT for lung cancer in certain groups of
smokers, as well as new guidelines on lipid and hypertension management, there
is little time to reflect on recommendations (the research/evidence) and discuss
with colleagues whether to move and how to move forward. Often there are many
243
ambiguities in the guidelines, especially within the context of real people’s lives.”
–Physician 201
“I actually do not feel I can regurgitate the new guidelines at this time. I have very
many variations of printed information about the new guidelines that I have been
carrying around in order to read and synthesize better, but I have not had the time
and space to do so as charting and other patient-related paperwork/communication
are more pressing needs and are pressing into my weekend hours, causing me to
have very little free time. . . .That being said, I do continue to carry around my
printed explanations of the guidelines in hope of learning them, as I do hope to
comply. . . . The information overload in addition to increased patient panel
numbers and increasing patient expectations of immediate online access to me,
along with personalized phone conversations, has caused this job to rob my
oxygen and has resulted in a state of burnout and disenchantment.” –Physician
204
One explanation for these seemingly conflicting quantitative and qualitative
results is that while the physicians in the study population experienced communication
overload, their perceptions of overload did not significantly influence their knowledge
communication behavior around this particular set of best-practice guidelines. Because
the treatment of high cholesterol is a common and important task for primary care
physicians, the new cholesterol best-practice guidelines were highly salient to their
everyday work, and the physicians in the study population may have felt compelled to
learn about the guidelines despite experiencing overload. In other words, the influence of
communication load on knowledge network structure may be contingent on the
244
contextual importance of the knowledge. In this case, the new cholesterol guidelines may
have been too important for physicians to ignore. Research on the related concept of
relational carrying capacity has similarly suggested that maximum carrying capacity for a
particular set of actors and relationships may be contingent upon contextual factors such
as the cultural norms, values, and regulatory pressures associated with those relationships
(H. Aldrich & Ruef, 2006).
Several other factors also may have contributed to the non-significant quantitative
results for communication load. First, communication load is by its nature challenging to
study via survey research methods, given that the survey respondents most impacted by
communication overload may be those who are least likely to take the time to respond to
a research survey. In the present study, physicians with the highest levels of
communication load may not have responded to the survey, causing load scores to be
lower in the study than in reality. Second, perceptions of what is and what is not
communication overload are likely influenced by organizational and professional
cultures. What is considered to be an egregiously heavy workload and high
communication load in one organization or in one profession may be perceived as a
normal, acceptable amount of work and communication load in another organization or
profession. Because the communication load measure used in this study was not
developed for, and has not previously been tested in, physician populations or health care
organizations, it is possible that the measure does not adequately assess communication
load for the organizational and professional cultures of the present study.
A second notable finding from the hypothesis testing was that physicians’
perceptions of the legitimacy of one of the organizational authors of the cholesterol best
245
practice guidelines—the American College of Cardiology—exerted significant positive
influence on physician-to-physician knowledge communication. Physicians who
perceived ACC legitimacy to be greater were more likely to seek or receive knowledge
about the ACC/AHA guidelines from other physicians. This finding was the only
significant result among the hypotheses proposed regarding knowledge source legitimacy
and credibility. The legitimacy of the other best-practice author, the American Heart
Association, did not significantly influence knowledge network structure, nor did the
credibility of the best-practice authors, nor did the legitimacy and credibility of the
artifacts. Results did suggest that the credibility of physician knowledge sources
influenced the knowledge network, but this finding must be interpreted with the caveat
that the measurement of physician credibility was not optimal.
Several observations regarding the ACC legitimacy finding are warranted. First,
only the legitimacy of the ACC, and not the AHA, influenced the knowledge network.
The relative importance of the ACC’s legitimacy to the physicians in the study
population may reflect the fact that the ACC is a professional organization devoted
exclusively to the concerns of physicians, whereas the AHA has the more diffuse mission
of serving patients with heart disease, the general public, and physicians. Because of the
ACC’s more focused mission of supporting physicians and promoting physician
professional norms—its normative isomorphic influence, according to DiMaggio and
Powell (1983)—its legitimacy may be more salient to these physicians than the
legitimacy of the AHA.
Second, ACC legitimacy affected knowledge ties between physicians only, and
did not influence knowledge ties between physicians and artifacts. In fact, none of the
246
hypothesized legitimacy and credibility variables influenced physician-artifact
knowledge ties. It is possible that the legitimacy of best-practice authors—that is, the
primary sources of the best-practice knowledge—was more important to physicians than
the legitimacy or credibility of the media artifacts—the secondary sources—that report
on and summarize the knowledge. This explanation is supported by the qualitative
physician responses. When physicians referred in the qualitative data to their social
judgments of knowledge sources, they most frequently emphasized their concerns about
the legitimacy of the authors of the best-practice guidelines. For example:
“Not all clinical best practice guidelines are the same. The cholesterol guidelines
are not as evidence-based as they appear. . . . The influence of pharmaceutical
companies and other industries is still very pervasive and deep-rooted. . . . I take
[guidelines authored by] the ACP [American College of Physicians] and the
USPTF [U.S. Preventive Services Task Force] as a better gold standard than
[those authored by] specialty groups.” –Physician 27
“I am not convinced that pharmaceutical companies have not played a role in the
new lipid guidelines since they are so prescription-oriented” –Physician 33
“The AHA and ACC guidelines are suspect, as many cardiologists disagreed and
the data were not vetted. Also there is heavy pharma influence on the guidelines
and often no evidence.” –Physician 132
These comments reflect concern over the process by which the best-practice authors
formulated the new cholesterol guidelines. They reference the legitimacy of the best-
practice authors as knowledge sources, as opposed to referencing the legitimacy and
credibility of the secondary media artifact sources. These qualitative results thus may
247
help explain why ACC legitimacy in particular was important to the physicians in this
study population.
It is also possible that physicians’ standards for judging sources that they used to
become aware about the best-practice guidelines were different from their standards for
judging sources they used to learn about the guidelines. Research in both the diffusion
and health behavior change literatures make distinctions between the less rigorous
process of gaining awareness about an innovation, practice, or behavior and the more
effortful processes of substantively learning about it and applying it in practice (Glanz,
Rimer, & Viswanath, 2008; Greenhalgh, et al., 2005; Rogers, 2003).
Given these distinctions, it is likely that people judge knowledge sources used for
learning with different legitimacy and credibility standards than those they apply to
knowledge sources used for awareness. These different standards of judgment for
awareness versus learning may partly explain why legitimacy and credibility did not
influence ties between physicians and artifacts in the present study. Physicians may have
used many of the media artifacts in the study to gain awareness of the guidelines only,
and used their colleagues to more substantively learn about the guidelines. Physician 81,
for example, described such a process:
“It is often the case that I hear in the usual media sources that a new guideline or
important study has been published. They [these media sources] may give the
broad strokes of the data or guideline. I then want to locate something close to, or
the actual original, guideline or article. I then end up discussing it with my
colleagues. It is helpful when someone within the organization who has more time
248
and less pressing clinical responsibility has digested the info and presents it in a
concise format.”
Future research could test whether physicians use different standards of judgment for
“awareness” knowledge sources versus “learning” knowledge sources.
A third notable finding from the hypothesis testing was the empirical evidence
that legitimacy and credibility may be distinct but related concepts, worthy of further
scholarly comparison and differentiation. The fact that best-practice author legitimacy,
but not credibility, exerted influence on the knowledge network, offered evidence of an
empirical difference between the legitimacy and credibility concepts. This empirical
result supports theoretical understanding of the two concepts and represents a preliminary
step toward further elaboration of the under-examined relationship between them.
Fourth, the hypothesis testing offered qualified support for the proposition, stated
in Hypothesis 9, that the multilevel sociomaterial network explained more variance in
knowledge network structure than the unipartite “social” network alone or the bipartite
“material” network alone. On one hand, comparison of the goodness of fit results across
the three models indicated that the unipartite model had the fewest poorly fitting
parameters and the lowest Mahalanobis distance score. These results suggested that the
unipartite model of physician-only best-practice communication provided the best
explanation for the observed knowledge network operating within Alpha Medical Group.
On the other hand, using goodness of fit statistics to compare models with
differing numbers of modes and substantially different numbers of parameters—versus,
for example, using them to compare three unipartite models with slightly different
parameters—may be problematic. This is because more complex models with greater
249
numbers of parameters are inherently more difficult to fit and more likely to have some
poorly fitting parameters (Robins & Lusher, 2013), and because the Mahalanobis
distance score, like many such goodness of fit heuristics, rewards parsimony and
penalizes for the use of greater numbers of parameters (Wang, et al., 2009). Better, more
precise tools and techniques are needed to evaluate goodness of fit across models of
varying complexity and size, particularly in the present context of estimating multilevel
network models.
Further, although comparison of the goodness-of-fit results suggested that the
unipartite model offered a better explanation of the observed network than the multilevel
model, comparison of the significance of the parameter estimates across the three
network models indicated that there were two non-significant parameters in the unipartite
model that became significant in the multilevel model. Taken together, the results of
testing Hypothesis 9, while not conclusive, lend credence to the overarching thesis of this
dissertation, that examination of sociomaterial networks holds substantial promise for
extending our understanding of knowledge networks specifically and of network structure
generally.
Limitations
This study was characterized by a number of limitations. As noted throughout, the
way that physician credibility was measured was not ideal given the lack of credibility
data available for all physicians in the study population. Asking physicians to assess the
credibility of all 185 of their colleagues would have resulted in excessive respondent
burden, but the absence of these data limited the degree to which conclusions could be
drawn from the physician credibility parameter estimates. Additionally, the measurement
250
of the credibility of the physician and artifact knowledge sources using monadic (as
opposed to dyadic) attributes applied to the source nodes was problematic. The monadic
approach involved converting the physician and artifact credibility scores from the dyadic
form in which they were collected into a monadic form by averaging the credibility
scores a physician or artifact source received from all physician respondents. I chose to
measure these variables using monadic attributes, because MPNet currently does not
accommodate analysis of dyadic versions of attributes in multilevel models, and because
it was not feasible to collect physicians’ credibility perceptions about all of their
colleagues and all of the knowledge artifacts they and their colleagues used. Thus use of
dyadic credibility scores for physicians and artifacts was not technically possible.
However, theoretical consensus, as discussed in Chapter 4, suggests that whereas
legitimacy is a collective evaluation that may be possible in many contexts to
appropriately measure monadically, credibility is an individual evaluation. One could
therefore argue that physician and artifact source credibility would be more appropriately
measured using a dyadic attribute, which considers only the credibility judgment of the
focal actor in the dyad, rather than the aggregate credibility judgment of all actors in the
network.
The inability in MPNet to measure physician and artifact credibility using dyadic
attributes is reflective of the broader challenges posed in this study by using the MPNet
analytical software. One of the key advantages of using multilevel ERGM in MPNet to
analyze network phenomena is that it allows for much greater model flexibility and
complexity than was previously possible. Despite this advantage, there were challenges
to fitting the data and hypotheses of the present project into the restrictions inherent in the
251
currently-available version of MPNet. For example, I collected data on primary care
physicians’ knowledge communication ties with specialist physicians (e.g., cardiologists)
and pharmacists. I identified these specialist physicians and pharmacists as additional
important sources of best-practice knowledge through preliminary interviews with Alpha
Medical Group executives, and their importance was confirmed in physicians’ qualitative
responses collected via the survey instrument. After much research, however, I concluded
that there was no way to appropriately conceptualize the multilevel network in MPNet
that would allow me to include the specialists and pharmacists as nodes in the network. In
addition to the challenges posed by MPNet modeling restrictions, multilevel network
model estimation in MPNet requires an investment of substantial computing power and
time that is likely not feasible in many research settings.
A third limitation of the present project is the classic weakness of cross-sectional
research designs. Cross-sectional designs cannot determine causality or verify the
appropriateness of a researcher’s choice to apply a social selection versus social influence
model to empirical network data. The problem of cross-sectional designs is particularly
acute in the network literature, where the paucity of longitudinal research is glaring and
the methodological tools available to analyze longitudinal data have become increasingly
sophisticated and easy to use (Ripley, et al., 2013). One could productively extend the
findings of the present research by collecting a second wave of data and investigating, for
example, how communication load, legitimacy, and credibility affect not only best-
practice knowledge communication but also knowledge implementation, physician
behavior, and patient outcomes.
252
Contributions and Directions for Future Research
Despite these limitations, this study makes a number of contributions, organized
here according to the audience to which each contribution is most relevant.
First, for health care practitioners and policymakers, the study results both
confirm and contradict conventional wisdom and scholarly research about the
communication of best-practice knowledge. In health care, as in many other fields, there
is great emphasis on and supply of best-practice knowledge, and at the same time, much
concern over the challenges of communicating such knowledge. The large quantity and
uncertain quality of best-practice knowledge are frequently discussed as two key barriers
to its dissemination and diffusion, and I operationalized these barriers in the present study
as physician communication load and knowledge source legitimacy and credibility. Study
results indicated, in partial contradiction of extant understanding, that although
physicians are concerned about communication overload, their perceptions of load did
not influence communication about the 2013 cholesterol guidelines, perhaps because of
the high salience of these guidelines to primary care physicians’ daily work. Study results
also indicated, in confirmation of extant understanding, that physicians’ perceptions of
the legitimacy of the organizational authors of the cholesterol guidelines significantly
influenced knowledge network structure. The legitimacy and credibility of the specific
media artifacts that communicated knowledge of the guidelines, however, were not
significant influences, suggesting that audience perceptions of the original organizational
authors and authorship process are more important for subsequent knowledge diffusion
efforts than audience perceptions of the many secondary, artifactual sources of best
practice information available to health care practitioners.
253
Taken together, the results on communication load, legitimacy, and credibility
suggest that—in the case of highly salient best practices, at least—the most important
determinant of a physician’s engagement in best-practice communication is his or her
perception of the legitimacy of the organizational authors of the best practice. Future
research could build upon these results by investigating if organizational author
legitimacy also plays a key role in whether a physician changes his or her professional
behavior, by adopting/adapting and implementing the new practice as a result of
communicating about it. Future research could also refine and further test these results by
investigating the influence of communication load, legitimacy, and credibility on best-
practice communication in networks of physicians working in different specialties, in
different organizational contexts, and in rural (as opposed to urban) geographic areas, as
well as in networks of health care practitioners who are not physicians. Because the
physicians in this study worked in a large health care provider organization with a
sophisticated information technology infrastructure, and because they worked in a
geographic region in which the supply of primary care physicians is relatively greater and
the volume of health care services provided to patients is relatively higher, the factors
that influenced the Alpha Medical Group physicians to form best-practice knowledge ties
may be different from the factors that influence other physician populations. Therefore,
future research could also test this study’s research hypotheses in other types of
organizational and geographic contexts.
Second, for organizational learning scholars, this project highlights the somewhat
forgotten concept of communication load as being worth a second look. The study of
communication load and overload originates, at least in part, in scholars’ early interest in
254
the difficulty and messiness of the processes of organizational decision-making and
learning (Case, 2008; Cyert & March, 1963; Friedman, et al., 2005; Galbraith, 1977;
Huber, 1991; Simon, 1957, 1979; Stephens, 2012). This dissertation’s integration of
organizational learning’s concept of communication load with network science represents
a potentially fresh and productive approach to examining load, overload, and related
concepts such as carrying capacity (H. Aldrich & Ruef, 2006; Hannan & Freeman, 1989;
P. Monge, et al., 2008; Oldroyd & Morris, 2012), absorptive capacity (Cohen &
Levinthal, 1990; Zahra & George, 2002), and attention (Ocasio, 2011). In fact, network
science, with its focus on characteristics of the relationships between two or more actors
in a network, is well-suited to the study of communication load and overload. Network
analysis allows researchers to examine how the frequency or strength of relations
between actors affects knowledge communication. That network science has not been
used more frequently to study communication overload, generally thought of as a
negative phenomenon or outcome, may be the result of the early (and ongoing) emphasis
in network science on the positive, rather than negative, outcomes of tie formation
between two actors (Labianca & Brass, 2006). An approach that integrates organizational
learning theory’s concept of communication load with network science’s methodological
tools may thus generate potentially productive lines of research for both fields.
