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
/
The effects of marketing communication on consumers' choice behavior: the case of pharmaceutical industry
(USC Thesis Other)
The effects of marketing communication on consumers' choice behavior: the case of pharmaceutical industry
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
THE EFFECTS OF MARKETING COMMUNICATION ON CONSUMERS’
CHOICE BEHAVIOR:
THE CASE OF PHARMACEUTICAL INDUSTRY
by
Ramkumar Janakiraman
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirement for the Degree
DOCTOR OF PHILOSOPHY
(BUSINESS ADMINISTRATION)
August 2006
Copyright 2006 Ramkumar Janakiraman
ii
ACKNOWLEDGEMENTS
This dissertation could not have been completed without the support of many
people who are gratefully acknowledged here.
I am grateful to Shantanu Dutta, Catarina Sismeiro, Rakesh Niraj, Fred Zufryden
and Geert Ridder who served on my dissertation committee. I further thank Phil
Stern for providing the data.
I greatly appreciate the support of my parents, who have always encouraged me in
my goals and for always believing in me. Finally, I cannot thank enough my wife
Rishika for her constant support, and words of encouragement. Special thanks to
her for being patient and putting up with all the “Later”, “How about next
weekend” as I pondered my way through the doctoral study.
iii
TABLE OF CONTENTS
ACKNOWLEDGEMENTS iii
LIST OF TABLES iv
ABSTRACT v
CHAPTER 1: INTRODUCTION 7
Overview 7
Background 10
Proposed Approach 16
CHAPTER 2: MODEL DEVELOPMENT 18
Learning, Searching and Thinking 19
Drug Choice Models 27
Heterogeneity and Estimation 32
CHAPTER 3: DATA AND MARKET DESCRIPTION 35
CHAPTER 4: RESULTS 42
Model Comparison 42
Physician Persistence 46
Persistence Predictors 47
Prescription Choice 50
CHAPTER 5: DISCUSSION 54
CHAPTER 6: CONCLUSION 60
BIBLIOGRAPHY 66
iv
LIST OF TABLES
Table 1: Descriptive Statistics of Sample............................................................ 37
Table 2: Marketing Communications Summary Statistics ................................. 38
Table 3: Descriptive Statistics of Physician Characteristics................................ 39
Table 4: Variable Description.............................................................................. 40
Table 5: Comparison of Models .......................................................................... 43
Table 6: Estimation Results: Persistence Predictors............................................ 45
Table 7 Estimation Results: Drug Choice Model for the Non-Persistent State... 45
Table 8: Estimation Results: Drug Choice Model for the Persistent State.......... 46
Table 9: Short-Run Effects of Marketing Communications on Drug Choice ..... 56
v
ABSTRACT
There is a growing interest in modeling the demand for prescription drugs and
understanding physicians’ response to marketing communication. Measuring the
response to marketing communication without accounting for persistence (and
heterogeneity) can significantly bias the estimates of physician response and lead
to ineffective budget allocations. Though studies in the medical literature have
suggested that physician decision making is often inertial, it is yet unclear
whether persistence is an important phenomenon in physician drug choice.
Capitalizing on a rich panel dataset of physician prescriptions and of competitive
marketing activity, I investigate physician persistence in drug choice. I propose
that physicians might exhibit persistence to minimize on search and learning
costs. For each new patient visit, I then jointly model the likelihood of physician
persistence and the physician’s drug choice decision. Unlike Bayesian learning
models, the proposed individual-level, heterogeneous and dynamic two-state
model does not assume all physicians are constantly engaged in costly
information search and processing. I explore the role of marketing communication
on persistence and drug choice, and I investigate whether physicians who exhibit
persistence respond differently to three forms of marketing communication: one-
vi
to-one meetings (detailing), outside the office meetings, and symposium
meetings.
The results show significant levels of persistence in drug choice. Physicians do
not frequently change state and they tend to be either persistent or non-persistent.
Non-persistent physicians appear to be responsive to detailing and symposium
meetings, whereas persistent physicians seem to be responsive only to symposium
meetings. Outside the office events, such as golf or lunch, have no effect on
physician choice. Finally, I find that (1) detailing and symposiums can have long-
lasting effects, (2) older physicians and those who work in smaller practices are
more likely to be persistent, and (3) the more a physician is willing to receive
sales force representatives, the higher the likelihood of being persistent. Finally, I
discuss alternatives to improve marketing communication allocations from my
rich set of results.
7
CHAPTER 1: INTRODUCTION
Overview
Recent medical reports have established that a significant number of physicians
do not adhere to clinical guidelines. These guidelines are systematically
developed to assist practitioners and patients in their decisions of appropriate
health care for specific clinical circumstances. A study by American Medical
Association (Cabana et al. 1999) reports that, “…a surprising percentage of
doctors are not following guidelines that could help them treat patients better
because they don't have enough information, time, or readiness to change”. A
more recent study by the University of Pittsburgh Medical Center (2002) indicates
also that many critical care doctors do not apply new life-saving findings in their
intensive care units, though doctors could be aware of such new findings
published in leading medical journals.
The phenomenon of non-adherence to practice guidelines by physicians is well
documented in the literature (Philips et al. 2001, Mottur-Pilson et al. 2001,
Alvanzo et al. 2003). Previous studies have suggested that physicians might not
change previous practice due to inertia or lack of motivation to change. If inertia
and motivational explanations play an important role in physicians’ decision
making, physicians’ drug choice is also likely affected and current drug choices
8
could structurally depend on previously prescribed drugs. Such structural
dependence of sequential choices is known in the marketing literature as
persistence or structural state dependence (Seetharaman 2004) and it has been
widely documented to impact optimal marketing allocation decisions and the
measurement of marketing response for packaged goods. In contrast, for
prescription drug choice, it is yet unclear how persistence impacts the marketing
activities of pharmaceuticals, or even if it is present at all.
In this paper, I investigate physicians’ persistence in prescription choice behavior,
how it interacts with the marketing activities of pharmaceutical companies, and its
implications for the measurement of marketing long-term effects. If significant,
physicians’ persistence in prescription choice would have important implications
for the marketing of pharmaceutical drugs. Pharmaceutical firms target physicians
with various means of marketing communication and other promotional efforts to
influence their prescription behavior. In the United Kingdom, some
pharmaceutical firms spend up to $20,000 per general physician on promoting
their brands (Reich 2005). In the United States, more than $11 billion is spent
each year in various promotional tools, $5 billion of which goes to sales
representatives, with about $8000 to $13,000 spent per year on each physician
(Wazana 2000). By providing better estimates of the effects of marketing
9
activity, the study of physicians’ persistence in prescription choice could improve
sales force management and the allocation of marketing efforts. Optimal decisions
of how much to invest in customer acquisition versus retention and the
measurement of the response to marketing efforts by pharmaceutical companies
will depend on physician persistence and on whether such behavior is adequately
accounted for. Finally, persistence in prescription behavior will affect also
competitive market structure and the optimal response to competitive marketing
activities.
My objectives are to (1) study the determinants of persistence in drug choice,
while controlling for the impact of promotional activity on prescribing behavior,
(2) segment physicians based on their dynamic persistence profile, and (3)
investigate whether the response to promotional activities of persistent physicians
differs from that of non-persistent physicians. I develop a two-state dynamic
model of physicians’ drug choice rooted in decision making theory. I then
estimate and test my proposed model on a rich panel data set of physician
prescription choices and competitive promotional activity. These data comprise
information on different forms of marketing communication including detailing,
outside the office meetings, and symposiums. My modeling results provide a
10
number of findings regarding persistence in prescription behavior and I discuss
the resulting managerial implications for pharmaceutical companies.