Third, for institutional theorists, this study makes contributions to two current
trends in the research literature. In its focus on how individuals’ legitimacy judgments
influence organizational knowledge communication, the study demonstrates how the
combination of institutional theory with network science offers a promising approach to
elucidating the microprocesses and microfoundations of organizational institutionalism, a
255
popular topic of late in the institutional theory literature (Bechky, 2011; Bitektine, 2011;
Bitektine & Haack, 2015; Felin, et al., 2015; Gondo & Amis, 2013; R. Greenwood, et al.,
2008; Powell & Colyvas, 2008; Suddaby, et al., 2010; Tost, 2011). Relatedly, the present
study also represents a first step in clarifying the distinctions between the concepts of
legitimacy and credibility, which have been silo-ed off into their respective disciplinary
domains of organizational theory and communication and persuasion, with very little
cross-pollination. Given the flurry of recent work done by organizational theorists to
productively disentangle the concepts of legitimacy, status, and reputation (Bitektine,
2011; Deephouse & Carter, 2005; Deephouse & Suchman, 2008; Devers, et al., 2009;
Tost, 2011), a logical next step is to incorporate credibility into the integrated picture,
gradually coming into focus, of individual and social judgments that influence
organizational knowledge communication.
Fourth, for social network research and network science, this study builds on a
small body of recent theoretical (Contractor, 2009; Contractor, et al., 2011) and empirical
(Keegan, et al., 2012; Su & Contractor, 2011) work by investigating the structure of
knowledge communication in a network composed of both human and nonhuman actors.
There have been hundreds of insightful studies published on social-only knowledge
networks, but fewer studies that consider multiple types of actors, human or otherwise, in
knowledge networks. Employing a sociomaterial ontology, this study offered a
conceptualization of organizational knowledge networks that may lead to a richer, more
complete understanding of the structure of knowledge communication. Research on
sociomaterial knowledge networks offers the promise of insights on the similarities and
differences across networks with different types of nodes, and on what is distinctive
256
between social knowledge networks and sociomaterial knowledge networks, thus
enhancing our understanding of both.
A number of future directions in the realm of network analysis are possible. A
follow-up to the present study, for example, could take advantage of the full potential of
MPNet by analyzing a unipartite artifact-artifact network, in addition to the present
study’s unipartite physician-physician network, bipartite physician-artifact network, and
combined multilevel network. Such a unipartite artifact-artifact network could represent
knowledge sharing between artifactual knowledge sources, building on recent related
work on agenda-setting networks (Ognyanova, 2012) and news media hyperlink
networks (Weber, Chung, & Park, 2012). Future work could also apply a sociomaterial
ontology to examine other networks, in addition to knowledge networks, that are
characterized by a heterogeneous actors. For example, researchers might productively
include animals (e.g., pets) as well as humans in investigations of people’s “social”
support networks (Beck & Meyers, 1996; Siegel, 1990; Wood, Giles-Corti, & Bulsara,
2005).
Fifth, for the still-emerging field of sociomateriality, the present project provides
an example both of how existing organizational theory may be productively paired with
the research on sociomateriality, and of how a sociomaterial ontology and epistemology
may be productively applied to empirical data. As several scholars have noted, there is a
need for the integration of sociomateriality with established organizational theories
(Leonardi, 2013a; Robey, et al., 2012). Likewise, scholars have argued for
sociomaterialists to move beyond publication of densely philosophical debates to the
application of sociomaterial concepts to practical, empirical contexts (Leonardi, 2013b).
257
Doing so holds the promise of demonstrating to a greater number of scholars the utility of
sociomateriality as an ontology for network as well as other social science research.
Conclusion
In this study, I theorized and empirically tested a sociomaterial model of
organizational knowledge communication. I used sociomateriality as an ontological
foundation to conceptualize best-practice knowledge communication as occurring in a
network in which the actors were both people and artifacts. I drew on propositions from
organizational learning theory and institutional theory to investigate how three forms of
sociomaterial agency—the communication load of knowledge consumers, and the
legitimacy and credibility of knowledge sources—influenced the structure of the
knowledge network. I tested these hypothesized influences with recently-developed
exponential random graph models for multilevel networks, producing new insights on (1)
the contingent nature of communication load effects on knowledge networks; (2) the
importance of best-practice-author legitimacy in structuring knowledge communication;
(3) the distinctions between the concepts of legitimacy and credibility; and (4) the
potential for multilevel sociomaterial networks to explain more variance in knowledge
network structure than unipartite or bipartite networks alone.
Sociomaterial models of organizational knowledge networks, like the one
presented in this study, advance our understanding of the structure and operation of real
people in real organizations. They represent a new approach in organizational
communication and network theory by extending existing conceptualizations of
knowledge networks from the social-only to the sociomaterial. In theorizing and testing a
sociomaterial model here, my objective was to offer a model of organizational knowledge
258
that is more realistic for representing the numerous empirical contexts in which both
human and material knowledge sources communicate.
259
REFERENCES
Aakhus, M., Ballard, D., Flanagin, A. J., Kuhn, T., Leonardi, P. M., Mease, J., et al.
(2011). Communication and materiality: A conversation from the CM café.
Communication Monographs, 78(4), 557-568. doi:
10.1080/03637751.2011.618358
Abrahamson, E. (1996). Management fashion. Academy of Management Review, 21(1),
254-285.
Abramson, J. D., & Redberg, R. F. (2013, November 13). Don't give more patients
statins, New York Times. Retrieved from
http://www.nytimes.com/2013/11/14/opinion/dont-give-more-patients-
statins.html?ref=opinion
Adler, P. S., & Kwon, S. (2002). Social capital: Prospects for a new concept. Academy of
Management Review, 27, 17-40. doi: 10.2307/4134367
Agneessens, F., & Wittek, R. (2012). Where do intra-organizational advice relations
come from? The role of informal status and social capital in social exchange.
Social Networks, 34, 333-345. doi: 10.1016/j.socnet.2011.04.002
Aldrich, H., & Ruef, M. (2006). Organizations evolving (2nd ed.). Thousand Oaks, CA:
Sage.
Aldrich, H. E., & Fiol, C. M. (1994). Fools rush in? The institutional context of industry
creation. Academy of Management Review, 19(4), 645-670. doi:
10.5465/AMR.1994.9412190214
Allison, P. D. (2010). Missing data. In P. V. Marsden & J. D. Wright (Eds.), Handbook of
survey research (2nd ed., pp. 631-658). Bingley, UK: Emerald Group.
Alwin, D. F. (2010). How good is survey measurement? Assessing the reliability and
validity of survey measures. In P. V. Marsden & J. D. Wright (Eds.), Handbook of
survey research (2nd ed., pp. 405-436). Bingley, UK: Emerald Group Publishing.
Argote, L. (2005). Reflections on two views of managing learning and knowledge in
organizations. Journal of Management Inquiry, 14(1), 43-48. doi:
10.1177/1056492604273179
Argote, L., Ingram, P., Levine, J. M., & Moreland, R. L. (2000). Knowledge transfer in
organizations: Learning from the experience of others. Organizational Behavior
and Human Decision Processes, 82(1), 1-8. doi: 10.1006/obhd.2000.2883
Argote, L., McEvily, B., & Reagans, R. (2003). Managing knowledge in organizations:
An integrative framework and review of emerging themes. Management Science,
49(4), 571-582. doi: 10.1287/mnsc.49.4.571.14424
260
Argote, L., & Miron-Spektor, E. (2011). Organizational learning: From experience to
knowledge. Organization Science, 22(5), 1123-1137. doi:
10.1287/orsc.1100.0621
Asch, S. E. (1951). Effects of group pressure upon the modification and distortion of
judgments. In H. Guetzkow (Ed.), Groups, leadership and men: Research in
human relations (pp. 177-188). Pittsburgh, PA: Carnegie Press.
Ashcraft, K. L., Kuhn, T., & Cooren, F. (2009). Constitutional amendments:
"Materializing" organizational communication. Academy of Management Annals,
3(1), 1-64. doi: 10.1080/19416520903047186
Ashforth, B. E., & Gibbs, B. W. (1990). The double-edge of organizational legitimation.
Organization Science, 1(2), 177-194. doi: 10.1287/orsc.1.2.177
Avorn, J. (2013, June 12). Healing the overwhelmed physician, The New York Times, p.
A27.
Baker, W. E., & Faulkner, R. R. (2002). Interorganizational networks. In J. A. C. Baum
(Ed.), The Blackwell companion to organizations (pp. 520-540). Oxford, UK:
Blackwell Publishers.
Balas, E. A., & Boren, S. A. (2000). Managing clinical knowledge for health care
improvement. In J. H. van Bemmel & A. T. McCray (Eds.), Yearbook of medical
informatics 2000: Patient-centered systems (pp. 65-70). Stuttgart, Germany:
Schattauer.
Ballard, D. I., & Seibold, D. R. (2006). The experience of time at work: Relationship to
communication load, job satisfaction, and interdepartmental communication.
Communication Studies, 57(3), 317-340. doi: 10.1080/10510970600845974
Banks, S. P., & Riley, P. (1993). Structuration theory as an ontology for communication
research. Communication Yearbook, 16, 167-196.
Barad, K. (2003). Posthumanist performativity: Toward an understanding of how matter
comes to matter. Signs, 28(3), 801-831. doi: 10.1086/345321
Barad, K. (2007). Meeting the universe halfway: Quantum physics and the entanglement
of matter and meaning. Durham, NC: Duke University Press.
Barbour, J. B. (2010). On the institutional moorings of talk in health care organizations.
Management Communication Quarterly, 24(3), 449-456.
Barbour, J. B., & Lammers, J. C. (2007). Health care institutions, communication, and
physicians' experience of managed care: A multilevel analysis. Management
Communication Quarterly, 21(2), 201-231. doi: 10.1177/0893318907308747
261
Barley, S. R., Meyerson, D. E., & Grodal, S. (2011). E-mail as a source and symbol of
stress. Organization Science, 22(4), 887-906. doi: 10.1287/orsc.1100.0573
Battilana, J. (2011). The enabling role of social position in diverging from the
institutional status quo: Evidence from the UK National Health Service.
Organization Science, 22(4), 817-834. doi: 10.1287/orsc.1100.0574
Bawden, D., & Robinson, L. (2009). The dark side of information: overload, anxiety and
other paradoxes and pathologies. Journal of Information Science, 35(2), 180-191.
doi: 10.1177/0165551508095781
Bechky, B. A. (2003). Object lessons: Workplace artifacts as representations of
occupational jurisdiction. American Journal of Sociology, 109(3), 720-752.
Bechky, B. A. (2011). Making organizational theory work: Institutions, occupations, and
negotiated orders. Organization Science, 22(5), 1157-1167. doi:
10.1287/orsc.1100.0603
Beck, A. M., & Meyers, N. M. (1996). Health enhancement and companion animal
ownership. Annual Review of Public Health, 17, 247-257.
Bell, G. G., & Zaheer, A. (2007). Geography, networks, and knowledge flow.
Organization Science, 18(6), 955-972. doi: 10.1287/orsc.1070.0308
Bennett, W. L., & Manheim, J. B. (2006). The one-step flow of communication. Annals
of the American Academy of Political and Social Science, 608, 213-232. doi:
10.1177/0002716206292266
Bitektine, A. (2011). Toward a theory of social judgments of organizations: The case of
legitimacy, reputation, and status. Academy of Management Review, 36(1), 151-
179. doi: 10.5465/AMR.2011.55662572
Bitektine, A., & Haack, P. (2015). The "macro" and the "micro" of legitimacy: Toward a
multilevel theory of the legitimacy process. Academy of Management Review,
40(1), 49-75. doi: 10.5465/amr.2013.0318
Blackler, F. (1995). Knowledge, knowledge work and organizations: An overview and
interpretation. Organization Studies, 16(6), 1021-1046. doi:
10.1177/017084069501600605
Blair, A. M. (2003). Reading strategies for coping with information overload ca. 1550-
1700. Journal of the History of Ideas, 64(1), 11-28.
Blair, A. M. (2010). Too much to know: Managing scholarly information before the
modern age. New Haven, CT: Yale University Press.
Bodenheimer, T. (2006). Primary care -- Will it survive? New England Journal of
Medicine, 355(9), 861-864.
262
Bodenheimer, T., & Pham, H. H. (2010). Primary care: Current problems and proposed
solutions. Health Affairs, 29(5), 799-805. doi: 10.1377/hlthaff.2010.0026
Bollen, K. A. (1989). Structural equations with latent variables. New York: John Wiley
& Sons.
Borgatti, S. P., & Cross, R. (2003). A relational view of information seeking and learning
in social networks. Management Science, 49(4), 432-445.
Borgatti, S. P., & Foster, P. C. (2003). The network paradigm in organizational research:
A review and typology. Journal of Management, 29(6), 991-1013.
Borgatti, S. P., & Halgin, D. S. (2011). On network theory. Organization Science, 22(5),
1168-1181. doi: 10.1287/orsc.1100.0641
Brandes, U., Robins, G., McCranie, A., & Wasserman, S. (2013). What is network
science? Network Science, 1(1), 1-15. doi: 10.1017/nws.2013.2
Brashers, D. E., Goldsmith, D. J., & Hsieh, E. (2002). Information seeking and avoiding
in health contexts. Human Communication Research, 28(2), 258-271. doi:
10.1111/j.1468-2958.2002.tb00807.x
Brass, D. J. (1985). Men's and women's networks: A study of interaction patterns and
influence in an organization. Academy of Management Journal, 28(2), 327-342.
doi: 10.2307/256204
Brass, D. J., Galaskiewicz, J., Greve, H. R., & Tsai, W. (2004). Taking stock of networks
and organizations: A multilevel perspective. Academy of Management Journal,
47(6), 795-817. doi: 10.2307/20159624
Brennecke, J., & Rank, O. N. (in press). The interplay between formal project
memberships and informal advice seeking in knowledge-intensive firms: A
multilevel network approach. Social Networks.
Brewer, D. D. (2000). Forgetting in the recall-based elicitation of personal and social
networks. Social Networks, 22(1), 29-43. doi: 10.1016/S0378-8733(99)00017-9
Brown, J. S., & Duguid, P. (2001). Knowledge and organization: A social-practice
perspective. Organization Science, 12(2), 198-213.
Brown, L. E. (Ed.) (1993) The new shorter Oxford English dictionary (4th ed., Vols. 1).
Oxford: Clarendon Press.
Brown, T. A. (2006). Confirmatory factor analysis for applied research. New York:
Guilford Press.
263
Browne, M. W., & Cudeck, R. (1993). Alternate ways of assessing model fit. In K. A.
Bollen & J. S. Long (Eds.), Testing structural equation models (pp. 136-162).
Newbury Park, CA: Sage.
Bryant, F. B., & Satorra, A. (2012). Principles and practice of scaled difference chi-
square testing. Structural Equation Modeling, 19(3), 372-398. doi:
10.1080/10705511.2012.687671
Bryant, F. B., & Satorra, A. (2013). EXCEL macro file for conducting scaled difference
chi-square tests via LISREL 8, LISREL 9, EQS, and Mplus. Retrieved from
http://www.econ.upf.edu/~satorra/
Bryant, J., & Oliver, M. B. (Eds.). (2009). Media effects: Advances in theory and
research (3rd ed.). New York: Routledge.
Burns, L. R., & Wholey, D. R. (1993). Adoption and abandonment of matrix
management programs: Effects of organizational characteristics and
interorganizational networks. Academy of Management Journal, 36(1), 106-138.
Burt, R. S. (1987). Social contagion and innovation: Cohesion versus structural
equivalence. American Journal of Sociology, 92, 1287-1335.
Burt, R. S. (1992). Structural holes. Cambridge, MA: Harvard University Press.
Butts, C. T. (2009). Revisiting the foundations of network analysis. Science, 325, 414-
416. doi: 10.1126/science.1171022
Byström, K. (2002). Information and information sources in tasks of varying complexity.