Background
Persistence in individual-specific choices over time (also called state dependence)
is said to occur when the individual’s prior purchase experience with specific
brands influences future brand purchase propensity. This structural state
dependence can be positive or negative in which cases it is called inertia (Jeuland
1979) and variety seeking (McAlister 1982), respectively. The presence of inertia
motivates marketers’ employment of promotional schemes, such as free sampling,
in the hope of “hooking” households to the brand for the long-term. The presence
of variety seeking motivates, for example, the lengthening of product lines.
The widespread access to scanner panel data in the past 20 years has allowed a
thorough study of choice persistence in the context of packaged goods (for a
review see Seetharaman 2004 and Keane 1997). The consensus that has emerged
in the literature is that, even if unobserved heterogeneity is adequately controlled
for, households’ brand choices are indeed persistent (Keane 1997, Abramson et. al
2000) and that ignoring one or more sources of state dependence produces
suboptimal inputs for managerial decision making (e.g., it understates marketing-
11
mix elasticities and underestimates the total incremental impact of a sales
promotion).
Though state dependence has been widely studied in the context of packaged
goods, and though there is a growing interest in modeling the demand for
prescription drugs in marketing and economics, it is yet unclear if persistence is
also present in physician choice behavior. However, few studies in the economics
literature have documented the effect of habit persistence in physician decision
making. Hellerstein (1998) studies physician choice of branded versus generic
drugs. The author finds some evidence of habit persistence in the prescription
behavior of physicians, even after controlling for unobserved characteristics of
physicians. Coscelli (2000) studies the drug switching for the Italian anti-ulcer
market and concludes that physicians are not indifferent across brands and that
they exhibit habit persistence. Though these studies represent a first look at
physician inertia in physician choice, both Coscelli (2000) and Hellerstein (1998)
do not study, nor do they control for, the effect of marketing efforts. As a result,
they cannot determine whether persistent doctors and non-persistent doctors
respond differently to marketing communications, nor can they determine how
state dependence interacts with marketing activities. In addition, these studies do
not control for therapy continuation or therapy change to returning patients.
12
Hence, it is difficult to conclude whether structural state dependence is the actual
force driving their results.
In the marketing literature, previous pharmaceutical research has focused broadly
on two aspects of physicians’ prescription behavior: physicians’ learning about
the quality of drug (Narayanan et al. 2004, Mukherji et al. 2004), and physicians’
response to marketing efforts (Parsons and Abeele 1981, Lilien et al. 1981, Gönül
et al. 2001, Manchanda and Chintagunta 2004, Manchanda et al 2004, Mizik and
Jacobson 2004). Learning models are based on Bayesian updating: at the end of
each time period, physicians learn and update the experience characteristic
1
of a
drug based on the patient’s consumption experience, the physician’s clinical
experience, and the advertising exposure (Currie and Park 2002, Coscelli and
Shum 2004, Mukherji et al. 2003, Narayanan et al, 2004). These models do not
model persistence and implicitly assume that a physician always tries to achieve
an apparently effortless fit between the specific condition of a patient and the
characteristics of the drugs prescribed.
Early studies of physicians’ response to marketing efforts in the marketing
literature (e.g., Parsons and Abeele 1981, Lilien et al. 1981) analyzed aggregate
1
Experience characteristics refer to unobservable product characteristics. Experience goods such
as prescription drugs are differentiated by observable product characteristics and unobservable
experience characteristics.
13
sales-territory level data to investigate the effects of sales force activity on sales.
Though the effects of previous prescription levels are often allowed to influence
current levels of prescription activity (through lagged terms), heterogeneity at the
doctor level is not accounted for. This could lead to the overestimation of state
dependence which is often referred to as spurious state dependence (Heckman
1981).
More recently three studies have provided significant insights on longitudinal
dynamics of prescription decisions, while studying the effects of pharmaceutical
promotional activities and controlling for physician heterogeneity. Making use of
a dynamic fixed effects distributed lag regression model, Mizik and Jacobson
(2004) empirically assess the effect of detailing and sampling on physician
prescribing behavior for three drugs. These authors model the monthly dynamics
of prescriptions and estimate the decay structure of the promotional effects over
time. Manchanda et al. (2004) and Manchanda and Chintagunta (2004) analyze
the impact of sales calls on physician prescription behavior using physician-level
data and accounting for physician-level heterogeneity.
These studies provide valuable results on the influence of promotional activities
on prescriptions. However, the nature of the data used by these authors does not
14
allow them to model structural persistence. First, the authors use time-aggregated
data instead of actual choice data. Mizik and Jacobson (2004) and Manchanda et
al. (2004) work with monthly data, Manchanda and Chintagunta (2004) model the
number of quarterly prescriptions. The three studies use either previous period
prescription (Manchanda and Chintagunta 2004; Manchanda et al. 2004) or a
series of lagged prescription levels for the particular drug to capture persistence.
Second, the authors use standard “new prescription” data. These data exclude all
refills given to a patient though each “new prescription” does not necessarily
correspond to a “new” and unique patient: once the initial refills are over, a
patient continuing a therapy will return to the physician and often will receive a
new prescription that is then reported in the “new prescription” variable.
Structural state-dependence implies that drug choice for a patient at visit t
structurally depends on the drug chosen for a different patient seen at visit t–1. As
a result, traditional monthly- or quarterly-level data that do not distinguish new
versus continuing patients will not be able to capture structural persistence.
Finally, these previous studies do not take into account the actions of competing
firms. Whereas Manchanda and Chintagunta (2004) and Manchanda et al (2004)
restrict their analysis to one single drug, Mizik and Jacobson (2004) study the
impact of detailing on physician prescribing behavior for three different drugs of a
15
single manufacturer. None of these studies includes competing detailing or other
competing marketing activities.
Access to disaggregate-level data on drug choice to new patients with competitive
information is essential to correctly estimate persistence levels and their impact
on marketing communication. A previous study that analyzes drug choice instead
of time-aggregated prescription data, and that investigates the effects of
pharmaceutical marketing communications, is the work of Gönül et al. (2001).
These authors provide the first exploratory study in marketing to investigate
whether and how pricing and promotional activities by pharmaceutical companies
(specifically detailing visits and the dispensing of free samples) influence
prescription choice behavior. The authors find that physicians are characterized
by fairly limited price sensitivity, detailing and samples have mostly informative
effects on physicians, and physicians with relatively large numbers of Medicare or
health maintenance organization patients are less influenced by promotion than
the remaining physicians. Though the authors use a multinomial logit choice
model to analyze a comprehensive panel of physicians, the data used in the study
do not include information on every patient seen by a physician; instead,
physicians were asked to fill out survey sheets for a typical week of the month. As
16
a result, the study cannot shed light on physician’s longitudinal drug choice
behavior nor on structural state dependence.
Proposed Approach
I exploit my unique and rich panel data of physician prescription choices and
competitive promotional activity for new patients to model physician persistence
while taking into account the competitive marketing efforts of the various firms. I
develop a two-state dynamic model of physician drug choice rooted in decision
making theory, and I investigate physicians’ persistence in prescription choice
behavior. The proposed joint model of physician persistence and drug choice
allows us to study the determinants of persistence in drug choice, while
controlling for the impact of promotional activity on prescribing behavior (I
observe different forms of marketing communication including detailing, outside
the office meetings, and symposiums). I further segment physicians based on their
dynamic persistence profile, and investigate whether the response to promotional
activities of persistent physicians differs from that of non-persistent physicians.
My model results indicate significant levels of persistence in drug choice with
about 66% of physicians classified as persistent (physicians do not frequently
change states and they tend to be persistent or non-persistent across all of their
17
prescription occasions). In addition, non-persistent physicians appear to be
responsive to detailing and symposium meetings, whereas persistent physicians
seem to be responsive only to symposium meetings. Outside the office events,
such as golf or lunch, seem to have no effect on physicians, irrespective of their
persistence profile. Finally, I find that (1) detailing and symposiums have long-
lasting effects, (2) older physicians and those who work in smaller practices are
more likely to exhibit persistence in drug choice, and (3) that the more a physician
is willing to receive sales force representatives of all competing companies, the
higher the likelihood of being persistent.
These results have important implications for pharmaceutical firms. The model
estimates provide a rich set of results that are useful for improving marketing
communications allocation decisions. For example, even though persistent
physicians will not be influenced by detailing, these physicians can still be
reached through the use of symposium meetings. Hence, these results are highly
encouraging for pharmaceutical firms because they suggest that they can reach the
different types of physician with effective tools and that the effects of some of
these actions can be long-lasting.
18
Chapter 2: MODEL DEVELOPMENT
In this section, I present the proposed joint model of physician persistence and
drug choice. The model has two components: a dynamic two-state model of
persistence and a drug choice model that will depend on the persistence profile of
the physician at each patient visit. I assume that physicians are either in a
“persist” or a “non-persist” state at each patient consultation. In addition if
physicians are in the “persist” state for a given patient visit t, the physician choice
will be affected by the choice made at patient visit t–1. Consequently, I expect
that physician drug choice will be influenced by different factors when in a
“persist” versus a “non-persist” mode.
Next I will present the theories and findings from the literature in support of
persistence in physician drug choice. Based on these theories and findings, I
propose a set of covariates for the two-state model of persistence and for drug
choice model. I then discuss the importance of accounting for unobserved
physician heterogeneity in a model of physician persistence and present the
adopted formulation of physician heterogeneity.
19
Learning, Searching, and Thinking
Previous research on Bayesian learning applied to pharmaceuticals implicitly
assumed that, in all prescription occasions, physicians try to achieve the best fit
between the specific condition of a patient and the characteristics of the drug
prescribed (Currie and Park 2002, Coscelli and Shum 2004, Mukherji et al. 2003,
Narayanan et al. 2004). It is in this apparently costless process of achieving the
best fit that physicians learn about the experience characteristic of the drug.
From a physician’s point of view, providing the best prescription and adhering to
best practice guidelines involves significant learning, search, and thinking costs
(Smith 1996; Stern and Trajtenberg 1998). In order to prescribe the best possible
drug to a new patient the physician will have to (1) accumulate medical
knowledge about the various drugs and treatments (potential interactions and side
effects, efficacy, etc.), (2) gather information about the particular patient (looking
at the patient’s health history, finding out the patient’s allergic conditions,
whether the patient is taking other medications, etc.), and (3) match the medical
knowledge to the individual needs of a patient (Smith 1996; Gorman 1995). Laine
and Weinberg (1999) report, for example, that accumulating medical knowledge
and matching it to the individual needs are some of the most difficult challenges
physicians face. Though physicians can rely on medical literature such as
20
journals, review articles and clinical guidelines, these sources are not usually
relied upon in daily practice (Avorn et al. 1982). The low signal to noise ratio, the
high pace of information change in the field of biomedical science, and the sheer
volume of readings are often mentioned as reasons for the lack of practical
usefulness of these sources (Wyatt 1991, Pauker et al. 1976, Smith 1996; Laine
and Weinberg 1999). Consequently, most physicians find getting adequate
information from the literature to be a major problem (Williamson et al. 1989),
tend to feel overwhelmed by new scientific information, and are not good at
finding
new information (Smith 1996).
I differ from Bayesian learning approaches by proposing that physicians might
exhibit persistence in drug choice in order to minimize on search, learning and
thinking costs (Klemperer 1987; Shugan 1980) so that the physician will treat
most patients as an “average” patient (Stern and Trajtenberg 1998). I assume that
there are two possible physician states—persistent and non-persistent—and that,
at each prescription occasion t, physician i has a certain likelihood of being
persistent (given the physician’s search and learning costs). I also assume that
physicians could switch between the two states over time. I do not force any a
priori segmentation of physicians into persistent versus non-persistent modes, nor
do I require physicians to be of one state or the other across all prescription
21
occasions. (This more flexible framework will allow us to test whether persistence
in prescribing behavior is more of a cross-sectional or more of a longitudinal
market characteristic.) I then model the physician’s likelihood of being in a
persistent state, () Persist
it
Pr , as a binary logit such that (for physician i = 1,…,N
and appointment t = 1,…,T
i
where T
i
is the number of patients seen by physician
i):
()
) exp( 1
) exp(
Pr
it
it
it
S
S
Persist
+
= .
(2.1)
The term S
it
represents physician’s i attractiveness of being persistent at
prescription occasion t by reflecting the motivation and the learning, search, and
thinking costs of each physician in each prescription occasion. I draw on previous
research from the medical literature to determine potential factors that might
predict whether physicians are or not in a persistent state (i.e., whether
physicians’ previous prescription determines the following prescription decision).
I propose practice size, professional experience, and the number of detailing calls
the physician accepts from all competing pharmaceutical firms as valuable
predictors of physician persistence. In addition, I use demographic physician
characteristics to control for observed heterogeneity in physician persistence
likelihood.
22
Cumulative Detailing Across Competing Drugs As an alternative to traditional
sources of medical information (journals, review articles, clinical guidelines, etc.),
physicians could rely on activities sponsored by pharmaceutical companies to get
informed on new treatments and dosages, new approved uses for existing medicines,
new ways to monitor patients, drug interactions and contraindications. Obtaining
information from drug representatives in detailing meetings requires minimal
effort and can be a valuable source of information for a busy doctor. For example,
Shaughnessy and Slawson (1996) argue that although the primary goal of drug
representatives is to promote a product, active approach on the part of physicians
can transform the detailing meeting into a useful and accurate source of
information. The authors suggest that physicians can put drug representatives to
work for them by making them check for new information about their drug that is
both relevant and valid. Marketing activities of pharmaceutical companies could
then be considered as a mechanism for speeding up the adoption of new and better
treatments, which leads to better patient outcomes (Spilker 2002). In addition, though
physicians are equivocal in their beliefs that sales representatives could provide
unbiased information on alternative drugs, previous studies do report that
physicians believe sales representatives provide accurate information about the
drugs they represent (Caudill et al. 1996; Hodges 1995; Strang et al. 1996).
The medical literature provides also evidence against the use of detailing as a
source of information. Pharmaceutical sales representatives are neither experts, nor
23
necessarily trustworthy; information might be biased toward the promoted drug and is
unlikely to be objective, or even accurate (Connelly at al. 1990). Attribution theory
would then predict that with low source credibility the arguments in a sales
representative message will be discounted (Eagley and Chaiken 1975; Dholakia and
Sternthal 1977) and that sales representatives will have little influence on physicians.
Physicians do view sales representatives with skepticism (Lichstein, Turner,and
O’Brien 1992; McKinley at al. 1990) and rated them twelfth in usefulness, out of
fifteen potential information sources (Peay and Peay 1990).
I hypothesize that regardless of the value of the actual information obtained in
detailing meetings, those physicians who have the time to meet with
pharmaceutical sales representatives (from all competing drugs) are also more
likely to have the time to use whatever new information they gather (in the
meetings or outside the meetings) in their drug choice decisions. In addition,
physicians who are more willing to have detailing meetings are also more likely
to trust the information provided, and hence apply it in medical practice. What
makes detailing meetings such a good signal of lower time constraints and interest
in the information provided is that detailing provides few other benefits other than
listening to drug-related information. Other events sponsored by pharmaceutical
24
companies, like symposiums and out-of-the-office meetings for lunch or golf,
provide additional benefits to physicians, like entertainment and networking.