Journal of the American Society for Information Science and Technology, 53(7),
581-591. doi: 10.1002/asi.10064
Cabana, M. D., Rand, C. S., Powe, N. R., Wu, A. W., Wilson, M. H., Abboud, P.-A. C.,
et al. (1999). Why don't physicians follow clinical practice guidelines? A
framework for improvement. Journal of the American Medical Association,
282(15), 1458-1465. doi: 10.1001/jama.282.15.1458
Calhoun, M. A., Starbuck, W. H., & Abrahamson, E. (2011). Fads, fashions, and the
fluidity of knowledge: Peter Senge's 'The Learning Organization'. In M. Easterby-
Smith & M. A. Lyles (Eds.), Handbook of organizational learning and knowledge
management (2nd ed., pp. 225-248). Chichester, UK: John Wiley & Sons.
Callon, M. (1986). Some elements of a sociology of translation: Domestication of the
scallops and the fishermen of St. Brieuc Bay. In J. Law (Ed.), Power, action and
belief: A new sociology of knowledge? (pp. 196-223). London: Routledge.
Callon, M. (1991). Techno-economic networks and irreversibility. In J. Law (Ed.), A
sociology of monsters: Essays on power, technology and domination (pp. 131-
161). London: Routledge.
264
Callon, M. (2004). The role of hybrid communities and socio-technical arrangements in
the participatory design. Journal of the Center for Information Studies, 5(3), 3-10.
Callon, M., & Muniesa, F. (2005). Peripheral vision: Economic markets as calculative
collective devices. Organization Studies, 26(8), 1229-1250. doi:
10.1177/0170840605056393
Canary, H. E. (2011). Knowledge types in cross-system policy knowledge construction.
In H. E. Canary & R. D. McPhee (Eds.), Communication and organizational
knowledge: Contemporary issues for theory and practice (pp. 244-263). New
York: Routledge.
Canary, H. E., & McPhee, R. D. (2011). Toward a communicative perspective on
organizational knowledge. In H. E. Canary & R. D. McPhee (Eds.),
Communication and organizational knowledge: Contemporary issues for theory
and practice (pp. 1-14). New York: Routledge.
Carpenter, M. A., Li, M., & Jiang, H. (2012). Social network research in organizational
contexts: A systematic review of methodological issues and choices. Journal of
Management, 28(4), 1328-1361. doi: 10.1177/0149206312440119
Casciaro, T., & Lobo, M. S. (2008). When competence is irrelevant: The role of
interpersonal affect in tast-related ties. Administrative Science Quarterly, 53(4),
655-684. doi: 10.2189/asqu.53.4.655
Case, D. O. (2008). Looking for information: A survey of research on information
seeking, needs, and behavior (2nd ed.). Bingley, UK: Emerald Group.
Cash, D. W., Clark, W. C., Alcock, F., Dickson, N. M., Eckley, N., Guston, D. H., et al.
(2003). Knowledge systems for sustainable development. Proceedings of the
National Academy of Sciences, 100(14), 8086-8091. doi:
10.1073pnas.1231332100
Castor, T., & Cooren, F. (2006). Organizations as hybrid forms of life: The implications
of the selection of agency in problem formulation. Management Communication
Quarterly, 19(4), 570-600. doi: 10.1177/0893318905284764
Certo, S. T., & Hodge, F. (2007). Top management team prestige and organizational
legitimacy: An examination of investor perceptions. Journal of Managerial
Issues, 19(4), 461-477.
Chaffee, S. H., & Metzger, M. J. (2001). The end of mass communication? Mass
Communication and Society, 4(4), 365-379. doi: 10.1207/S15327825MCS0404_3
Chen, W., & Lee, K.-H. (2013). Sharing, liking, commenting, and distressed? The
pathway between facebook interaction and psychological distress.
Cyberpsychology, Behavior, and Social Networking, 16(10), 728-734. doi:
10.1089/cyber.2012.0272
265
Cheney-Lippold, J. (2011). A new algorithmic identity: Soft biopolitics and the
modulation of control. Theory, Culture & Society, 28(1). doi:
10.1177/0263276411424420
Cho, J., Ramgolam, D. I., Schaefer, K. M., & Sandlin, A. N. (2011). The rate and delay in
overload: An investigation of communication overload and channel synchronicity
on identification and job satisfaction. Journal of Applied Communication
Research, 39(1), 38-54. doi: 10.1080/00909882.2010.536847
Chung, C. J., & Goldhaber, G. (1991). Measuring communication load: A three-
dimensional instrument. Paper presented at the International Communication
Association Conference, Chicago.
Chung, J. Y., Berger, B. K., & DeCoster, J. (2011). Developing measurement scales of
organizational and issue legitimacy in issue management: A case of direct-to-
consumer advertising in the pharmaceutical industry. Paper presented at the
International Communication Association Conference, Boston, MA.
Cohen, W. M., & Levinthal, D. A. (1990). Absorptive capacity: A new perspective on
learning and innovation. Administrative Science Quarterly, 35(1), 128-152.
Coleman, J. S., Katz, E., & Menzel, H. (1966). Medical innovation: A diffusion study.
New York: Bobbs Merrill.
Collins, H. M. (1993). The structure of knowledge. Social Research, 60(1), 95-116.
Colyvas, J. A. (2007). From divergent meanings to common practices: The early
institutionalization of technology transfer in the life sciences at Stanford
University. Research Policy, 36, 456-476. doi: 10.1016/j.respol.2007.02.019
Colyvas, J. A., & Powell, W. W. (2006). Roads to institutionalization: The remaking of
boundaries between public and private science. Research in Organizational
Behavior, 27, 305-353. doi: 10.1016/S0191-3085(06)27008-4
Contractor, N. (2009). The emergence of multidimensional networks. Journal of
Computer-Mediated Communication, 14(3), 743-747.
Contractor, N., Monge, P. R., & Leonardi, P. M. (2011). Multidimensional networks and
the dynamics of sociomateriality: Bringing technology inside the network.
International Journal of Communication, 5, 682-720.
Cook, S. D. N., & Brown, J. S. (1999). Bridging epistemologies: The generative dance
between organizational knowledge and organizational knowing. Organization
Science, 10(4), 381-400. doi: 10.1287/orsc.10.4.381
Cooren, F. (2004). Textual agency: How texts do things in organizational settings.
Organization, 11(3), 373-393. doi: 10.1177/1350508404041998
266
Cooren, F. (2006). The organizational world as a plenum of agencies. In F. Cooren, J. R.
Taylor & E. J. Van Every (Eds.), Communication as organizing: Empirical and
theoretical explorations in the dynamic of text and conversation (pp. 81-100).
Mahway, NJ: Lawrence Erlbaum Associates.
Corman, S. R., & Dooley, K. J. (2011). Problems and promises of managing explicit
knowledge: The ideal case of university research. In H. E. Canary & R. D.
McPhee (Eds.), Communication and organizational knowledge: Contemporary
issues for theory and practice (pp. 151-170). New York: Routledge.
Cornelissen, J. P., Durand, R., Fiss, P. C., Lammers, J. C., & Vaara, E. (2015). Putting
communication front and center in institutional theory and analysis. Academy of
Management Review, 40(1), 10-27. doi: 10.5465/amr.2014.0381
Coromina, L., & Coenders, G. (2006). Reliability and validity of egocentered network
data collected via Web: A meta-analysis of multilevel multitrait multimethod
studies. Social Networks, 28(3), 209-231. doi: 10.1016/j.socnet.2005.07.006
Council on Graduate Medical Education. (2007). New paradigms for physician training
for improving access to health care, 18th report. Rockville, MD: U.S. Health
Resources and Services Administration.
Cross, R., Borgatti, S. P., & Parker, A. (2001). Beyond answers: Dimensions of the
advice network. Social Networks, 23, 215-235.
Cross, R., & Sproull, L. (2004). More than an answer: Information relationships for
actionable knowledge. Organization Science, 15(4), 446-462. doi: 10. 1287/orsc.
1040.0075
Cummings, J. N. (2004). Work groups, structural diversity, and knowledge sharing in a
global organization. Management Science, 50(3), 352-364.
Cyert, R. M., & March, J. G. (1963). A behavioral theory of the firm. Englewood Cliffs,
NJ: Prentice-Hall.
Czarniawska, B. (2008). How to misuse institutions and get away with it: Some
reflections on institutional theory(ies). In R. Greenwood, C. Oliver, K. Sahlin &
R. Suddaby (Eds.), The SAGE Handbook of Organizational Institutionalism (pp.
769-782). Los Angeles: Sage.
D'Adderio, L. (2011). Artifacts at the centre of routines: Performing the material turn in
routines theory. Journal of Institutional Economics, 7(2), 197-230. doi:
10.1017/S174413741000024X
Dabbish, L. A., Kraut, R. E., Fussell, S., & Kiesler, S. (2005). Understanding email use:
Predicting action on a message. Paper presented at the CHI 2005, Portland, OR.
267
Darr, E. D., & Kurtzberg, T. R. (2000). An investigation of partner similarity dimensions
on knowledge transfer. Organizational Behavior and Human Decision Processes,
82(1), 28-44. doi: 10.1006/obhd.2000.2885
Davenport, T. H., & Prusak, L. (1998). Working knowledge: How organizations manage
what they know. Boston, MA: Harvard Business School Press.
Davidson, E. R., & Chismar, W. G. (2007). The interaction of institutionally triggered
and technology-triggered social structure change: An investigation of
computerized physician order entry. MIS Quarterly, 31(4), 739-758.
Davis, G. F., & Greve, H. R. (1997). Corporate elite networks and governance changes in
the 1980s. American Journal of Sociology, 103(1), 1-37.
Dearing, J. W. (2004). Improving the state of health programming by using diffusion
theory. Journal of Health Communication, 9, 21-36.
Deephouse, D. L. (1996). Does isomorphism legitimate? Academy of Management
Journal, 39(4), 1024-1039. doi: 10.2307/256722
Deephouse, D. L., & Carter, S. M. (2005). An examination of differences between
organizational legitimacy and organizational reputation. Journal of Management
Studies, 42(2), 329-360.
Deephouse, D. L., & Suchman, M. (2008). Legitimacy in organizational institutionalism.
In R. Greenwood, C. Oliver, K. Sahlin & R. Suddaby (Eds.), The SAGE handbook
of organizational institutionalism (pp. 49-77). Los Angeles: Sage.
Deng, L., & Poole, M. S. (2011). Knowledge utilization in electronic networks of
practice. In H. E. Canary & R. D. McPhee (Eds.), Communication and
organizational knowledge: Contemporary issues for theory and practice (pp. 209-
220). New York: Routledge.
DeSanctis, G., & Poole, M. S. (1994). Capturing the complexity in advanced technology
use: Adaptive structuration theory. Organization Science, 5(2), 121-147.
DeVellis, R. F. (2003). Scale development: Theory and applications (2nd ed.). Thousand
Oaks: Sage Publications.
Devers, C. E., Dewett, T., Mishina, Y., & Belsito, C. A. (2009). A general theory of
organizational stigma. Organization Science, 20(1), 154-171.
DiMaggio, P. J., & Powell, W. W. (1983). The iron cage revisited: Institutional
isomorphism and collective rationality in organizational fields. American
Sociological Review, 48, 147-160.
268
DiMaggio, P. J., & Powell, W. W. (1991). Introduction. In W. W. Powell & P. J.
DiMaggio (Eds.), The new institutionalism in organizational analysis (pp. 1-38).
Chicago: University of Chicago Press.
Dowling, J., & Pfeffer, J. (1975). Organizational legitimacy: Social values and
organizational behavior. Pacific Sociological Review, 18(1), 122-136. doi:
10.2307/1388226
Downs, J., & Good, C. (2014). New cholesterol guidelines: Has Godot finally arrived?
Annals of Internal Medicine, 160(5), 354-356. doi: 10.7326/M13-2850
Dunn, M. B., & Jones, C. (2010). Institutional logics and institutional pluralism: The
contestation of care and science logics in medical education, 1967-2005.
Administrative Science Quarterly, 55, 114-149. doi: 10.2189/asqu.2010.55.1.114
Dyer, J. H., & Nobeoka, K. (2000). Creating and managing a high-performance
knowledge-sharing network: The Toyota case. Strategic Management Journal,
21(3), 345-367.
Easterby-Smith, M., & Lyles, M. A. (2011a). The evolving field of organizational
learning and knowledge management. In M. Easterby-Smith & M. A. Lyles
(Eds.), Handbook of organizational learning and knowledge management (2nd
ed., pp. 1-22). Chichester, UK: John Wiley & Sons.
Easterby-Smith, M., & Lyles, M. A. (Eds.). (2011b). Handbook of organizational
learning and knowledge management (2nd ed.). Chichester, UK: John Wiley &
Sons.
Easterby-Smith, M., Lyles, M. A., & Tsang, E. W. K. (2008). Inter-organizational
knowledge transfer: Current themes and future prospects. Journal of Management
Studies, 45(4), 677-690.
Elsbach, K. D. (1994). Managing organizational legitimacy in the California cattle
industry: The construction and effectiveness of verbal accounts. Administrative
Science Quarterly, 39(1), 57-88. doi: 10.2307/2393494
Eppler, M. J., & Mengis, J. (2004). The concept of information overload: A review of
literature from organization science, accounting, marketing, MIS, and related
disciplines. The Information Society, 20(5), 325-344. doi:
10.1080/01972240490507974
Esterling, K. M. (2004). The political economy of expertise: Information and efficiency in
American national politics. Ann Arbor: University of Michigan Press.
Farace, R. V., Monge, P. R., & Russell, H. M. (1977). Communicating and organizing.
Reading, MA: Addison-Wesley.
269
Faraj, S., & Azad, B. (2012). The materiality of technology: An affordance perspective.
In P. M. Leonardi, B. A. Nardi & J. Kallinikos (Eds.), Materiality and organizing:
Social interaction in a technological world (pp. 237-258). Oxford: Oxford
University Press.
Faulkner, P., & Runde, J. (2012). On sociomateriality. In P. M. Leonardi, B. A. Nardi &
J. Kallinikos (Eds.), Materiality and organizing: Social interaction in a
technological world (pp. 49-66). Oxford: Oxford University Press.
Faust, K. (2011). Animal social networks. In J. Scott & P. J. Carrington (Eds.), The SAGE
handbook of social network analysis (pp. 148-166). London: SAGE Publications.
Feldman, M. S., & Orlikowski, W. J. (2011). Theorizing practice and practicing theory.
Organization Science, 22(5), 1240-1253. doi: 10.1287/orsc.1100.0612
Felin, T., Foss, N. J., & Ployhart, R. E. (2015). The microfoundations movement in
strategy and organization theory. Academy of Management Annals, 9(1), 575-632.
doi: 10.1080/19416520.2015.1007651
Ferlie, E., Fitzgerald, L., Wood, M., & Hawkins, C. (2005). The nonspread of
innovations: The mediating role of professionals. Academy of Management
Journal, 48(1), 117-134.
Field, M. J., & Lohr, K. N. (Eds.). (1990). Clinical practice guidelines: Directions of a
new program. Institute of Medicine. Washington, DC: National Academy Press.
Fiol, C. M., & Lyles, M. A. (1985). Organizational learning. Academy of Management
Review, 10(4), 803-813.
Fiol, C. M., & O'Connor, E. (2006). Stuff matters: Artifacts, social identity, and
legitimacy in the U.S. medical profession. In A. Rafaeli & M. G. Pratt (Eds.),
Artifacts and organizations: Beyond mere symbolism (pp. 241-258). New York:
Psychology Press.
Fisher, E. S., Goodman, D., Skinner, J., & Bronner, K. (2009). Health care spending,
quality, and outcomes: A Dartmouth Atlas Project topic brief. Hanover, NH:
Dartmouth Institute for Health Policy and Clinical Practice.
Fisher, E. S., Wennberg, D. E., Stukel, T. A., Gottlieb, D. J., Lucas, F. L., & Pinder, E. L.