2
Professional Colleagues For busy and time constrained physicians, professional
colleagues could also play an important a role in keeping knowledge current. As a
result, physicians who work in bigger practices have an easier access to
information that could be used in daily practice because they have an easier
access to other physicians (Laine and Weinberg 1999). The effect of social
contagion and opinion leaders in the industry has been well documented in the
literature (e.g. Bauer and Wortzel 1966; Coleman, Menzel and Katz 1966; Lilien
et al. 1981; Van den Bulte and Lilien 2001). In addition, bigger practices are more
likely to exhibit considerable patient heterogeneity: in bigger practices it will be
more likely to find patients farther away from the “average patient” simply due to
the size of the patient pool. If considerable patient heterogeneity is present, it will
be more difficult to keep simple prescribing rules that would minimize on search
costs and still keep adequate patient welfare. Hence, I hypothesize that physicians
from bigger practices are less likely to be in a persistent state.
2
An alternative reason for physicians to meet pharmaceutical sales representatives suggested in
the literature is to receive free-samples they can then distribute to patients (Medical Marketing and
Media 2000; Wolf 1998). In the context of my empirical application that alternative explanation is
not a valid one as physicians receive no samples in their detailing meetings. In case samples are
present, it is possible to include an interaction between samples and detailing meetings to capture
the possible counter effect of samples on persistence.
25
Professional Experience Older physicians are usually thought of having
superior clinical abilities due to tacit knowledge and skills accumulated during the
years. The literature in medicine reports results that suggest the opposite.
Choudhry et al. (2005) review empirical studies evaluating the relationship
between clinical experience and performance and find that older physicians
possess less factual knowledge, are less likely to adhere to appropriate standards
of care, and may also have poorer patient outcomes. Older physicians seem also
less likely to adopt newly proven therapies (Freiman 1985, Hlatky et al. 1988) and
may be less receptive to new standards of care (Young et al. 1987). Based on
these findings, I hypothesize that older physicians are also more likely to be in a
persistent state because they are more adverse to change and have higher learning,
search, and thinking costs. I also hypothesize that professional experience will
interact with detailing meetings: older physicians who accept to see sales
representatives from any drug company are more likely to be using those
meetings to extract information relevant for clinical practice. Older physicians
who meet sales representatives will also be more experienced in extracting
information sought (Shaughnessy and Slawson 1996) and hence are less likely to
be persistent.
26
Gender I include information on whether the physician was male or female in
order to account for differences across physicians in their persistence likelihood. I
do not have specific hypotheses for the direction or magnitude of gender effects.
Though previous research in marketing has shown systematic differences in how
males and females process and are affected by persuasive messages (e.g., Meyers-
Levy 1988; Meyers-Levy and Sternthal 1991), previous research has provided
little evidence on the direction of the effect of gender on choice persistence.
In sum, I specify S
it
from Equation 2.1 as follows:
i i
it i it it
Gender ze PracticeSi
Experience Experience CumulDet CumulDet
5 4
3 2 1 0 it
S
ρ ρ
ρ ρ ρ ρ
+ +
+ × + + =
,
(2.2)
where, CumulDet
it
are the cumulative detailing calls that a physician i received
from all competing pharmaceutical drugs until time t, Experience
i
is the
professional experience of physician i, PracticeSize
i
is the size of the practice
physician i belongs to (measured as number of doctors), and Gender
i
is a dummy
variable that takes the value of one if physician i is male, and zero otherwise.
This model formulation has two important features. First, the likelihood that a
physician will be in a persistent state varies not only across physicians but also
across time. Hence, physicians can be in or out of a persistent state in different
prescription occasions. Second, I leverage on the important distinction between
27
detailing and other forms of marketing communication to model the probability of
a physician being in a persistent or in a non-persistent state. Unlike other forms of
marketing communication (e.g., print advertising, outdoor displays, etc.)
physicians cannot simply be “exposed” to detailing. If driving in the city we
happen to see advertising on billboards it is not because we have accepted or
chosen to see them, it is usually because they are there. Physicians, on the other
hand, have to accept to be seen by pharmaceutical sales representatives in order
for a detailing meeting to occur. Even though modeling the full physician decision
is beyond the scope of this paper (and would require data not available to the
authors), I am able to account for this physician component by allowing
physicians’ previous detailing meetings to serve as a signal of how available and
interested they are in having such meetings (and hence how likely they are to
accept subsequent detailing encounters).
Drug Choice Models
Following previous research, I apply a multinomial logit to model physicians’
drug choice (Gönül et al. 2001). Also consistent with previous research, I assume
that the marketing actions of pharmaceutical companies might have an impact on
physician decision making. Though previous research has focused mostly on the
impact of detailing and sampling on physician prescribing (Lurie et al. 1990;
28
Lexchin 1989; Gönül et al. 2001, Manchanda and Chintagunta 2004, Mizik and
Jacobson 2004), other marketing communication events might also have an
impact on physician decision making. These other communication tools, which
include symposiums and out-of-the-office meetings
3
with other physicians,
represent a much smaller portion of the pharmaceutical marketing expenditures
when compared to detailing (Wittink 2002) and have remained largely unstudied.
I propose that these other forms of marketing communication can exert an
influence on physicians’ decision making and I will allow them to impact drug
choice, irrespective of physicians’ persistence state.
Finally, I allow the choice process to differ for a physician in the “persist” versus
the “non-persist” state at each prescription occasion. I assume the drug choice of
physicians in a persistent state at patient visit t will be affected by the choice
made at patient visit t–1 (irrespective of patient symptoms and demographic
profile). Physicians in a non-persistent state will not be influenced by the previous
drug choice. As a result, the deterministic component of the indirect utility for a
physician in each persistence state is as follows:
( )
ijt i ij
persist Non
ijt
omm MarketingC f V
* *
+ =
−
α , and
(2.3)
3
Detailing sessions are face-to-face meetings between a physician and a sales-representative in the
physician’s office; out-of-the-office meetings are meetings between a sales-representative and one
or more physicians and are usually held in a more pleasant context (e.g., over lunch or golf).
29
( )
ijt i ijt i ij
Persist
ijt
omm MarketingC f I V + + =
−1
δ α ,
(2.4)
where
s
ijt
V is the indirect utility of each alternative j (j = 1,…,J where J is the
number of alternative drugs) for state s (s = {“Persist”, “Non-Persist”}). The
terms
*
ij
α and
ij
α are physician and drug specific constants; I allow the intrinsic
preference of physicians for each drug to influence drug choice. This intrinsic
preference can be the result of the physician’s previous experience with each drug
and could be determined by the specific composition of the physician’s patient
base. The terms (.)
*
i
f and (.)
i
f are functions of current and previous levels of
marketing communication (detailing, symposiums, and out-of-the-office
meetings), possibly non-linear and with physician-specific parameters; δ
i
is the
state dependence parameter and I
ijt-1
is a dummy variable that takes the value one
if physician i prescribed the drug j to the previous patient (t–1), and zero
otherwise (the state dependence term influences only drug choice in the persistent
state).
Those physicians not taking into consideration patient welfare or medical
guidelines regarding alternative treatments could then be very persistent
physicians by prescribing always the same drug, regardless of patient symptoms.
In that case δ
i
would be positive and significant. However, being in a persistent
30
state does not necessarily mean that physicians are not taking into consideration
patient welfare, nor does it mean that physicians are not gathering relevant
information for diagnosis. Being in a persistent state might mean, for example,
that physicians have set-up some sort of simple prescribing rules that could be the
most effective for their own patients and at the same time minimize the learning,
search and thinking costs involved in drug decision making. The physician could
always start their patients on the drug with the least side effects (even though not
as powerful or not as effective). Physicians would then only switch if in
subsequent visits the patient would report undesired side-effects or simply no
effects at all. Then the physician could try the second level drug (usually more
powerful but also with more side effects), and so on. This pattern of prescribing
would lead to persistence but does not necessarily mean physicians are not taking
in consideration the desires and needs of the patient (e.g., the physician is
considering the importance of reducing side effects to improve the patients quality
of life, but at the same time minimizing search costs due to a busy schedule).
I adopt the standard formulation of Nerlove-Arrow (1962) to define the functions
(.)
*
i
f and (.)
i
f for each marketing communication tool. Consider the case of
detailing (I apply similar formulations for each form of marketing
31
communications). The detailing stock at time t for a brand j and physician i, D
ijt
,
is defined as:
1 −
+ =
ijt d ijt ijt
D d D λ ,
(2.5)
where λ
d
is the carry-over parameter of detailing and d
ijt
are the detailing visits
from drug j sales representatives that physician i has received in the month prior
to the prescription occasion t. This formulation allows detailing to be a stock
related to the flow of current and past detailing, allowing us to account for the
carry-over effects of detailing. This stock depreciates over time, reflecting the
loss in effectiveness of past detailing, and lets us calculate the long term effects of
detailing (this specification is also called the linear-distributed lag formulation
and has been used previously in the context of pharmaceutical industry; e.g.
Narayanan et al. 2004). I specify stocks for outside the office meeting and
symposium in a similar manner
Given the two model components (the two-state component and the drug choice
component), the probability that physician i will choose alternative j at time t, Pr
ijt
is given by:
)} | ( Pr ) ( Pr
) | ( Pr )) ( Pr 1 {( Pr
Persist j Persist
Persist Non j Persist
it it
it it ijt
× +
− × − =
.
(2.6)
32
The term Pr
ijt
(j|s), is the probability that doctor i chooses drug j for prescription
occasion t given the physician’s state s. This probability is given by the familiar
multinomial logit expression (McFadden 1978):
∑
+
+
= −
j
ijt i ij
ijt i ij
ijt
omm MarketingC f
omm MarketingC f
persist Non j
)) ( exp(
)) ( exp(
) | ( Pr
* *
* *
α
α
, and
(2.7)
()
∑
+ +
+ +
=
−
−
j
1
1
i
) ) ( exp(
) ) ( exp(
| Pr
ijt i ijt i ij
ijt i ijt i ij
jt
omm MarketingC f I
omm MarketingC f I
Persist j
δ α
δ α
.
(2.8)
Assuming independence across individual physicians, the likelihood function is
simply the product of the probabilities of each physician’s choices:
( ) () ( )
() ( )
∏∏∏
−= =
× +
− × −
=
I
i
T
t
J
j it it
it it
i
Persist j Persist
persist Non j Persist
L
11 1
| Pr Pr
| Pr Pr 1
) ( θ
(2.9)
Heterogeneity and Estimation
Individual decision makers might have different preferences over the alternatives
under consideration. In our pharmaceutical context, it is possible that a physician
might be prescribing the same drug repeatedly because her practice is in an area
where all the patients have similar clinical and demographic profile. The
prescription of the same drug to two or more new patients in a row could mean
33
that patients are very similar and not that the physician is persistent and
minimizing on learning, search, and thinking costs. These differences across
physicians can produce choice patterns similar to those of persistent decision
makers but, instead of endogenous, these differences would be determined by
exogenous factors not directly related to the history of the individual’s choices
(this confound created by heterogeneity is known as spurious state dependence
Heckman 1981). As a result, if heterogeneity is not adequately accounted for, we
will overestimate structural state dependence (Keane 1997).
To address this issue, I account for unobserved heterogeneity via a random
coefficient formulation: in general, a coefficient for a particular physician, β
i
, is
the sum of a population-level mean and a stochastic deviation representing the
physician i’s preference differences relative to the average preferences in the
population, that is, β
i
= β + η
i
where β is the population mean and η
i
is the
stochastic deviation. The stochastic deviations η
i
allow the unobserved portion of
the utilities to be correlated over alternatives and time, which in turn eliminates
the restrictive assumption of independence from irrelevant alternatives present in
standard logits. This random effects estimator is often referred to as the mixed
logit estimator and is widely used in the marketing and economics literature (see
McFadden and Train 2000, Brownstone and Train 1999 for details).
34
I estimate the proposed two-state dynamic mixed logit model via simulated
maximum likelihood. The simulated maximum likelihood estimator is consistent,
asymptotically normal, and efficient (Train 2003). In order to increase
computation efficiency in the estimation I use Halton draws instead of random
draws to compute the simulated probabilities. Halton draws achieve greater
precision and coverage for a given number of draws than normal random draws
(Train 2003). Hence, Halton draws allow us to increase efficiency without
sacrificing precision (e.g., for a mixed logit model, 100 Halton draws provide
more accurate results than 1000 random draws; Bhat 2001).
35
CHAPTER 3: DATA AND MARKET DESCRIPTION
This study employs a unique panel data set for the antidepressant therapeutic
market in the United Kingdom from 1996 to 1999 (the physician panel of general
physicians, GPs, is representative of the UK physician population). When
choosing an antidepressant the key criteria are efficacy, adverse effects, safety in
overdose, potential drug interactions, and withdrawal symptoms. Though all
antidepressants may appear to be equivalent in efficacy, each has a different
pharmacological profile with specific side effects and pharmacodynamics. There
is no single “best antidepressant” but consideration of each of the criteria for an
individual patient will help the physician find the antidepressant best suited to
each patient (Baldwin 2003).
During the period of this study there were more than 50 molecules available in the
market (henceforth, I use the terms brand, drug and molecule interchangeably).
Due to the large number of alternatives, I restrict my analysis to those molecules
with at least 5% of market share. This yielded a set of five molecules—Dotheipin,
Sertraline, Amitriptyline, Fluoxetine, and Paroxetine—belonging to two different
therapeutic subclasses. The first two molecules are tricyclic antidepressants
(TCAs) and the remaining three drugs are elective serotonin reuptake inhibitors
36
(SSRIs)
4
. The later entrants in the market, Fluoxetine, Paroxetine, and Sertraline,
were under patent protection during the time period of this study and had no
generic versions. The remaining molecules were not under patent protection and
were offered by several pharmaceutical companies. Because all drugs have been
in the market for at least five years, I can conclude that there are no considerable
differences in search and learning costs across the drugs.
The panel data I analyze include information on (1) physician characteristics
(gender, practice size, age), (2) competing marketing communication events for
the five retained drugs, and (3) the drug prescribed to each new patient. This is
quite distinct from data sets used in previous research. First, in previous research
“new prescriptions” did not necessarily mean new patients to the doctor (previous
data included also new prescription scripts written to returning patients and
patients continuing therapy). Second, I track all prescriptions of all competing
drugs written to each new patient by the panel of physicians. Third, I have
information on the marketing communications activity targeted to the panel of
4
Antidepressants can be categorized into different therapeutic subclasses based on the chemical
structure of the drug and its biological action. The subclasses are tricyclic antidepressants (TCAs),
monoamine oxidase inhibitors (MAOIs), selective serotonin reuptake inhibitors (SSRIs) and other
antidepressants (which may include tetracyclic compounds and other atypical antidepressants).
TCA molecules were introduced in 1959, year of the introduction of the therapeutic area, followed
by MAOI and then by SSRI molecules.
37
physicians from all competing firms. Finally, the marketing communication
activities include not only detailing information but also symposiums and out-of-
the-office meetings sponsored by pharmaceutical firms (free samples are not used
as a marketing tool in this market).
The data set consists of 9672 prescriptions written by 108 physicians over the
four-year period. Table 1 presents the prescription share of the five drugs, their
therapeutic subclass, and their year of entry.