(2003a). The implications of regional variations in Medicare spending. Part 1:
The content, quality, and accessibility of care. Annals of Internal Medicine, 138,
273-287. doi: 10.7326/0003-4819-138-4-200302180-00006
Fisher, E. S., Wennberg, D. E., Stukel, T. A., Gottlieb, D. J., Lucas, F. L., & Pinder, E. L.
(2003b). The implications of regional variations in Medicare spending. Part 2:
Health outcomes and satisfaction with care. Annals of Internal Medicine, 138,
288-298. doi: 10.7326/0003-4819-138-4-200302180-00007
270
Fiske, S. T., & Taylor, S. E. (2013). Social cognition: From brains to culture (2nd ed.).
Los Angeles: Sage.
Flanagin, A. J. (2002). The elusive benefits of the technological support of knowledge
management. Management Communication Quarterly, 16(2), 242-248. doi:
10.1177/089331802237237
Flanagin, A. J., & Bator, M. (2011). The utility of information and communication
technologies in organizational knowledge management. In H. E. Canary & R. D.
McPhee (Eds.), Communication and organizational knowledge: Contemporary
issues for theory and practice (pp. 173-190). New York: Routledge.
Flanagin, A. J., & Metzger, M. J. (2000). Perceptions of internet information credibility.
Journalism & Mass Communication Quarterly, 77(3), 515-540. doi:
10.1177/107769900007700304
Flanagin, A. J., & Metzger, M. J. (2007). The role of site features, user attributes, and
information verification behaviors on the perceived credibility of web-based
information. New Media & Society, 9(2), 319-342. doi:
10.1177/1461444807075015
Flanagin, A. J., & Metzger, M. J. (2008). Digital media and youth: Unparalleled
opportunity and unprecedented responsibility. In M. J. Metzger & A. J. Flanagin
(Eds.), Digital media, youth, and credibility (pp. 5-28). Cambridge, MA: MIT
Press.
Flanagin, A. J., & Metzger, M. J. (in press). Digital media and perceptions of source
credibility in political communication. In K. Kenski & K. Hall Jamieson (Eds.),
The Oxford handbook of political communication. Oxford: Oxford University
Press.
Fogg, B. J., Marshall, J., Laraki, O., Osipovich, A., Varma, C., Fang, N., et al. (2001).
What makes Web sites credible?: A report on a large quantitative study.
Proceedings of the SIGCHI Conference on Human Factors in Computing
Systems, 61-68.
Fogg, B. J., Soohoo, C., Danielson, D. R., Marable, L., Stanford, J., & Trauber, E. R.
(2003). How do users evaluate the credibility of Web sites? A study with over
2,500 participants. Proceedings of the ACM Conference on Designing for User
Experiences, 1-15.
Friedman, V. J., Lipshitz, R., & Popper, M. (2005). The mystification of organizational
learning. Journal of Management Inquiry, 14(1), 19-30. doi:
10.1177/1056492604273758
Friemel, T. N. (2012). Network dynamics of television use in school classes. Social
Networks, 34, 346-358. doi: 10.1016/j.socnet.2011.08.002
271
Fulk, J., Heino, R., Flanagin, A. J., Monge, P. R., & Bar, F. (2004). A test of the
individual action model for organizational information commons. Organization
Science, 15(5), 569-585.
Galbraith, J. (1974). Organization design: An information processing view. Interfaces,
3(28-36). doi: 10.1287/inte.4.3.28
Galbraith, J. (1977). Organizational design. Reading, MA: Addison-Wesley.
Gaskin, J., Berente, N., Lyytinen, K., & Yoo, Y. (2014). Toward a generalizable
sociomaterial inquiry: A computational approach for zooming in and out of
sociomaterial routines. MIS Quarterly, 38(3), 849-871.
Gherardi, S. (2006). Organizational knowledge: The texture of workplace learning.
Oxford: Blackwell Publishing.
Giddens, A. (1990). The consequences of modernity. Stanford, CA: Stanford University
Press.
Glanz, K., Rimer, B. K., & Viswanath, K. (Eds.). (2008). Health behavior and health
education: Theory, research, and practice (4th ed.). San Francisco: Jossey-Bass.
Glasgow, R. E., & Emmons, K. M. (2007). How can we increase translation of research
into practice? Types of evidence needed. Annual Review of Public Health, 28,
413-433. doi: 10.1146/annurev.publhealth.28.021406.144145
Go, A. S., Mozaffarian, D., Roger, V. L., Benjamin, E. J., Berry, J. D., Borden, W. B., et
al. (2013). Heart disease and stroke statistics--2013 update: A report from the
American Heart Association. Circulation, 127, e6-e245. doi:
10.1161/CIR.0b013e31828124ad
Gondo, M. B., & Amis, J. M. (2013). Variations in practice adoption: The roles of
conscious reflection and discourse. Academy of Management Review, 38(2), 229-
247. doi: 10.5465/amr.2010.0312
Gorman, P. N. (1995). Information needs of physicians. Journal of the American Society
for Information Science, 46(10), 729-736. doi: 10.1002/(SICI)1097-
4571(199512)46:10<729::AID-ASI3>3.0.CO;2-2
Granovetter, M. (1973). The strength of weak ties. American Journal of Sociology, 78(6),
1360-1380.
Granovetter, M. (1985). Economic action and social structure: The problem of
embeddedness. American Journal of Sociology, 91(3), 481-510.
Green, L. W., Ottoson, J. M., García, C., & Hiatt, R. A. (2009). Diffusion theory and
knowledge dissemination, utilization, and integration in public health. Annual
272
Review of Public Health, 30, 151-174. doi:
10.1146/annurev.publhealth.031308.100049
Greenhalgh, T., Robert, G., Bate, P., Macfarlane, F., & Kyriakidou, O. (2005). Diffusion
of innovations in health service organisations: A systematic literature review.
Malden, MA: Blackwell.
Greenwood, R., Oliver, C., Sahlin, K., & Suddaby, R. (2008). Introduction. In R.
Greenwood, C. Oliver, K. Sahlin & R. Suddaby (Eds.), The SAGE handbook of
organizational institutionalism (pp. 1-46). Los Angeles: Sage.
Greenwood, R., Suddaby, R., & Hinings, C. R. (2002). Theorizing change: The role of
professional associations in the transformation of institutionalized fields.
Academy of Management Journal, 45(1), 58-80. doi: 10.2307/3069285
Grol, R. (2001). Improving the quality of medical care: Building bridges among
professonal pride, payer profit, and patient satisfaction. Journal of the American
Medical Association, 286(20), 2578-2585.
Groleau, C., & Demers, C. (2012). Pencils, legos, and guns: A study of artifacts used in
architecture. In P. M. Leonardi, B. A. Nardi & J. Kallinikos (Eds.), Materiality
and organizing: Social interaction in a technological world (pp. 259-286).
Oxford: Oxford University Press.
Groves, R. M., Fowler, F. J., Jr., Couper, M. P., Lepkowski, J. M., Singer, E., &
Tourangeau, R. (2009). Survey methodology (2nd ed.). Hoboken, NJ: John Wiley
& Sons.
Guallar, E., & Laine, C. (2014). Controversy over clinical guidelines: Listen to the
evidence, not the noise. Annals of Internal Medicine, 160(5), 361-362. doi:
10.7326/M14-0112
Gulati, R., Dialdin, D. A., & Wang, L. (2002). Organizational networks. In J. A. C. Baum
(Ed.), The Blackwell companion to organizations (pp. 281-303). Oxford, UK:
Blackwell Publishers.
Haas, M. R., & Hansen, M. T. (2007). Different knowledge, different benefits: Toward a
productivity perspective on knowledge sharing in organizations. Strategic
Management Journal, 28, 1133-1153.
Hannan, M. T., & Freeman, J. (1989). Organizational ecology. Cambridge: Harvard
University Press.
Hansen, M. T. (1999). The search-transfer problem: The role of weak ties in sharing
knowledge across organization subunits. Administrative Science Quarterly, 44(1),
82-111.
273
Hansen, M. T. (2002). Knowledge networks: Explaining effective knowledge sharing in
multiunit companies. Organization Science, 13(3), 232-248.
Hansen, M. T., & Haas, M. R. (2001). Competing for attention in knowledge markets:
Electronic document dissemination in a management consulting company.
Administrative Science Quarterly, 46, 1-28.
Hansen, M. T., Mors, M. L., & Løvås, B. (2005). Knowledge sharing in organizations:
Multiple networks, multiple phases. Academy of Management Journal, 48(5),
776-793. doi: 10.5465/AMJ.2005.18803922
Haunschild, P. R., & Chandler, D. (2008). Institutional-level learning: Learning as a
source of institutional change. In R. Greenwood, C. Oliver, R. Suddaby & K.
Sahlin (Eds.), The SAGE Handbook of Organizational Institutionalism (pp. 624-
649). Thousand Oaks, CA: Sage.
Haveman, H. A., & David, R. J. (2008). Ecologists and institutionalists: Friends or foes?
In R. Greenwood, C. Oliver, R. Suddaby & K. Sahlin (Eds.), The SAGE
Handbook of Organizational Institutionalism (pp. 573-595). Los Angeles: Sage.
Hayes, N. (2011). Information technology and the possibilities for knowledge sharing. In
M. Easterby-Smith & M. A. Lyles (Eds.), Handbook of organizational learning
and knowledge management (2nd ed., pp. 83-104). Chichester, UK: John Wiley &
Sons.
Hilligoss, B., & Rieh, S. Y. (2008). Developing a unifying framework of credibility
assessment: Construct, heuristics, and interaction in context. Information
Processing and Management, 44(1467-1484). doi: 10.1016/j.ipm.2007.10.001
Hovland, C. I., Janis, I. L., & Kelley, H. H. (1953). Communication and persuasion. New
Haven, CT: Yale University Press.
Hsiao, C.-J., Cherry, D. K., Beatty, P. C., & Rechsteiner, E. A. (2010). National
Ambulatory Medical Care Survey: 2007 summary. National Health Statistics
Reports, No. 27. Hyattsville, MD: National Center for Health Statistics, U.S.
Centers for Disease Control and Prevention. .
Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure
analysis: Conventional criteria versus new alternatives. Structural Equation
Modeling, 6, 1-55. doi: 10.1080/10705519909540118
Huber, G. P. (1991). Organizational learning: The contributing processes and the
literatures. Organization Science, 2(1), 88.
Huy, Q. N., Corley, K. G., & Kraatz, M. S. (2014). From support to mutiny: Shifting
legitimacy judgments and emotional reactions impacting the implementation of
radical change. Academy of Management Journal, 57(6), 1650-1680. doi:
10.5465/amj.2012.0074
274
Iacobucci, D., & Wasserman, S. (1990). Social networks with two sets of actors.
Psychometrika, 707-720. doi: 10.1007/BF02294618
Ibarra, H. (1992). Homophily and differential returns: Sex differences in network
structure and access in an advertising firm. Administrative Science Quarterly,
37(3), 422-447. doi: 10.2307/2393451
Ibarra, H. (1997). Paving an alternative route: Gender differences in managerial
networks. Social Psychology Quarterly, 60(1), 91-102. doi: 10.2307/2787014
Ibarra, H., Kilduff, M., & Tsai, W. (2005). Zooming in and out: Connecting individuals
and collectivities at the frontiers of organizational network research. Organization
Science, 16(4), 359-371. doi: 10.1287/orsc.1050.0129
Ingram, P., & Roberts, P. W. (2000). Friendships among competitors in the Sydney hotel
industry. American Journal of Sociology, 106(2), 387-423.
Institute of Medicine. (2011). Clinical practice guidelines we can trust. Washington, DC:
National Academies Press.
Iverson, J. O., & McPhee, R. D. (2002). Knowledge management in communities of
practice: Being true to the communicative character of knowledge. Management
Communication Quarterly, 16(2), 259-266.
Jackson, M. H., & Williamson, J. (2011). Challenges of implementing systems for
knowledge management: Static systems and dynamic practices. In H. E. Canary &
R. D. McPhee (Eds.), Communication and organizational knowledge:
Contemporary issues for theory and practice (pp. 53-68). New York: Routledge.
Janis, I. L. (1972). Victims of groupthink: A psychological study of foreign-policy
decisions and fiascoes. Boston: Houghton Mifflin.
Jensen, R., & Szulanski, G. (2004). Stickiness and the adaptation of organizational
practices in cross-border knowledge. Journal of International Business Studies,
35(6), 508-523.
Jöreskog, K. G. (2004). On chi-squares for the independence model and fit measures in
LISREL. Retrieved from www.ssicentral.com/lisrel/techdocs/ftb.pdf
Kaghan, W. N., & Lounsbury, M. (2006). Artifacts, articulation work, and institutional
residue. In A. Rafaeli & M. G. Pratt (Eds.), Artifacts and organizations: Beyond
mere symbolism (pp. 259-278). New York: Psychology Press.
Kallinikos, J., Leonardi, P. M., & Nardi, B. A. (2012). The challenge of materiality:
Origins, scope, and prospects. In P. M. Leonardi, B. A. Nardi & J. Kallinikos
(Eds.), Materiality and organizing: Social interaction in a technological world
(pp. 3-22). Oxford: Oxford University Press.
275
Kane, G. C., & Alavi, M. (2007). Information technology and organizational learning: An
investigation of exploration and exploitation processes. Organization Science,
18(5), 796-812.
Kaptelinin, V., & Nardi, B. A. (2006). Acting with technology: Activity theory and
interaction design. Cambridge, MA: MIT Press.
Karr-Wisniewski, P., & Lu, Y. (2010). When more is too much: Operationalizing
technology overload and exploring its impact on knowledge worker productivity.
Computers in Human Behavior, 26, 1061-1072. doi: 10.1016/j.chb.2010.03.008
Kautz, K., & Jensen, T. B. (2013). Sociomateriality at the royal court of IS: A jester's
monologue. Information and Organization, 23, 15-27. doi:
10.1016/j.infoandorg.2013.01.001
Keaney, J. F., Curfman, G. D., & Jarcho, J. A. (2014). A pragmatic view of the new
cholesterol treatment guidelines. New England Journal of Medicine, 370(3), 275-
278. doi: 10.1056/NEJMms1314569
Keating, N. L., Ayanian, J. Z., Cleary, P. D., & Marsden, P. V. (2007). Factors affecting
influential discussions among physicians: A social network analysis of a primary
care practice. Journal of General Internal Medicine, 22, 794-798. doi:
10.1007/s11606-007-0190-8
Keegan, B., Gergle, D., & Contractor, N. (2012). Do editors or articles drive
collaboration? Multilevel statistical network analysis of Wikipedia coauthorship.
Proceedings of the 15th ACM Conference on Computer-Supported Cooperative
Work and Social Computing (CSCW). Seattle, WA.
Kilduff, M., & Brass, D. J. (2010). Organizational social network research: Core ideas
and key debates. Academy of Management Annals, 4(1), 317-357. doi:
10.1080/19416520.2010.494827
Kilduff, M., Tsai, W., & Hanke, R. (2006). A paradigm too far? A dynamic stability
reconsideration of the social network research program. Academy of Management
Review, 31(4), 1031-1048. doi: 10.5465/AMR.2006.22528168
Kline, R. B. (2011). Principles and practice of structural equation modeling (3rd ed.).
New York: Guilford Press.
Knoke, D. (1985). The political economies of associations. Research in Political
Sociology, 1, 211-242.
Knorr Cetina, K. (1997). Sociality with objects: Social relations in postsocial knowledge
societies. Theory, Culture & Society, 14(1), 1-30. doi:
10.1177/026327697014004001
276
Ko, D.-G., Kirsch, L. J., & King, W. R. (2005). Antecedents of knowledge transfer from
consultants to clients in enterprise system implementations. MIS Quarterly, 29(1),
59-85.
Kolata, G. (2013, November 17). Risk calculator for cholesterol appears flawed, The New
York Times. Retrieved from www.nytimes.com/2013/11/18/health/risk-calculator-
for-cholesterol-appears-flawed.html?_r=0
Koskinen, J., & Daraganova, G. (2013). Exponential random graph model fundamentals.