Table 1: Descriptive Statistics of Sample
Group Molecule
Volume of
Prescriptions
Market
Share
Year of
Entry
SSRI Fluoxetine 5099 30.03 1989
TCA Dotheipin 4723 27.82 1969
SSRI Paroxetine 4415 26.01 1991
SSRI Sertraline 1169 6.89 1991
SSRI Amitriptyline 1571 9.25 1961
Table 2 presents overall sample statistics of the three types of marketing
communication and the share of marketing events for each drug. I can see from
these tables that “face-to-face” detailing accounts for about 90% of all marketing
communication activities in my dataset and that most of these activities are for the
38
branded drugs (these are the brands under patent protection, Fluoxetine,
Paroxetine, and Sertraline).
Table 2: Marketing Communications Summary Statistics *
Share (%)
Molecule
Detailing
Meetings
Outside-the-Office
Meetings
Symposium
Meetings
Fluoxetine 34.81 19.30 19.36
Dotheipin 0.03 0.93 0.00
Paroxetine 31.32 57.19 53.46
Sertraline 33.84 22.57 27.19
Total 89.30 8.00 2.70
*The drug Amitriptyline was not supported by marketing activities during
the time period of this study.
In Table 3, I present summary statistics of physician-specific variables. All
variables and their operationalization are listed in Table 4.
39
Table 3: Descriptive Statistics of Physician Characteristics
Physicians
Variable
Number Percentage
15-25 1686 17.43
26-30 4714 48.74
31-35 855 8.84
Experience
(Years)
>35 2417 24.99
Male 7957 82.27
Gender
Female 1715 17.73
1 619 6.40
2 1493 15.44
3 380 3.93
4 1651 17.07
5 1776 18.36
6 983 10.16
Practice
Size
7 2770 28.64
40
Table 4: Variable Description
Variable Operationalization
CumulDet
it
Total detailing visits that the physician i received from all
competing drugs until the prescription time period t (I used an
initialization period of 12 months prior to for all physicians).
Experience
i
Number of years of physician i's professional experience (years
since graduation).
PracticeSize
i
The number of doctors in the practice that physician i belongs to.
Gender
i
Dummy variable that takes the value of one if physician i is a
male, and zero if female.
d
ijt
Number of drug j detailing visits received by physician i one
month prior to prescription time t.
o
ijt
Number of outside the office meetings sponsored by drug j and
attended by physician i one month prior to prescription time t.
s
ijt
Number of symposiums sponsored by drug j and attended by
physician i one month prior to prescription time t.
D
ijt
Stock of detailing for drug j, physician i and prescription occasion
t following the Nerlove-Arrow formulation (I used the first 12
months to initialize the variable for all physicians).
O
ijt
Stock of outside the office meetings for drug j, physician i and
prescription occasion t following the Nerlove-Arrow formulation
(I used the first 12 months to initialize the variable for all
physicians).
S
ijt
Stock of symposium meetings for drug j, physician i and
prescription occasion t following the Nerlove-Arrow formulation
(I used the first 12 months to initialize the variable for all
physicians).
41
Key institutional features related to the UK market deserve special attention. First,
direct to consumer advertising (DTC Advertising) of prescription drugs is not
allowed in the U.K. and in the entire European Union. As a result, I can assume
that patients do not know much about the drugs and do not exert strong influence
on physicians. It further helps us to focus on physicians’ prescription behavior.
Second, prescription drugs are covered by the National Health System (NHS) for
most of the patients. They pay a flat fee per prescription regardless of the cost of
the drug. There is no variation in insurance status across patients. Therefore, drug
price and effects of insurance are not key issues. This key feature of the market
avoids the agency problems associated with different third-party payers as
physicians face a uniform incentive scheme.
42
CHAPTER 4: RESULTS
In this section, I discuss in detail the model results and their managerial
implications. First I present the fit comparison of the proposed model and of
alternative model formulations. Then I describe the results regarding the two-state
component of the model followed by the results from the drug choice component.
Finally, I discuss the managerial implications of my results.
Model Comparison
In order to test the proposed two-state dynamic mixed logit formulation
(2STATE), I estimate three alternative models. The first null model (NULL1) is a
traditional mixed logit formulation with random effects heterogeneity in all
parameters that does not separate physician states. The second null model
(NULL2) is the proposed two-state dynamic logit model estimated without
heterogeneity. Both NULL1 and NULL2 allow for physician persistence. The
third null model (NULL3) is a latent class logit model that also allows for
physician persistence (I tested for the number of segments required to account for
physician heterogeneity and based on Bayesian Information Criterion, BIC, the
43
two-segment solution provided the best fit). Table 5 provides the fit comparison
of alternative models.
Table 5: Comparison of Models
Model Description of the model
Log
Likelihood
BIC
2STATE
Proposed two state dynamic mixed
logit model
-13,040.00 13,255.66
NULL1
Traditional mixed logit model that does
not separate the two physician states
-13297.80 13,398.75
NULL2
Proposed two state dynamic logit
model without heterogeneity
-13500.50 13,619.80
NULL3
Latent class logit model that allows for
persistence
-13555.02 13,655.97
Several model comparisons deserve special attention. First, comparing the latent
class model (NULL3) with the proposed model without heterogeneity (NULL2)
allows us to isolate the effect of the two-state structure from the influence of
alternative heterogeneity methods (the 2STATE model allows for two segments
and for random effects heterogeneity in each segment; the NULL2 allows only for
the proposed two segment structure—persistent and non-persistent—without
accounting for heterogeneity within each segment). Second, comparing the
proposed two-state model with heterogeneity (2STATE) and a mixed logit that
does not separate physician states but has random effects heterogeneity in all
44
parameters (NULL1) allows us to determine the effect of the two-state structure
after accounting for random effects heterogeneity.
As can be seen from Table 5, both comparisons are in favor of the two-state
structure: the BIC of the two-segment latent class model (NULL3) is 13,655.97
and the BIC of the proposed two-state dynamic logit model estimated without
heterogeneity (NULL2) is 13,619.80; the BIC of the traditional mixed logit that
does not separate physician states (NULL1) is 13,398.75 and the BIC of the
proposed two-state model with heterogeneity (2STATE) is 13,255.66.
5
Finally, comparing the fit of models NULL1 (the traditional mixed logit that does
not separate physician states) and NULL3 (the latent class model) I can determine
the improvement in fit provided by the random-effects heterogeneity formulation
adopted in my study: the BIC of the model NULL1 is 13,398.75 (random effects
formulation) and is lower than the BIC of the model NULL3 (latent class
formulation), which is 13,655.97.
In Tables 6, 7 and 8, I present the results for the best fitting model, the proposed
2STATE model. I report the estimates and t-statistics for the persistence
5
The likelihood improvement between the traditional mixed logit model and the proposed two-
state model is statistically significant at the 1% significance level. The Wald test rejects the
traditional model in favor of the proposed model ( χ
2
statistic is 515 while the critical value is
χ
2
(25, 0.01) is 44.31).
45
predictors and for the drug choice component of the proposed model (these
include an estimate of the population level means for each parameter, and the
corresponding estimate of the population standard deviation).
Table 6: Estimation Results: Persistence Predictors
Parameter Description Estimate t-statistic
CumulDet -0.25 -4.03
Interaction (CumulDet x Experience) -0.23 -5.18
Experience 2.85 6.72
Practice Size -1.41 -5.05
Gender 0.26 -0.60
Table 7: Estimation Results: Drug Choice Model for the Non-Persistent State
Mean of Physician
Coefficients
Standard Deviation of
Physician Coefficients
Parameter Description
Estimate t-statistic Estimate t-statistic
Fluoxetine intercept 1.30 6.39 1.34 10.48
Dotheipin intercept 2.23 11.04 0.84 10.01
Paroxetine intercept 1.19 5.77 0.75 6.98
Sertraline intercept -4.78 -0.44 0.90 6.22
Detailing coefficient 0.37 2.61 1.06 3.03
Detailing carryover 0.51 3.18 1.30 7.86
Outside meetings coefficient -0.01 -0.38 0.94 1.80
Outside meetings carryover 0.51 1.13 0.91 1.98
Symposium coefficient 0.12 2.04 0.89 2.11
Symposium carryover 0.64 1.64 0.95 9.29
46
Table 8: Estimation Results: Drug Choice Model for the Persistent State
Mean of Physician
Coefficients
Standard Deviation of
Physician Coefficients
Parameter Description
Estimate t-statistic Estimate t-statistic
Fluoxetine intercept 0.74 15.79 0.97 67.23
Dotheipin intercept 0.67 13.43 0.99 19.80
Paroxetine intercept 0.58 12.11 0.99 7.84
Sertraline intercept -0.38 -6.15 1.00 4.38
Detailing coefficient 0.06 0.53 0.99 2.73
Detailing carryover 0.18 1.71 1.00 6.06
Outside meetings coefficient 0.08 1.04 0.90 5.62
Outside meetings carryover 0.37 1.45 1.02 4.78
Symposium coefficient 0.11 2.51 1.65 5.54
Symposium carryover 0.08 6.11 0.22 2.03
Persistence 1.37 35.13 0.85 6.40
Physician Persistence
From the estimates of persistence predictors (Table 6), I computed the persistence
probability given by Equation 2.1 for each prescription occasion. I find that
physicians exhibit persistence in about 58% of the prescription occasions (the
remaining 42% are non-persistent prescriptions). I then determine whether the
majority of these persistent prescriptions are associated to few physicians (in
which case persistence is a more cross-sectional feature of the market under
analysis), or whether physicians then to exhibit a mix of persistent and non-
47
persistent prescribing behavior (in which case persistence is a more longitudinal
feature of the market under analysis). To see this I calculate how many physicians
have persistent prescribing behavior and how often. I find that only about 2.5% of
the physicians in my data set switch states (from persistent to non-persistent)
during the period of my study. Hence, physicians can be classified into a
persistent (about 66%) and a non-persistent type (about 34%), that is, persistence
is a physician-specific feature.
Persistence Predictors
I hypothesized that cumulative detailing across all competing drugs (CumulDet
it
),
practice size (PracticeSize
i
), professional experience of a physician (Experience
i
),
and gender (Gender
i
) could play an important role as predictors of physician
persistence in a given prescription occasion. I also hypothesized an interaction
between Experience
i
and CumulDet
it
. From all of the hypothesized predictors only
gender provided no significant results (the coefficient estimate for Gender
i
is –
0.26 with a t-statistic of –0.60), suggesting that gender does no play a significant
role in the persistence likelihood for a given prescription occasion. Next, I discuss
the results for the remaining variables, all of which are significant
PracticeSize
i
I find a negative and significant practice size coefficient as
hypothesized (the coefficient estimate is –1.41 with a t-statistic of –5.05), that is,
48
physicians who work in larger practices are less persistent (all else constant). On
the one hand, physicians in larger practices are likely to see more heterogeneous
patients simply by facing a larger patient pool. This hinders the physicians’
strategy of applying simple prescribing rules to minimize search, learning, and
thinking costs while providing adequate care, which is consistent with my results.
On the other hand, my results are also consistent with the use of professional
colleagues as a low-cost and accessible information source. By talking to
colleagues in the same practice, physicians search, learning and thinking costs are
reduced and the use of simple prescribing rules becomes less relevant (reducing
persistence). From my dataset, I am unable to disentangle the two effects. I have
no information regarding how heterogeneous each patient pool is nor do I know
the frequency of physician contact, the specificities of physician network within
and across practices, and how physicians use the information provided by
colleagues. Analyzing how physician contact and horizontal communication
among physician might impact clinical practice is an important area for future
research.
CumulDet
it
The coefficient of cumulative detail is negative and significant as
hypothesized (the coefficient estimate is –0.25 with a t-statistic of –4.03). This
indicates that physicians receiving sales representatives more often (over time and
across pharmaceutical firms) tend to be of the non-persistent type. This further
49
suggests that detailing may play a significant role as a convenient and low-cost
source of medical information (Spilker 2002).
Experience
i
The positive and significant estimate of professional experience
(the coefficient estimate is 2.85 with a t-statistic of 6.72) indicates that physicians
with more years of experience tend to exhibit persistence. This is consistent with
the findings reported by medical studies that older physicians exhibit inertia in
their practice and that they may be less receptive to new standards of care.
However, the coefficient on the interaction term between cumulative detailing and
experience was negative and significant (the coefficient estimate is –0.23 with a t-
statistic of –5.18). Physicians with more experience will be able to better extract
the information sought from sales representatives (Shaughnessy and Slawson
1996). Hence, if physicians with more experience accept to see sales
representatives (from any drug company) it is perhaps because they trust the
detailing information and are more likely to be using the detailing meetings to
extract the information relevant for clinical practice. Consequently, these
physicians, as indicated by my results, are less likely to be persistent. I can then
see that older and more experienced physicians who rely on detailing meetings for
information gathering (convenient and low in search and thinking costs) are less
persistent by relying less on simple prescribing rules and by prescribing drugs that
are apparently more tailored to the specific situation of each new patient.
50
Prescription Choice
Tables 7 and 8 present the drug choice parameters for non-persistent and
persistent states respectively. The mean persistence parameter in the choice model
of the persistent state, δ
i
, is positive and highly significant with an estimate of 1.37
and a t-statistic of 35.13 (for the non-persistent state drug choice is not a function
of the drug prescribed to the previous patient). This means that there are clearly
two types of prescribing situations or, as in my case, two types of physicians with
clearly different decision processes: those persistent and those not persistent.
Regarding the effects of marketing communications, I estimate the impact of
detailing, symposiums, and out-of-the-office meetings under both states. Below I
present the results for each one of the marketing communications tools.
Detailing Meetings Comparing the estimates and significance of the detailing
parameters for the persistent versus the non-persistent states, I detect two
important differences. First, detailing meetings have a positive and significant
effect on non-persistent doctors but they have no impact on drug choice of
persistent doctors. The estimate of the mean coefficient for the non-persistent
state is 0.37 with a t-statistic of 2.61, and for the persistent state is 0.06 with a t-
statistic of 0.53. In addition, the carry-over parameters associated with detailing is
significant for the non-persistent state and not significant for the persistent state.
51
These results suggest that physicians who tend not to exhibit persistence are
responsive to detailing, whereas physicians who exhibit persistence are not
responsive to detailing.
This is a result consistent with my previous findings on cumulative detailing and
persistence. My results had revealed that the more physicians were willing to
accept detailing visits (from any pharmaceutical company), the lower the
likelihood that those physicians will be persistent in a given prescription occasion.
I then suggested that by accepting to receive sales representatives, physicians
demonstrate their willingness to use the information obtained from detailing
meetings as a convenient, low-cost solution to the information gathering problem
in daily practice. As a result, physicians receiving more sales representatives were
also less persistent. My additional finding that non-persistent physicians are
responsive to detailing, and that persistent physicians are not responsive, further
corroborate my previous conclusions.
Out-of-the-office Meetings There has been a lot of debate worldwide about the
value of pharmaceutical firm’s hospitality and gifts to physicians. Titled “No
more free lunches”, a recent editorial column of British Medical Journal, (Abbasi
and Smith 2003), argues that doctors, drug companies and most importantly
52
patients would all benefit from a greater distance between doctors and drug
companies. The American College of Physicians also discourages the acceptance
of individual gifts, hospitality, and trips with the belief that acceptance of even
small gifts affects clinical judgment and heightens the perception of a conflict of
interest (Hall 2001). As a result companies have recently started eliminating perks
to physicians that could be viewed as inappropriate (that is the example of Merck
& Co; see Petersen 2002a). Despite all the debate around the value of such forms
of marketing communication, my findings suggest that for both the persistent and
non-persistent states, out-of-the-office meetings do not exert a significant
influence on physicians prescription behavior. The estimate of the mean
coefficient for the persistent state is 0.08 with a t-statistic of 1.04, and for the non-
persistent state is –0.01 with a t-statistic of –0.38 (the carry-over parameters
associated with out-of-the-office meetings in both states are also not significant).