In D. Lusher, J. Koskinen & G. Robins (Eds.), Exponential random graph models
for social networks: Theory, methods, and applications (pp. 49-76). Cambridge:
Cambridge University Press.
Koskinen, J., & Snijders, T. A. B. (2013). Simulation, estimation, and goodness of fit. In
D. Lusher, J. Koskinen & G. Robins (Eds.), Exponential random graph models
for social networks (pp. 141-166). Cambridge: Cambridge University Press.
Kostova, T., & Zaheer, S. (1999). Organizational legitimacy under conditions of
complexity: The case of the multinational enterprise. Academy of Management
Review, 24(1), 64-81. doi: 10.5465/AMR.1999.1580441
Krumholz, H. M. (2013, November 12). Three things to know about the new cholesterol
guidelines, New York Times. Retrieved from
http://well.blogs.nytimes.com/2013/11/12/3-things-to-know-about-the-new-
cholesterol-guidelines/?_php=true&_type=blogs&_r=0
Kuhn, T., & Porter, A. J. (2011). Heterogeneity in knowledge and knowing: A social
practice perspective. In H. E. Canary & R. D. McPhee (Eds.), Communication
and organizational knowledge: Contemporary issues for theory and practice (pp.
17-34). New York: Routledge.
Labianca, G., & Brass, D. J. (2006). Exploring the social ledger: Negative relationships
and negative asymmetry in social networks in organizations. Academy of
Management Review, 31(3), 596-614. doi: 10.5465/AMR.2006.21318920
Lammers, J. C., & Barbour, J. B. (2006). An institutional theory of organizational
communication. Communication Theory, 16, 356-377. doi: 10.1111/j.1468-
2885.2006.00274.x
Latour, B. (1992). Where are the missing masses? The sociology of a few mundane
artefacts. In W. E. Bijker & J. Law (Eds.), Shaping technology/building society:
Studies in sociotechnical change (pp. 225-258). Cambridge, MA: MIT Press.
Latour, B. (1994). On technical mediation--philosophy, sociology, genealogy. Common
Knowledge, 3(2), 29-64.
Latour, B. (2005). Reassembling the social: An introduction to actor-network-theory.
Oxford: Oxford University Press.
277
Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation.
Cambridge, UK: Cambridge University Press.
Lavis, J. N., Robertson, D., Woodside, J. M., McLeod, C. B., & Abelson, J. (2003). How
can research organizations more effectively transfer research knowledge to
decision makers? Milbank Quarterly, 81(2), 221-248.
Law, J. (1992). Notes on the theory of the actor-network: Ordering, strategy and
heterogeneity. Systems Practice, 5, 379-393.
Law, J. (2009). Actor network theory and material semiotics. In B. S. Turner (Ed.), The
new Blackwell companion to social theory (pp. 141-158). West Sussex, UK: John
Wiley & Sons.
Lazega, E., Mounier, L., Snijders, T. A. B., & Tubaro, P. (2012). Norms, status and the
dynamics of advice networks: A case study. Social Networks, 34, 323-332. doi:
10.1016/j.socnet.2009.12.001
Lazega, E., & Pattison, P. E. (1999). Multiplexity, generalized exchange and cooperation
in organizations: A case study. Social Networks, 21, 67-90. doi: 10.1016/S0378-
8733(99)00002-7
Lazega, E., & van Duijn, M. (1997). Position in formal structure, personal characteristics
and choices of advisors in a law firm: A logistic regression model for dyadic
network data. Social Networks, 19(375-397). doi: 10.1016/S0378-8733(97)00006-
3
Lee, G. K., & Cole, R. E. (2003). From a firm-based to a community-based model of
knowledge creation: The case of the Linux kernal development. Organization
Science, 14(6), 633-649.
Lehr, J. K., & Rice, R. E. (2002). Organizational measures as a form of knowledge
management: A multitheoretic, communication-based exploration. Journal of the
American Society for Information Science and Technology, 53(12), 1060-1073.
Lenzer, J., Hoffman, J. R., Furberg, C. D., & Ioannidis, J. P. A. (2013). Ensuring the
integrity of clinical practice guidelines: A tool for protecting patients. British
Medical Journal, 347, 5535-5545. doi: 10.1136/bmj.f5535
Leonardi, P. M. (2011). When flexible routines meet flexible technologies: Affordance,
constraint, and the imbrication of human and material agencies. MIS Quarterly,
36(1), 147-167.
Leonardi, P. M. (2012). Materiality, sociomateriality, and socio-technical systems: What
do these terms mean? How are they different? Do we need them? In P. M.
Leonardi, B. A. Nardi & J. Kallinikos (Eds.), Materiality and organizing: Social
interaction in a technological world (pp. 25-48). Oxford: Oxford University
Press.
278
Leonardi, P. M. (2013a). The emergence of materiality within formal organizations. In P.
R. Carlile, D. Nicolini, A. Langley & H. Tsoukas (Eds.), How matter matters:
Objects, artifacts, and materiality in organization studies (pp. 142-170). Oxford:
Oxford University Press.
Leonardi, P. M. (2013b). Theoretical foundations for the study of sociomateriality.
Information and Organization, 23, 59-76. doi: 10.1016/j.infoandorg.2013.02.002
Leonardi, P. M., & Barley, S. R. (2008). Materiality and change: Challenges to building
better theory about technology and organizing. Information and Organization, 18,
159-176. doi: 10.1016/j.infoandorg.2008.03.001
Leonardi, P. M., & Barley, S. R. (2010). What's under construction here? Social action,
materiality, and power in constructivist studies of technology and organizing.
Academy of Management Annals, 4(1), 1-51. doi: 10.1080/19416521003654160
Leonardi, P. M., & Barley, W. C. (2011). Materiality as organizational communication:
Technology, intent, and delegation in the production of meaning. In T. Kuhn
(Ed.), Matters of communication: Political, cultural, and technological challenges
to communication theorizing (pp. 101-122). Cresskill, NJ: Hampton Press.
Leskovec, J., Kleinberg, J., & Faloutsos, C. (2007). Graph evolution: Densification and
shrinking diameters. ACM Transactions on Knowledge Discovery from Data,
1(1), Article 2. doi: 10.1145/1217299.1217301
Levitt, B., & March, J. G. (1988). Organizational learning. Annual Review of Sociology,
14, 319-340.
Lomi, A., Lusher, D., Pattison, P., & Robins, G. (2014). The focused organization of
advice relations: A study in boundary crossing. Organization Science, 25(2), 438-
457. doi: 10.1287/orsc.2013.0850
Lusher, D., Koskinen, J., & Robins, G. (Eds.). (2013). Exponential random graph models
for social networks: Theory, methods, and applications. Cambridge: Cambridge
University Press.
Lusher, D., & Robins, G. (2013a). Example exponential random graph model analysis. In
D. Lusher, J. Koskinen & G. Robins (Eds.), Exponential random graph models
for social networks (pp. 37-48). Cambridge: Cambridge University Press.
Lusher, D., & Robins, G. (2013b). Formation of social network structure. In D. Lusher, J.
Koskinen & G. Robins (Eds.), Exponential random graph models for social
networks: Theory, methods, and applications (pp. 16-28). Cambridge: Cambridge
University Press.
Marsden, P. V. (1990). Network data and measurement. Annual Review of Sociology, 16,
435-463. doi: 10.1146/annurev.so.16.080190.002251
279
Marsden, P. V. (2005). Recent developments in network measurement. In P. J.
Carrington, J. Scott & S. Wasserman (Eds.), Models and methods in social
network analysis (pp. 8-30). Cambridge: Cambridge University Press.
Marsden, P. V. (2011). Survey methods for network data. In J. Scott & P. J. Carrington
(Eds.), The SAGE handbook of social network analysis (pp. 370-388). Los
Angeles: SAGE.
Martin, S. S., & Blumenthal, R. S. (2014). Concepts and controversies: The 2013
American College of Cardiology/American Heart Association risk assessment and
cholesterol treatment guidelines. Annals of Internal Medicine, 160(5), 356-358.
doi: 10.7326/M13-2805
Mayer, K. J., & Sparrowe, R. T. (2013). Integrating theories in AMJ articles. Academy of
Management Journal, 56(4), 917-922. doi: 10.5465/amj.2013.4004
McCroskey, J. C., & Teven, J. J. (1999). Goodwill: A reexamination of the construct and
its measurement. Communication Monographs, 66, 90-103. doi:
10.1080/03637759909376464
McEvily, B., Perrone, V., & Zaheer, A. (2003). Trust as an organizing principle.
Organization Science, 14(1), 91-103. doi: 10.1287/orsc.14.1.91.12814
McEvily, B., Soda, G., & Tortoriello, M. (2014). More formally: Rediscovering the
missing link between formal organization and informal social structure. Academy
of Management Annals, 8(1), 299-345. doi: 10.1080/19416520.2014.885252
McPhee, R. D., Canary, H. E., & Iverson, J. O. (2011). Moving forward with
communicative perspectives on organizational knowledge. In H. E. Canary & R.
D. McPhee (Eds.), Communication and organizational knowledge: Contemporary
issues for theory and practice (pp. 304-313). New York: Routledge.
McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: Homophily in
social networks. Annual Review of Sociology, 27, 415-444.
Mennis, J., & Mason, M. J. (2012). Social and geographic contexts of adolescent
substance use: The moderating effects of age and gender. Social Networks, 34,
150-157. doi: 10.1016/j.socnet.2010.10.003
Metzger, M. J., Flanagin, A. J., & Medders, R. B. (2010). Social and heuristic approaches
to credibility evaluation online. Journal of Communication, 60, 413-439. doi:
10.1111/j.1460-2466.2010.01488.x
Meyer, J. W., & Rowan, B. (1977). Institutionalized organizations: Formal structure as
myth and ceremony. American Journal of Sociology, 83(2), 340-363.
Meyer, J. W., & Scott, J. (Eds.). (1983). Organizational environments: Ritual and
rationality. Beverly Hills, CA: Sage.
280
Miner, A. S., & Mezias, S. J. (1996). Ugly duckling no more: Pasts and futures of
organizational learning research. Organization Science, 7(1), 88-99. doi:
10.1287/orsc.7.1.88
Mitka, M. (2013). Blood pressure, cholesterol guidelines face more delays. Journal of the
American Medical Association, 310(6), 568-569.
Mondak, J. J. (1990). Perceived legitimacy of Supreme Court decisions: Three functions
of source credibility. Political Behavior, 12(4), 363-384. doi:
10.1007/BF00992794
Monge, P., Heiss, B. M., & Margolin, D. B. (2008). Communication network evolution in
organizational communities. Communication Theory, 18, 449-477. doi:
10.1111/j.1468-2885.2008.00330.x
Monge, P. R., & Contractor, N. S. (2003). Theories of communication networks. Oxford:
Oxford University Press.
Morrison, E. W. (2002). Newcomers' relationships: The role of social network ties during
socialization. Academy of Management Journal, 45(6), 1149-1160. doi:
10.2307/3069430
Murphy, A. G., & Eisenberg, E. M. (2011). Coaching to the craft: Understanding
knowledge in health care organizations. In H. E. Canary & R. D. McPhee (Eds.),
Communication and organizational knowledge: Contemporary issues for theory
and practice (pp. 264-284). New York: Routledge.
Mutch, A. (2013). Sociomateriality--Taking the wrong turning? Information and
Organization, 23, 23-40. doi: 10.1016/j.infoandorg.2013.02.001
Nahon, K. (2011). Fuzziness of inclusion/exclusion in networks. International Journal of
Communication, 5, 756-772.
Nebus, J. (2006). Building collegial information networks: A theory of advice network
generation. Academy of Management Review, 31(3), 615-637. doi:
10.5465/AMR.2006.21318921
Nembhard, I. M., Alexander, J. A., Hoff, T. J., & Ramanujam, R. (2009). Why does the
quality of health care continue to lag? Insights from management research.
Academy of Management Perspectives, 24-42. doi: 10.5465/AMP.2009.37008001
Nerkar, A., & Paruchuri, S. (2005). Evolution of R&D capabilities: The role of
knowledge networks within a firm. Management Science, 51(5), 771-785. doi:
10.1287/mnsc.1040.0354
New York Times Editorial Board. (2013, November 18). Cholesterol guidelines under
attack, New York Times. Retrieved from
281
http://www.nytimes.com/2013/11/19/opinion/cholesterol-guidelines-under-
attack.html
Newman, M. E. J. (2010). Networks: An introduction. Oxford: Oxford University Press.
Niekamp, A.-M., Mercken, L. A. G., Hoebe, C. J. P. A., & Dukers-Muijrers, N. H. T. M.
(2013). A sexual affiliation network of swingers, heterosexuals practicing risk
behaviours that potentiate the spread of sexually transmitted infections: A two-
mode approach. Social Networks, 35, 223-236. doi: 10.1016/j.socnet.2013.02.006
Nonaka, I. (1994). A dynamic theory of organizational knowledge creation. Organization
Science, 5(1), 14-37.
Nonaka, I., & Takeuchi, H. (1995). The knowledge-creating company. New York: Oxford
University Press.
Nutley, S. M., Walter, I., & Davies, H. T. O. (2007). Using evidence: How research can
inform public services. Bristol, UK: Policy Press.
Nye, C. D., & Drasgow, F. (2011). Assessing goodness of fit: simple rules of thumb
simply do not work. Organizational Research Methods, 14(3), 548-570. doi:
10.1177/1094428110368562
O'Reilly, C. A. (1980). Individuals and information overload in organizations: Is more
necessarily better? Academy of Management Journal, 23(4), 684-696. doi:
10.2307/255556
Ocasio, W. (2011). Attention to attention. Organization Science, 22(5), 1286-1296. doi:
10.1287/orsc.1100.0602
Ognyanova, K. (2012). Intermedia agenda setting in an era of fragmentation:
Applications of network science in the study of mass communication. (Doctoral
dissertation). Available from ProQuest Dissertations and Theses database.
Oldroyd, J. B., & Morris, S. S. (2012). Catching falling stars: A human resource response
to social capital's detrimental effect of information overload on star employees.
Academy of Management Review, 37(3), 396-418. doi: 10.5465/amr.2010.0403
Orlikowski, W. J. (2002). Knowing in practice: Enacting a collective capability in
distributed organizing. Organization Science, 13(3), 249-273.
Orlikowski, W. J. (2007). Sociomaterial practices: Exploring technology at work.
Organization Studies, 28(9), 1435-1448. doi: 10.1177/0170840607081138
Orlikowski, W. J., & Barley, S. R. (2001). Technology and institutions: What can
research on information technology and research on organizations learn from each
other? MIS Quarterly, 25(2), 145-165.
282
Orlikowski, W. J., & Scott, S. V. (2009). Sociomateriality: Challenging the separation of
technology, work and organization. Academy of Management Annals, 2(1), 433-
474. doi: 10.1080/19416520802211644
Østbye, T., Yarnall, K. S. H., Krause, K. M., Pollak, K. I., Gradison, M., & Michener, J.
L. (2005). Is there time for management of patients with chronic diseases in
primary care? Annals of Family Medicine, 3(3), 209-214. doi: 10.1370/afm.310
Owen-Smith, J., & Powell, W. W. (2008). Networks and institutions. In R. Greenwood,
C. Oliver, K. Sahlin & R. Suddaby (Eds.), The SAGE handbook of organizational
institutionalism (pp. 596-623). Los Angeles: Sage.
Pentland, B. T., & Singh, H. (2012). Materiality: What are the consequences? In P. M.
Leonardi, B. A. Nardi & J. Kallinikos (Eds.), Materiality and organizing: Social
interaction in a technological world (pp. 287-295). Oxford: Oxford University
Press.
Perloff, R. M. (2010). The dynamics of persuasion: Communication and attitudes in the
21st century (4th ed.). New York: Routledge.
Phelps, C., Heidl, R., & Wadhwa, A. (2012). Knowledge, networks, and knowledge
networks: A review and research agenda. Journal of Management, 38(4), 1115-
1166. doi: 10.1177/0149206311432640
Polanyi, M. (1966). The tacit dimension. London: Routledge and Kegan Paul.
Pornpitakpan, C. (2004). The persuasiveness of source credibility: A critical review of
five decades' evidence. Journal of Applied Social Psychology, 34(2), 243-281.
doi: 10.1111/j.1559-1816.2004.tb02547.x
Powell, W. W., & Colyvas, J. A. (2008). Microfoundations of institutional theory. In R.
Greenwood, C. Oliver, K. Sahlin & R. Suddaby (Eds.), The SAGE handbook of
organizational institutionalism (pp. 276-298). Los Angeles: Sage.
Preciado, P., Snijders, T. A. B., Burk, W. J., Stattin, H., & Kerr, M. (2012). Does
proximity matter? Distance dependence in adolescent friendships. Social
Networks, 34, 18-31. doi: 10.1016/j.socnet.2011.01.002
Provan, K. G., Beagles, J. E., Mercken, L., & Leischow, S. J. (2012). Awareness of
evidence-based practices by organizations in a publicly funded smoking cessation
network. Journal of Public Administration Research and Theory. doi:
10.1093/jopart/mus011
Provan, K. G., Fish, A., & Sydow, J. (2007). Interorganizational networks at the network
level: A review of the empirical literature on whole networks. Journal of
Management, 33(3), 479-516. doi: 10.1177/0149206307302554
283
Pustejovsky, J. E., & Spillane, J. P. (2009). Question-order effects in social network
name generators. Social Networks, 31, 221-229.
Rafaeli, A., & Pratt, M. G. (Eds.). (2006). Artifacts and organizations: Beyond mere
symbolism. Mahwah, NJ: Lawrence Erlbaum Associates.
Raman, R., & Bharadwaj, A. (2012). Power differentials and performative deviation
paths in practice transfer: The case of evidence-based medicine. Organization
Science, 23(6), 1593-1621. doi: 10.1287/orsc.1110.0708
Rank, O. N., Robins, G. L., & Pattison, P. E. (2010). Structural logic of
intraorganizational networks. Organization Science, 21(3), 745-764. doi:
10.1287/orsc.1090.0450
Reagans, R., & McEvily, B. (2003). Network structure and knowledge transfer: The
effects of cohesion and range. Administrative Science Quarterly, 48(2), 240-267.
Ren, Y., & Argote, L. (2011). Transactive memory systems 1985-2010: An integrative
framework of key dimensions, antecedents, and consequences. Academy of
Management Annals, 5(1), 189-229. doi: 10.1080/19416520.2011.590300
Rhoads, K. V. L., & Cialdini, R. B. (2002). The business of influence: Principles that lead
to success in commercial settings. In J. P. Dillard & M. Pfau (Eds.), The
persuasion handbook: Developments in theory and practice (pp. 513-542).
Thousand Oaks, CA: Sage.
Ridker, P. M., & Cook, N. R. (2013). Statins: New American guidelines for prevention of
cardiovascular disease. Lancet, 382, 1762-1765. doi: 10.1016/S0140-
6736(13)62388-0
Rieh, S. Y., & Danielson, D. R. (2007). Credibility: A multidisciplinary framework. In B.
Cronin (Ed.), Annual review of information science and technology (Vol. 41, pp.
307-364). Medford, NJ: Information Today.
Rindova, V. P., Pollock, T. G., & Hayward, M. L. A. (2006). Celebrity firms: The social
construction of market popularity. Academy of Management Review, 31(1), 50-71.
doi: 10.5465/AMR.2006.19379624
Ripley, R. M., Snijders, T. A. B., Boda, Z., Vörös, A., & Preciado, P. (2013). Manual for
SIENA version 4.0 (version February 17, 2014). Oxford: University of Oxford,
Department of Statistics; Nuffield College. http://www.stats.ox.ac.uk/siena/.
Roberts, K. H., & O'Reilly, C. A. (1979). Some correlations of communication roles in
organizations. Academy of Management Journal, 22(1), 42-57. doi:
10.2307/255477
Robey, D., Raymond, B., & Anderson, C. (2012). Theorizing information technology as a
material artifact in information systems research. In P. M. Leonardi, B. A. Nardi
284
& J. Kallinikos (Eds.), Materiality and organizing: Social interaction in a
technological world (pp. 217-236). Oxford: Oxford University Press.
Robins, G., & Daraganova, G. (2013). Social selection, dyadic covariates, and geospatial
effects. In D. Lusher, J. Koskinen & G. Robins (Eds.), Exponential random graph
models for social networks: Theory, methods, and applications (pp. 91-101).
Cambridge: Cambridge University Press.
Robins, G., & Lusher, D. (2013). Illustrations: Simulation, estimation, and goodness of
fit. In D. Lusher, J. Koskinen & G. Robins (Eds.), Exponential random graph
models for social networks: Theory, methods, and applications (pp. 167-185).
Cambridge: Cambridge University Press.
Robins, G., Pattison, P., Kalish, Y., & Lusher, D. (2007). An introduction to exponential
random graph (p*) models for social networks. Social Networks, 29(2), 173-191.
doi: 10.1016/j.socnet.2006.08.002
Robins, G., Pattison, P., & Wang, P. (2009). Closure, connectivity and degree
distributions: Exponential random graph (p*) models for directed social networks.
Social Networks, 31, 105-117. doi: 10.1016/j.socnet.2008.10.006
Robins, G., Snijders, T. A. B., Wang, P., Handcock, M., & Pattison, P. (2007). Recent
developments in exponential random graph (p*) models for social networks.
Social Networks, 29, 192-215. doi: 10.1016/j.socnet.2006.08.003
Rogers, E. M. (2003). Diffusion of innovations (5th ed.). New York: Free Press.
Ruef, M., & Scott, W. R. (1998). A multidimensional model of organizational legitimacy:
hospital survival in changing institutional environments. Administrative Science
Quarterly, 43, 877-904.
Rycroft-Malone, J., Seers, K., Titchen, A., Harvey, G., Kitson, A., & McCormack, B.
(2004). What counts as evidence in evidence-based practice? Journal of Advanced
Nursing, 47(1), 81-90. doi: 10.1111/j.1365-2648.2004.03068.x
Sahlin-Andersson, K., & Engwall, L. (Eds.). (2002). The expansion of management
knowledge: Carriers, flows, and sources. Stanford: Stanford University Press.
Sahlin, K., & Wedlin, L. (2008). Circulating ideas: Imitation, translation and editing. In
R. Greenwood, C. Oliver, K. Sahlin & R. Suddaby (Eds.), The SAGE handbook of
organizational institutionalism (pp. 218-242). Los Angeles: Sage.
Saint-Charles, J., & Mongeau, P. (2009). Different relationships for coping with
ambiguity and uncertainty in organizations. Social Networks, 31, 33-39. doi:
10.1016/j.socnet.2008.09.001
Schatzki, T. R., Knorr Cetina, K., & von Savigny, E. (Eds.). (2001). The practice turn in
contemporary theory. London: Routledge.
285
Schuster, D. V., Valente, T. W., Skara, S. N., Wenten, M. R., Unger, J. B., Boley Cruz,
T., et al. (2006). Intermedia processes in the adoption of tobacco control activities
among opinion leaders in California. Communication Theory, 16, 91-117.
Scott, J. (2000). Social network analysis (2nd ed.). Los Angeles: SAGE Publications.
Scott, S. V., & Orlikowski, W. J. (2013). Sociomateriality--taking the wrong turning? A
response to Mutch. Information and Organization, 23, 77-80. doi:
10.1016/j.infoandorg.2013.02.003
Scott, W. R. (2013). Institutions and organizations: Ideas, interests, and identities (4th
ed.). Thousand Oaks, CA: Sage.
Scott, W. R., Reuf, M., Mendel, P. J., & Caronna, C. (2000). Institutional change and
health care organizations: From professional dominance to managed care.
Chicago: University of Chicago Press.
Selznick, P. (1957). Leadership in administration. Evanston, Ill: Row, Peterson.
Shaneyfelt, T. M., & Centor, R. M. (2009). Reassessment of clinical practice guidelines:
Go gently into that good night. Journal of the American Medical Association,
301(8), 868-869.
Shaneyfelt, T. M., Mayo-Smith, M. F., & Rothwangl, J. (1999). Are guidelines following
guidelines? The methodological quality of clinical practice guidelines in the peer-
reviewed medical literature. Journal of the American Medical Association,
281(20), 1900-1905.
Sharratt, M., & Usoro, A. (2003). Understanding knowledge-sharing in online
communities of practice. Electronic Journal on Knowledge Management, 1(2),
187-196.
Shipilov, A. V., Gulati, R., Kilduff, M., Li, S., & Tsai, W. (2014). Relational pluralism
within and between organizations. Academy of Management Journal, 57(2), 449-
459. doi: 10.5465/amj.2013.1145
Shoemaker, P. J. (1982). The perceived legitimacy of deviant political groups: Two
experiments on media effects. Communication Research, 9(2), 249-286. doi:
10.1177/009365082009002004
Shumate, M. (2011). Knowledge management systems and work teams. In H. E. Canary
& R. D. McPhee (Eds.), Communication and organizational knowledge:
Contemporary issues for theory and practice (pp. 191-208). New York:
Routledge.
Shumate, M., & Contractor, N. (2013). Emergence of multidimensional social networks.
In L. L. Putnam & D. K. Mumby (Eds.), The SAGE handbook of organizational
286
communication: Advances in theory, research, and methods (pp. 449-474).
Thousand Oaks, CA: SAGE.
Shumate, M., & Palazzolo, E. T. (2010). Exponential random graph (p*) models as a
method for social network analysis in communication research. Communication
Methods and Measures, 4(4), 341-371. doi: 10.1080/19312458.2010.527869
Siegel, J. M. (1990). Stressful life events and use of physician services among the elderly:
The moderating role of pet ownership. Journal of Personality and Social
Psychology, 58(6), 1081-1086.
Siles, I., & Boczkowski, P. (2012). At the intersection of content and materiality: A texto-
material perspective on the use of media technologies. Communication Theory,
22, 227-249. doi: 10.1111/j.1468-2885.2012.01408.x
Simon, H. A. (1955). A behavioral model of rational choice. Quarterly Journal of
Economics, 69, 99-118. doi: 10.2307/1884852
Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological
Review, 63, 129-138. doi: 10.1037/h0042769
Simon, H. A. (1957). Models of man: Social and rational. New York: Wiley.
Simon, H. A. (1979). Rational decision making in business organizations. American
Economic Review, 69(4), 493-513.
Sirovich, B. E., Gottlieb, D. J., Welch, H. G., & Fisher, E. S. (2006). Regional variations
in health care intensity and physician perceptions of quality of care. Annals of
Internal Medicine, 144, 641-649. doi: 10.7326/0003-4819-144-9-200605020-
00007
Snijders, T. A. B., Pattison, P., Robins, G., & Handcock, M. (2006). New specifications
for exponential random graph models. Sociological Methodology, 36, 99-153. doi:
10.1111/j.1467-9531.2006.00176.x
Soltis, S. M., Agneessens, F., Sasovova, Z., & Labianca, G. (2013). A social network
perspective on turnover intentions: The role of distributive justice and social
support. Human Resource Management, 52(4), 561-584. doi: 10.1002/hrm.21542
Song, L., & Chang, T.-Y. (2012). Do resources of network members help in help
seeking? Social capital and information search. Social Networks, 34, 658-669.
doi: 10.1016/j.socnet.2012.08.002
Soucek, R., & Moser, K. (2010). Coping with information overload in email
communication: Evaluation of a training intervention. Computers in Human
Behavior, 26, 1458-1466. doi: 10.1016/j.chb.2010.04.024
287
Spender, J.-C. (1996). Making knowledge the basis of a dynamic theory of the firm.
Strategic Management Journal, 17, 45-62.
Starfield, B. (1998). Primary care: Balancing health needs, services, and technology.
New York: Oxford University Press.
Starfield, B., Shi, L., & Macinko, J. (2005). Contribution of primary care to health
systems and health. Milbank Quarterly, 83(3), 457-502. doi: 10.1111/j.1468-
0009.2005.00409.x
Stephens, K. K. (2012). Theorizing beyond informational message characteristics: Social
considerations, materiality, and a contemporary view of communication overload.
Paper presented at the International Communication Association Conference,
Phoenix, AZ.
Stephens, K. K., & Davis, J. (2009). The social influences on electronic multitasking in
organizational meetings. Management Communication Quarterly, 23(1), 63-83.
doi: 10.1177/0893318909335417
Still, M. C., & Strang, D. (2009). Who does an elite organization emulate? Administrative
Science Quarterly, 54, 58-89.
Stone, N. J., Robinson, J. G., Lichtenstein, A. H., Bairey Merz, C. N., Blum, C. B., Eckel,
R. H., et al. (2013). 2013 ACC/AHA guideline on the treatment of blood
cholesterol to reduce atherosclerotic cardiovascular risk in adults: A report of the
American College of Cardiology/American Heart Association Task Force on
Practice Guidelines. Circulation, 00, 000-000. doi:
10.1161/01.cir.0000437738.63853.7a
Stone, N. J., Robinson, J. G., Lichtenstein, A. H., Bairey Merz, C. N., Blum, C. B., Eckel,
R. H., et al. (2014). 2013 ACC/AHA guideline on the treatment of blood
cholesterol to reduce atherosclerotic cardiovascular risk in adults: A report of the
American College of Cardiology/American Heart Association Task Force on
Practice Guidelines. Journal of the American College of Cardiology, 63(25),
2889-2934. doi: 10.1016/j.jacc.2013.11.002
Strang, D., & Macy, M. W. (2001). In search of excellence: Fads, success stories, and
adaptive emulation. American Journal of Sociology, 107(1), 147-182.
Strang, D., & Soule, S. A. (1998). Diffusion in organizations and social movements:
From hybrid corn to poison pills. Annual Review of Sociology, 24, 265-290.
Su, C., & Contractor, N. (2011). A multidimensional network approach to studying team
members' information seeking from human and digital knowledge sources in
consulting firms. Journal of the American Society for Information Science and
Technology, 62(7), 1257-1275.
288
Su, C., Huang, M., & Contractor, N. (2010). Understanding the structures, antecedents
and outcomes of organisational learning and knowledge transfer: A multi-
theoretical and multilevel network analysis. European Journal of International
Management, 4(6), 576-601.
Suchman, M. C. (1995). Managing legitimacy: strategic and institutional approaches.
Academy of Management Review, 20, 571-610.
Suddaby, R., Elsbach, K. D., Greenwood, R., Meyer, J. W., & Zilber, T. B. (2010).
Organizations and their institutional environments--bringing meaning, values, and
culture back in. Academy of Management Journal, 53(6), 1234-1240.
Sundar, S. S. (2008). The MAIN model: A heuristic approach to understanding
technology effects on credibility. In M. J. Metzger & A. J. Flanagin (Eds.),
Digital media, youth, and credibility (pp. 73-100). Cambridge, MA: MIT Press.
Sussman, S., Valente, T. W., Rohrbach, L. A., Skara, S., & Pentz, M. A. (2006).
Translation in the health professions: Coverting science into action. Evaluation
and the Health Professions, 29(1), 7-32.
Szulanski, G. (1996). Exploring internal stickiness: Impediments to the transfer of best
practice within the firm. Strategic Management Journal, 17, 27-43.
Tabachnick, B. G., & Fidell, L. S. (2007). Using multivariate statistics (5th ed.). Boston:
Pearson.
Taylor, J. R. (2001). Toward a theory of imbrication and organizational communication.
American Journal of Semiotics, 17(2), 269-298.
Taylor, J. R., & Van Every, E. J. (2000). The emergent organization: Communication as
site and surface. Mahwah, NJ: Lawrence Erlbaum Associates.
Timmermans, S., & Mauck, A. (2005). The promises and pitfalls of evidence-based
medicine. Health Affairs, 24(1), 18-28. doi: 10.1377/hlthaff.24.1.18
Tolbert, P. S., & Zucker, L. G. (1983). Institutional sources of change in the formal
structure of organizations: The diffusion of civil service reform, 1880-1935.
Administrative Science Quarterly, 28, 22-39.
Tortoriello, M., Reagans, R., & McEvily, B. (2012). Bridging the knowledge gap: The
influence of strong ties, network cohesion, and network range on the transfer of
knowledge between organizational units. Organization Science, 23(4), 1024-
1039. doi: 10.1287/orsc.1110.0688
Tost, L. P. (2011). An integrative model of legitimacy judgments. Academy of
Management Review, 36(4), 686-710. doi: 10.5465/amr.2010.0227
289
Tsai, W. (2002). Social structure of "coopetition" within a multiunit organization:
Coordination, competition, and intraorganizational knowledge sharing.
Organization Science, 13(2), 179-190. doi: 10.1287/orsc.13.2.179.536
Tsoukas, H. (1996). The firm as a distributed knowledge system: A constructionist
approach. Strategic Management Journal, 17, 11-25.
Tsoukas, H. (2011). How should we understand tacit knowledge? A phenomenological
view. In M. Easterby-Smith & M. A. Lyles (Eds.), Handbook of organizational
learning and knowledge management (2nd ed., pp. 453-476). Chichester, UK:
John Wiley & Sons.
Tsoukas, H., & Vladimirou, E. (2001). What is organizational knowledge? Journal of
Management Studies, 38(7), 973-993. doi: 10.1111/1467-6486.00268
Tucker, A. L., Nembhard, I. M., & Edmondson, A. C. (2007). Implementing new
practices: An empirical study of organizational learning in hospital intensive care
units. Management Science, 53(6), 894-907. doi: 10.1287/mnsc.1060.0692
Tushman, M., & Nadler, D. (1978). Information processing as an integrating concept in
organizational design. Academy of Management Review, 3, 613-624.
Tyler, T. R. (2006). Psychological perspectives on legitimacy and legitimation. Annual
Review of Psychology, 57, 375-400. doi:
10.1146/annurev.psych.57.102904.190038
Uzzi, B. (1996). The soures and consequences of embeddedness for the economic
performance of organizations: The network effect. American Sociological Review,
61(4), 674-698. doi: 10.2307/2096399
Uzzi, B. (1997). Social structure and competition in interfirm networks: The paradox of
embeddedness. Administrative Science Quarterly, 42(1), 35-67. doi:
10.2307/2393808
Valente, T. W. (2010). Social networks and health: Models, methods, and applications.
New York: Oxford University Press.
Valente, T. W. (2012). Network interventions. Science, 337(49), 49-53. doi:
10.1126/science.1217330
van Wijk, R., Jansen, J. J. P., & Lyles, M. A. (2008). Inter- and intra-organizational
knowledge transfer: A meta-analytic review and assessment of its antecedents and
consequences. Journal of Management Studies, 45(4), 830-853.
Van Wijk, R., Van Den Bosch, F. A. J., & Volberda, H. W. (2011). Organizing
knowledge in social, alliance, and organizational networks. In M. Easterby-Smith
& M. A. Lyles (Eds.), Handbook of organizational learning and knowledge
management (2nd ed., pp. 477-504). Chichester, UK: Wiley.
290
Van Zandt, T. (2004). Overload in a network of targeted communication. RAND Journal
of Economics, 35(3), 542-560. doi: 10.2307/1593707
Veblen, T. (1899/2007). The theory of the leisure class. Oxford: Oxford University Press.
Vehovar, V., Manfreda, K. L., Koren, G., & Hlebec, V. (2008). Measuring ego-centered
social networks on the web: Questionnaire design issues. Social Networks, 30,
213-222. doi: 10.1016/j.socnet.2008.03.002
Vergne, J.-P. (2011). Toward a new measure of organizational legitimacy: Method,
validation, and illustration. Organizational Research Methods, 14(3), 484-502.
doi: 10.1177/1094428109359811
von Hippel, E. (1994). "Sticky information" and the locus of problem solving:
Implications for innovation. Management Science, 40(4), 429-439.
Vygotsky, L. S. (1978). Mind in society: The development of higher psychological
processes. Cambridge, UK: Cambridge University Press.
Wang, P. (2013). Exponential random graph model extensions: Models for multiple
networks and bipartite networks. In D. Lusher, J. Koskinen & G. Robins (Eds.),
Exponential random graph models for social networks: Theory, methods, and
applications (pp. 115-129). Cambridge: Cambridge University Press.
Wang, P., Pattison, P., & Robins, G. (2013). Exponential random graph model
specifications for bipartite networks--A dependence hierarchy. Social Networks,
35(2), 211-222. doi: 10.1016/j.socnet.2011.12.004
Wang, P., Robins, G., Pattison, P., & Koskinen, J. (2014). MPNet User Manual.
Retrieved from
http://sna.unimelb.edu.au/__data/assets/pdf_file/0004/1185745/MPNetManual.pdf
Wang, P., Robins, G., Pattison, P., & Lazega, E. (2013). Exponential random graph
models for multilevel networks. Social Networks, 35, 96-115. doi:
10.1016/j.socnet.2013.01.004
Wang, P., Robins, G., Pattison, P., Lazega, E., & Jourda, M.-T. (2013). Social selection
models for multilevel networks (MelNet Technical Report). Retrieved from
MelNet website:
http://sna.unimelb.edu.au/__data/assets/file/0006/787848/SSMM.pdf
Wang, P., Sharpe, K., Robins, G. L., & Pattison, P. E. (2009). Exponential random graph
(p*) models for affiliation networks. Social Networks, 31, 12-25. doi:
10.1016/j.socnet.2008.08.002
Washington, M., & Zajac, E. J. (2005). Status evolution and competition: Theory and
evidence. Academy of Management Journal, 48(2), 282-296. doi:
10.5465/AMJ.2005.16928408
291
Wasko, M. M., & Faraj, S. (2005). Why should I share? Examining social capital and
knowledge contribution in electronic networks of practice. MIS Quarterly, 29(1),
35-57.
Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications.
Cambridge: Cambridge University Press.
Wasserman, S., & Iacobucci, D. (1991). Statistical modelling of one-mode and two-mode
networks: simultaneous analysis of graphs and bipartite graphs. British Journal of
Mathematical and Statistical Psychology, 13-43. doi: 10.1111/j.2044-
8317.1991.tb00949.x
Wathen, C. N., & Burkell, J. (2002). Believe it or not: Factors influencing credibility on
the Web. Journal of the American Society for Information Science and
Technology, 53(2), 134-144. doi: 10.1002/asi.10016
Watzlawick, P., Beavin, J. H., & Jackson, D. D. (1967). Pragmatics of human
communication: A study of interactional patterns, pathologies, and paradoxes.
New York: W. W. Norton & Company.
Weber, M. S., Chung, C. J., & Park, H. W. (2012). Editors' introduction to the special
issue: The hyperlinked society: Understanding the changing nature of
communication in online environments. Journal of Computer-Mediated
Communication, 17(2), 117-119. doi: 10.1111/j.1083-6101.2011.01570.x
Wellstein, B., & Kieser, A. (2011). Trading "best practices"--a good practice? Industrial
and Corporate Change, 20(3), 683-719. doi: 10.1093/icc/dtr011
Wenger, E. (1998). Communities of practice: Learning, meaning, and identity.
Cambridge, UK: Cambridge University Press.
Westerwick, A. (2012). Effects of sponsorship, Web site design, and Google ranking on
the credibility of online information. Journal of Computer-Mediated
Communication, 18(194-211). doi: 10.1111/jcc4.12006
Westphal, J. D., Gulati, R., & Shortell, S. M. (1997). Customization or conformity? An
institutional and network perspective on the content and consequences of TQM
adoption. Administrative Science Quarterly, 42(2), 366-394. doi:
10.2307/2393924
Whelan, E., & Teigland, R. (2013). Transactive memory systems as a collective filter for
mitigating information overload in digitally enabled organizational groups.
Information and Organization, 23, 177-197. doi: I 0.1 016/j-infoandorg.20
13.06.001
Whyte, J., & Harty, C. (2012). Socio-material practices of design coordination: Objects
as plastic and partisan. In P. M. Leonardi, B. A. Nardi & J. Kallinikos (Eds.),
292
Materiality and organizing: Social interaction in a technological world (pp. 196-
216). Oxford: Oxford University Press.
Wood, L., Giles-Corti, B., & Bulsara, M. (2005). The pet connection: Pets as a conduit
for social capital? Social Science and Medicine, 61, 1159-1173. doi:
10.1016/j.socscimed.2005.01.017
Yarnall, K. S. H., Pollak, K. I., Østbye, T., Krause, K. M., & Michener, J. L. (2003).
Primary care: Is there enough time for prevention? American Journal of Public
Health, 93(4), 635-641.
Yousefi-Nooraie, R., Dobbins, M., & Marin, A. (2014). Social and organizational factors
affecting implementation of evidence-informed practice in a public health
department in Ontario: A network modelling approach. Implementation Science,
9(29), 1-14. doi: 10.1186/1748-5908-9-29
Zahra, S. A., & George, G. (2002). Absorptive capacity: A review, reconceptualization,
and extension. Academy of Management Review, 27(2), 185-203.
Zappa, P. (2011). The network structure of knowledge sharing among physicians. Quality
& Quantity, 45, 1109-1126. doi: 10.1007/s11135-011-9494-1
Zimmerman, M. A., & Zeitz, G. J. (2002). Beyond survival: Achieving new venture
growth by building legitimacy. Academy of Management Review, 27(3), 414-431.
doi: 10.2307/4134387
Zipf, G. (1949). Human behavior and the principle of least effort: An introduction to
human ecology. New York: Addison-Wesley.
293
APPENDIX A: SURVEY INSTRUMENT
Q1
For the first set of questions, think about how you may have learned about the new cholesterol guidelines issued in November 2013 by
the American College of Cardiology (ACC) and the American Heart Association (AHA). For example, you may have asked a
colleague for their opinion about the new guidelines, or you may have received an e-mail from your department chief about the
guidelines, or you may have happened to see a summary of the guidelines online.
Have you learned about the 2013 ACC/AHA cholesterol guidelines from any [Alpha Medical Group] primary care physician (PCP)
colleagues? The method by which you and the other person communicated (face-to-face, phone, e-mail, etc.) does not matter.
Yes
No
>> If No Is Selected, Then Skip To Q4 [Skip Logic]
Q2
Please list in the spaces below the names of all of your [Alpha Medical Group] PCP colleagues from whom you have learned about
the 2013 ACC/AHA cholesterol guidelines. Select the person’s clinic location and name from the alphabetized drop-down menus
provided. List as many or as few names as appropriate, then scroll to the bottom of the page to move to the next question. If you wish
to list more names, there is an option to do so at the bottom of the page.
This question may be time-consuming, but your answers are important for helping us understand and improve how best-practice
guidelines travel through an organization. We are very grateful for your time.
[Alpha Medical Group] PCP #1
Clinic Location
Physician Name
294
[Alpha Medical Group] PCP #2
Clinic Location
Physician Name
[Alpha Medical Group] PCP #3
Clinic Location
Physician Name
[Alpha Medical Group] PCP #4
Clinic Location
Physician Name
[Alpha Medical Group] PCP #5
Clinic Location
Physician Name
► Please select one:
I have more names to provide
I am finished with this question
>> If I am finished with this question Is Selected, Then Skip To Q4 [Skip Logic]
295
Q3
List as many or as few names as appropriate, then scroll to the bottom of the page to click on the "Next" button.
[Alpha Medical Group] PCP #6
Clinic Location
Physician Name
[Alpha Medical Group] PCP #7
Clinic Location
Physician Name
[Alpha Medical Group] PCP #8
Clinic Location
Physician Name
[Alpha Medical Group] PCP #9
Clinic Location
Physician Name
[Alpha Medical Group] PCP #10
Clinic Location
Physician Name
296
Q4
Have you learned about the 2013 ACC/AHA cholesterol guidelines from any sources other than people, such as UpToDate, journal
articles, newspapers, etc.? The medium (website, e-mail, paper, etc.) by which you accessed this source does not matter.
Yes
No
>> If No Is Selected, Then Skip To Q6 [Skip Logic]
Q5
Please list in the spaces below all of those sources other than people, from which you have learned about the 2013 ACC/AHA
cholesterol guidelines. These may include UpToDate; the official guidelines articles published in the journals JACC and Circulation;
mass media such as newspapers, TV, and radio; websites; or other sources. The medium (website, e-mail, paper, etc.) by which you
accessed this source does not matter.
Please provide the specific name of the source (for example, "JACC" or "Boston Globe") rather than a category (for example,
"journal" or "newspaper"). List as many or as few sources as appropriate, then scroll to the bottom of the page to move to the next
question.
Source Name (please type in)
Please provide the specific name, rather than a category
Source #1
Source #2
Source #3
Source #4
Source #5
Source #6
297
Source Name (please type in)
Please provide the specific name, rather than a category
Source #7
Source #8
Source #9
Source #10
Q6
In the next set of questions, please provide your opinion about the ACC, the AHA, and any physicians or other sources of information
about the 2013 cholesterol guidelines that you named in the previous questions. You will be asked to specify to what extent you agree
or disagree with a statement about each source of information.
Your answers to these questions, as well as all other questions in the survey, will be completely confidential. All survey results will be
anonymized, so that your name and the names of any colleagues you provide in your answers will be replaced with identification
numbers. Only the study external principal investigator (Amanda Beacom) will see and have access to the raw survey results before
they are anonymized. Any survey results presented or published in reports or research articles will be in the form of anonymized,
aggregated results that cannot be associated with any individual.
This source provides expert information:
[The following answer choices are populated from the respondent’s answers to the network questions Q2, Q3, and Q5.]
Strongly
Agree
Moderately
Agree
Mildly
Agree
Neither
Agree nor
Disagree
Mildly
Disagree
Moderately
Disagree
Strongly
Disagree
ACC (American College of Cardiology)
AHA (American Heart Association)
[PCP #1]
298
Strongly
Agree
Moderately
Agree
Mildly
Agree
Neither
Agree nor
Disagree
Mildly
Disagree
Moderately
Disagree
Strongly
Disagree
[PCP #2]
[PCP #3]
[PCP #4]
[PCP #5]
[PCP #6]
[PCP #7]
[PCP #8]
[PCP #9]
[PCP #10]
[Source #1]
[Source #2]
[Source #3]
[Source #4]
[Source #5]
[Source #6]
[Source #7]
[Source #8]
[Source #9]
[Source #10]
299
Q7
This source is well-informed:
[same answer choices as Q6]
Q8
This source conveys competence:
[same answer choices as Q6]
Q9
This source is trustworthy:
[same answer choices as Q6]
Q10
This source is unbiased:
[same answer choices as Q6]
Q11
This source is reliable:
[same answer choices as Q6]
Q12
People think this source is valuable to our society:
[The following answer choices are populated from the respondent’s answers to the network question Q5.]
Strongly
Agree
Moderately
Agree
Mildly
Agree
Neither
Agree nor
Disagree
Mildly
Disagree
Moderately
Disagree
Strongly
Disagree
ACC (American College of Cardiology)
300
Strongly
Agree
Moderately
Agree
Mildly
Agree
Neither
Agree nor
Disagree
Mildly
Disagree
Moderately
Disagree
Strongly
Disagree
AHA (American Heart Association)
[Source #1]
[Source #2]
[Source #3]
[Source #4]
[Source #5]
[Source #6]
[Source #7]
[Source #8]
[Source #9]
[Source #10]
Q13
People in the general public would approve of this source if asked their opinion:
[same answer choices as Q12]
Q14
This source is viewed by industry experts as one of the top organizations in its field:
[same answer choices as Q12]
Q15
This source always follows laws and regulations:
[same answer choices as Q12]
301
Q16
People would be shocked to hear that this source violated any professional codes of conduct:
[same answer choices as Q12]
Q17
This source's organizational leaders believe in “playing by the rules” and following accepted operating guidelines:
[same answer choices as Q12]
Q18
Now, to help us understand how information overload may influence how physicians learn about best-practice guidelines, please
answer the following questions from this validated scale. Think about all of the information (best practices as well as all other types of
information) you need to do your clinical work. In a typical week in the previous month, how often:
Always Frequently Sometimes Rarely Never
Do you feel you have to engage
in too much communication (for
example, too many phone calls,
meetings, memos, letters, face-
to-face conversations, emails,
text messages, etc.)?
Do you receive more information
than you can process?
Do you receive more information
than you need in order to do your
job effectively?
302
Always Frequently Sometimes Rarely Never
Do you have more discussion
than you wish to about confusing
or ambiguous information?
Does your communicating with
others involve too many
decisions?
Do you receive information that
requires you to make too many
decisions?
Do you receive information that
needs too many explanations in
order for it to be useful to you?
Q19
In the space below, please provide any additional information you feel is important for us to better understand how you communicate
with your colleagues or use other sources to learn about clinical best practices. For example, you may wish to elaborate on one of your
above answers, or suggest important questions that we omitted but should have asked about this topic. If you don't have any comments
to share, you may skip this question and click on the "Next" button at the bottom of the page.
303
Q20
Lastly, in order to control for any potential differences between full- and part-time clinicians, please answer the following question.
Out of 10 total clinical sessions in a week, how many do you typically work?
1 2 3 4 5 6 7 8 9 10
304
APPENDIX B: CONFIRMATORY FACTOR ANALYSIS OF
COMMUNICATION LOAD
Table B1. Polychoric Correlations, Means, and Standard Deviations for Communication
Load
Item 2 Item 3Item 4Item 5Item 6 Item 7
Item 2 1.000
Item 3 0.691 1.000
Item 4 0.456 0.454 1.000
Item 5 0.613 0.643 0.704 1.000
Item 6 0.650 0.537 0.637 0.765 1.000
Item 7 0.656 0.611 0.614 0.702 0.857 1.000
M 4.02 3.85 3.19 3.05 3.19 3.35
SD 0.764 0.875 0.993 0.929 0.888 0.882
Item 2 (Estimate how often…) You receive more information than you can process
Item 3 You receive more information than you need in order to do your job effectively
Item 4 You have more discussion than you wish to about confusing or ambiguous information
Item 5 Your communicating with others involves too many decisions
Item 6 You receive information that requires you to make too many decisions
Item 7 You receive information that needs too many explanations in order for it to be useful
Note. Items were scored on a 1-5 scale, with higher scores indicating greater communication load.
Table B2. Global Goodness of Fit Indices for Communication Load
Criteria Model 1 (df = 14) Model 2 (df = 9) Final Model (df = 8)
SB scaled χ
2
37.93423.602 16.649
p > 0.05 0.001 0.005 0.034
SRMSR ≤ 0.08 0.068 0.057 0.038
CFI > 0.95 0.978 0.983 0.990
RMSEA ≤ 0.06 0.110 0.095 0.080
90% CI narrow 0.069 – 0.152 0.041 – 0.150 0.000 – 0.140
p > 0.05 0.015 0.038 0.134
Note. Boldface type indicates indices suggestive of satisfactory fit. SB = Satorra-Bentler; SRMSR
= standardized root mean square residual; CFI = Bentler’s comparative fit index; RMSEA = root
mean square error of approximation.
305
Communication
Load
Item 2
Item 3
Item 4
Item 5
Item 7
Item 6
1.00
(0.70)
1.03***
(0.72)
0.96***
(0.67)
1.24***
(0.87)
1.32***
(0.92)
1.27***
(0.89)
0.49
0.51
0.55
0.48
0.25
0.15
0.20
0.22
Figure B1. Unstandardized (and standardized) parameter estimates and measurement
errors for communication load.
*** p < .01
306
APPENDIX C: CONFIRMATORY FACTOR ANALYSIS OF LEGITIMACY
Table C1. Polychoric Correlations, Means, and Standard Deviations for Legitimacy
Item 1 Item 2 Item 3 Item 4 Item 6 Item 6
Item 1 1.000
Item 2 0.723 1.000
Item 3 0.532 0.591 1.000
Item 4 0.436 0.385 0.454 1.000
Item 5 0.459 0.500 0.423 0.606 1.000
Item 6 0.420 0.380 0.417 0.673 0.642 1.000
M 6.09 5.99 6.23 5.43 5.90 5.60
SD 0.997 1.053 1.055 1.298 1.233 1.242
Item 1 People think this source is valuable to our society
Item 2 People in the general public would approve of this source if asked their opinion
Item 3 This source is viewed by industry experts as one of the top organizations in its field
Item 4 This source always follows laws and regulations
Item 5
People would be shocked to hear that this source violated any professional codes of
conduct
Item 6
This source’s organizational leaders believe in ‘playing by the rules’ and following
accepted operating guidelines
Note. Items 1-3 are indicators of moral legitimacy, and Items 4-6 are indicators of regulatory
legitimacy. Items were scored on a 1-7 scale, with higher scores indicating greater legitimacy.
Table C2. Global Goodness of Fit Indices for Legitimacy
Criteria Model 1 (df = 8) Model 2 (df = 7) Final Model (df = 6)
SB scaled χ
2
25.04317.813 13.632
p > 0.05 0.002 0.013 0.034
SRMSR ≤ 0.08 0.035 0.027 0.021
CFI > 0.95 0.992 0.995 0.996
RMSEA ≤ 0.06 0.056 0.051 0.052
90% CI narrow 0.028 – 0.085 0.020 – 0.083 0.018 – 0.086
p > 0.05 0.196 0.378 0.473
Note. Boldface type indicates indices suggestive of satisfactory fit. SB = Satorra-Bentler; SRMSR
= standardized root mean square residual; CFI = Bentler’s comparative fit index; RMSEA = root
mean square error of approximation.
307
Regulatory
Legitimacy
Item 1
Item 2
Item 3
Item 4
Item 6
Item 5
1.00
(0.74)
1.01***
(0.75)
1.03***
(0.77)
1.00
(0.75)
1.12***
(0.84)
1.01***
(0.75)
0.55
0.45
0.41
0.44
0.45
0.30
0.43
0.15
Moral
Legitimacy
0.55
0.41
0.11
Figure C1. Unstandardized (and standardized) parameter estimates and measurement
errors for legitimacy.
*** p < .01
308
APPENDIX D: CONFIRMATORY FACTOR ANALYSIS OF CREDIBILITY
Table D1. Polychoric Correlations, Means, and Standard Deviations for Credibility
Item 1 Item 2 Item 3 Item 4 Item 6 Item 6
Item 1 1.000
Item 2 0.716 1.000
Item 3 0.686 0.904 1.000
Item 4 0.618 0.788 0.837 1.000
Item 5 0.413 0.510 0.573 0.704 1.000
Item 6 0.603 0.772 0.827 0.912 0.755 1.000
M 6.06 6.42 6.39 6.37 5.71 6.24
SD 1.085 0.833 0.898 0.914 1.299 0.967
Item 1 This source provides expert information
Item 2 This source is well-informed
Item 3 This source conveys competence
Item 4 This source is trustworthy
Item 5 This source is unbiased
Item 6 This source is reliable
Note. Items 1-3 are indicators of expertise, and Items 4-6 are indicators of trustworthiness. Items
were scored on a 1-7 scale, with higher scores indicating greater credibility.
Table D2. Global Goodness of Fit Indices for Credibility
Criteria Model 1 (df = 8)
SB scaled χ
2
29.281
p > 0.05 0.000
SRMSR ≤ 0.08 0.028
CFI > 0.95 0.996
RMSEA ≤ 0.06 0.065
90% CI narrow 0.041 – 0.090
p > 0.05 0.150
Note. Boldface type indicates indices suggestive of satisfactory fit. SB = Satorra-Bentler; SRMSR
= standardized root mean square residual; CFI = Bentler’s comparative fit index; RMSEA = root
mean square error of approximation.
309
Trustworthiness
Item 1
Item 2
Item 3
Item 4
Item 6
Item 5
1.00
(0.73)
1.34***
(0.98)
1.28***
(0.93)
1.00
(0.96)
0.76***
(0.73)
1.01***
(0.97)
0.92
0.47
0.14
0.04
0.08
0.47
0.06
Expertise 0.53
0.60
Figure D1. Unstandardized (and standardized) parameter estimates and measurement
errors for credibility.
*** p < .01
310
APPENDIX E: KNOWLEDGE SOURCE ARTIFACTS USED BY PHYSICIANS
Table E1. Artifacts Used By Physicians
No.
1
Artifact Description Media
2
41 UpToDate Clinical decision support 0
25 New England Journal of Medicine Medical journal/publication 0
22 New York Times Newspaper 1
17 Journal Watch Medical journal/publication 0
16 [Local newspaper] Newspaper 1
15 Journal of the American Medical Association Medical journal/publication 0
14 Journal of the ACC Medical journal/publication 0
12 Annals of Internal Medicine Medical journal/publication 0
6 ACC/AHA guidelines document Guidelines document 0
6 Circulation (AHA journal) Medical journal/publication 0
5 Journal(s) (not otherwise specified) Medical journal/publication 0
5 National Public Radio Radio 1
4 ACC/ACC website Professional organization 0
4 AHA website Professional organization 0
3 CME course (not otherwise specified) Educational meeting 0
3 E-mail (not otherwise specified) E-mail 0
3 Medscape Medical journal/publication 0
3 Prescriber's Letter Medical journal/publication 0
3 Wall Street Journal Newspaper 1
3 American College of Physicians Professional organization 0
2 Internal Medicine News Medical journal/publication 0
2 Cleveland Clinic Journal of Medicine Medical journal/publication 0
2 Physician’s First Watch Medical journal/publication 0
2 Newspaper (not otherwise specified) Newspaper 1
2 TV (not otherwise specified) TV 1
2 Google/Googled it Website 1
2 Mass media (not otherwise specified) Mass media 1
2 Cleveland Clinic/website Website 0
1 Diabetic conference Educational meeting 0
1
[Local medical school] sponsored internal
medicine review course
Educational meeting 0
311
1
[State] American Academy of Family
Physicians Spring Refresher Meeting
Educational meeting 0
1 Internal medicine update course Educational meeting 0
1 Grand Rounds at [local hospital] Educational meeting 0
1 [Alpha Medical Group] e-mail E-mail 0
1 American Family Physician Medical journal/publication 0
1 American Journal of Medicine Medical journal/publication 0
1 Oakstone Internal Medicine Review Medical journal/publication 0
1
American College of Physicians online Medical
Knowledge Self-Assessment Program update
board review course materials
Medical journal/publication 0
1 Mayo Clinic Journal of Medicine Medical journal/publication 0
1 ACP Internist Medical journal/publication 0
1 [State medical society] Journal Medical journal/publication 0
1 iPhone app (not otherwise specified) Mobile app 0
1 iPhone app released by ACC/AHA Mobile app 0
1 American Medical Association Professional organization 0
1
Professional organization website (not
otherwise specified)
Professional organization 0
1 U.S. Preventive Services Task Force Professional organization 0
1
Physicians Committee for Responsible
Medicine
Professional organization 0
1 Radio (not otherwise specified) Radio 1
1 CNN TV 1
1 TV news (not otherwise specified) TV 1
1 Dr Oz TV 1
1 Website (not otherwise specified) Website 0
1 WebMD Website 1
1 Yahoo News Website 1
1 Mass media websites (not otherwise specified) Website 1
1 Joslin Diabetes Center website Website 0
Note. Names enclosed in brackets and italicized were anonymized.
1
Indicates number of nominations the artifact received from physicians.
2
Indicates the type of media artifact, mass (1) or niche (0).
312
APPENDIX F: POORLY-FITTING GOODNESS OF FIT STATISTICS FOR
NETWORK MODELS
Table F1. Poorly-Fitting Goodness of Fit Statistics for Bipartite Model
Parameter Observed Mean SD t-ratio
IsolatesXB 010.7103.382 -3.170
SD DegreeXA 1.4937 1.2419 0.125 2.011
Table F2. Poorly-Fitting Goodness of Fit Statistics for Multilevel Model
Parameter Observed Mean SD t-ratio
Version 1
IsolatesXA 3219.0465.335 2.428
Mass Media_X4CycleB2 15 3.387 4.467 2.600
SD DegreeXA 1.4937 1.320 0.086 2.015
Version 2
IsolatesXA 3220.0175.369 2.232
IsolatesXB 011.4412.634 -4.343
Mass Media_XEdgeB 61 104.134 19.497 -2.212
SD DegreeXA 1.4937 1.2802 0.104 2.062
Abstract (if available)
Abstract
This study investigates how best-practice knowledge is communicated in an organizational network. Using sociomateriality as an ontological foundation, I conceptualized knowledge communication as occurring in an organizational network in which the actors are both people and artifacts. I drew on propositions from organizational learning theory and institutional theory to investigate how three forms of sociomaterial agency—the communication load of knowledge consumers, and the legitimacy and credibility of knowledge sources—influenced the structure of the knowledge network. I tested these hypothesized influences using data on the communication of new cholesterol treatment guidelines within an organizational network of primary care physicians and knowledge artifacts. Results from the estimation of recently-developed exponential random graph models for multilevel networks indicated that the legitimacy of the organizational authors of best-practice knowledge exerted a significant influence on network structure, whereas communication load had no effect. The results also offered qualified support for the premise that in many empirical contexts, multilevel sociomaterial networks hold the potential to explain more variance in the structure of knowledge communication than unipartite or bipartite networks alone.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
The coevolution of multimodal, multiplex, and multilevel organizational networks in development communities
PDF
Building social Legoland through collaborative crowdsourcing: marginality, functional diversity, and team success
PDF
The evolution of multilevel organizational networks in an online gaming community
PDF
Ecology and network evolution in online innovation contest crowdsourcing
PDF
Competing logics and organizational failure: the structuring of California prison health care
PDF
The evolution of knowledge creation online: what is driving the dynamics of Wikipedia networks
PDF
Knowledge, motivation, and organizational influences in aligning learning to business performance results: a study of a promising practice
PDF
A multitheoretical multilevel explication of crowd-enabled organizations: exploration/exploitation, social capital, signaling, and homophily as determinants of associative mechanisms in donation-...
PDF
Contagious: social norms about health in work group networks
PDF
Crisis and stasis: understanding change in online communities
PDF
Network influences in health initiatives: multimedia games for youth in Peru
PDF
Mobile self-tracking for health: validating predictors, effects, mediator, moderator, and social influence
PDF
Informing the exam room: understanding the process of translating evidence-based medical research into clinical care
PDF
Creating a community of practice: serving student veterans in higher education
PDF
Organizational mimicry in American social movement communities: an analysis of form communication effects on the evolution of crisis pregnancy centers, 1989-2009
PDF
The formation and influence of online health social networks on social support, self-tracking behavior and weight loss outcomes
PDF
On practical network optimization: convergence, finite buffers, and load balancing
PDF
Relationship formation and information sharing to promote risky health behavior on social media
PDF
The evolution of multidimensional and multilevel networks in online crowdsourcing
PDF
Intermedia agenda setting in an era of fragmentation: applications of network science in the study of mass communication
Asset Metadata
Creator
Beacom, Amanda M.
(author)
Core Title
Communicating organizational knowledge in a sociomaterial network: the influences of communication load, legitimacy, and credibility on health care best-practice communication
School
Annenberg School for Communication
Degree
Doctor of Philosophy
Degree Program
Communication
Publication Date
03/15/2016
Defense Date
01/28/2016
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
best practice,communication load,communication network,credibility,exponential random graph model,health,legitimacy,multilevel,OAI-PMH Harvest,organization,sociomateriality
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Riley, Patricia (
committee chair
), Monge, Peter (
committee member
), Valente, Thomas (
committee member
)
Creator Email
abeacom@gmail.com,abeacom@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c40-220030
Unique identifier
UC11278044
Identifier
etd-BeacomAman-4197.pdf (filename),usctheses-c40-220030 (legacy record id)
Legacy Identifier
etd-BeacomAman-4197.pdf
Dmrecord
220030
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Beacom, Amanda M.
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
best practice
communication load
communication network
credibility
exponential random graph model
health
legitimacy
multilevel
organization
sociomateriality