Though significant resources are still allocated towards this marketing tool, my
results suggest that physicians’ prescription behavior is not influenced by “wine
and dine” activities. Given the attention that this has received from
pharmaceutical firms, the industry itself, and the states (Petersen 2002b), my
study provides valuable new insights.
53
Symposium Meetings For both the persistent and non-persistent state,
symposium meetings have a positive and significant effect on physician drug
choice (at the 5% significance level). The estimate of the mean coefficient for the
persistent state is 0.11 with a t-statistic of 2.51, and for the non-persistent state is
0.12 with a t-statistic of 2.04. However, the carry-over parameter associated with
the symposium variable is not significant for the non-persistent state, though it is
significant for the persistent state. This means that the effect of symposiums
dissipates faster for non-persistent physicians than for persistent physicians. This
also means that the long term effects of symposium meetings are much smaller
for non-persistent than for persistent physicians, and that the profitability of
symposiums can vary greatly depending on the type of doctor invited. To the best
of my knowledge, this is the first paper to document that symposium meetings do
have an effect on physician drug choice using a panel data of physician
prescribing. It is also the first time that individual differences of long-term effects
of symposium meetings are documented in the literature.
54
CHAPTER 5: DISCUSSION
Physicians’ persistence with a particular drug implies that they may not be willing
to change their prescription behavior. This might suggest that those physicians
who are persistent are not responsive to a firm’s sales force efforts despite the
considerable sums pharmaceutical firms spend in promoting their drug (Wittink
2002). My results suggest that persistence is a mainly a cross-sectional market
characteristic, that is, physicians tend to be either persistent or non-persistent.
More importantly, I find that physicians who exhibit persistence respond
differentially to pharmaceutical marketing communications when compared to
non-persistent physicians. Persistent and non-persistent physicians not only
exhibit different intensity of response to marketing communications, they also
might be responsive to one tool and not responsive to another. First, I find that
out-of-the-office informal meetings with physicians have no influence on the
prescription choice behavior. Given that some of these meetings are very
expensive for pharmaceutical companies, it would be important to better
understand the impact of these meetings (e.g., perhaps they can stimulate category
growth, an issue not studied in the current paper, or perhaps such meetings are
intended for brand building and maintaining a good rapport with the physicians to
55
open the doors for future detailing meetings, a communications tool that can be
effective).
Second, non-persistent physicians are responsive to detailing efforts whereas
persistent physicians do not respond to detailing. Considering the significant
carry-over effect of detailing for the non-persistent group, I find that the long run
effect of detailing for the non-persistent segment is 0.76 (0.37×[1/(1–0.51)]). The
short-run effect sizes of detailing for the non-persist group are also not negligible.
Table 7 reports the detailing elasticities for the non-persistent state (detailing
elasticities for the persistent state are not significant). The own detailing
elasticities vary from 0.26 for Sertraline and about 0.07 for Fluoxetine and
Paroxetine, whereas the cross-detailing elastcities vary from –0.06 to –0.03. For
pharmaceutical companies these results are highly encouraging because they
suggest that (1) at least some physicians are responsive to detailing long after the
introduction of the drugs, (2) it is possible to identify and target the physicians to
make better use of resources (as also suggested by Manchanda et al. 2004), and
(3) detailing has an impact on switching across drugs and not only on the number
of prescription as previously reported in the literature (previous results on
detailing effectiveness confound primary and secondary demand effects due to the
56
nature of the data used; e.g., Manchanda et al 2004; Manchanda and Chintagunta
2004; Mizik and Jacbson 2004).
Table 9: Short-Run Effects of Marketing Communications on Drug Choice
*
Drug Changing 1% of Marketing
Communications
Description
Drug Being
Affected
Fluoxetine Paroxetine Sertraline
Fluoxetine 0.0715 -0.0371 0.0000
Dotheipin -0.0352 -0.0266 0.0000
Paroxetine -0.0489 0.0729 0.0000
Sertraline -0.0589 -0.0395 0.2579
Detailing Elasticities:
Non-Persistent State
Amitriptyline -0.0352 -0.0266 0.0000
Fluoxetine 0.0002 -0.0001 0.0000
Dotheipin -0.0001 -0.0002 0.0000
Paroxetine -0.0000 0.0008 0.0000
Sertraline -0.0001 -0.0002 0.0004
Symposium Elasticities:
Non-Persistent State
Amitriptyline -0.0001 -0.0002 0.0000
Fluoxetine 0.0001 -0.0001 0.0000
Dotheipin -0.0001 -0.0002 0.0000
Paroxetine -0.0000 0.0007 0.0000
Sertraline -0.0001 -0.0001 0.0004
Symposium Elasticities:
Persistent State
Amitriptyline -0.0000 -0.0001 0.0000
* The table gives the impact of 1% change in a column brand on the choice probability of a row
brand. For example, 1% change in Fluoxetine detailing will increase Fluoxetine own choice
probability by 0.0715% and decrease the choice probability of Dotheipin by 0.0352%. Note
that I do not report the elasticities of outside the office meetings variable because it was not
significant for both states. I have omitted the columns for the change in marketing effort of
Dotheipin and Amitriptyline because these have no impact on choice probabilities (own- and
cross-elasticities are zero).
57
Finally, even though the persistent physicians are not responsive to detailing they
can still be influenced through symposium meetings (both persistent and non-
persistent physicians are responsive to symposium events). The immediate effect
of symposium meetings on both physician groups is very small with a maximum
value of 0.0008 for the own symposium elasticity of Paroxetine (Table 9 reports
the symposium elasticities for persistent and non-persistent physicians). However,
symposium meetings have significant long-term effects for persistent physicians
of about 0.12 (symposiums also have an effect on non-persistent physicians,
though smaller and only short-term). This is quite meaningful for pharmaceutical
companies in their resource allocation decisions: symposiums provide drug
companies an alternative route of influencing persistent physicians, who are not
responsive to detailing but whose conversion could generate considerable
additional revenue. Hence, if physicians consistently refuse to see sales
representatives, an indication that they are of the persistent type, pharmaceutical
companies could invite these physicians for symposiums. Pharmaceutical
companies could also determine the persistent and non-persistent physicians,
using the full set of predictors available, and engage in a more targeted marketing
communications strategy.
58
An important take away from these results for managers is targeting the two
different segments differentially. Instead of targeting all the physicians equally,
the return on investment (ROI) on marketing communication would be higher if
the firms can identify those physicians who exhibit persistence from those who do
not. In contrast, pharmaceutical firms currently target physicians by the volume
of prescriptions they write (Brand and Kumar 2003, Manchanda et al. 2004).
Manchanda et al. 2004 report high volume physicians are detailed to a greater
extent than low volume physicians
6
. This is not an optimal way of detailing as it
does not take into account the responsiveness of physicians to detailing. Based on
the results of my model, I suggest that physicians should target the non-persist
physicians with detailing as they are more responsive to this form of marketing
communication. My results also help shed light on some of the demographic
variables that can identify a non-persistent physician. I find that various
physician characteristics such as professional experience and the size of their
practice influence the physicians’ propensity to exhibit persistence and
consequently their responsiveness to detailing.
Finally, physicians’ persistence has also important implications for understanding
the market structure and policy implementation. Persistence in choice is closely
related switching costs and hence has an important role in understanding market
6
I also find that physicians who write more prescriptions are detailed more in my dataset.
59
behavior in any industry. The persistence at the individual physician level leads to
persistence in market share that we observe at the macro level (Coscelli 2000)
which can create barriers to entry in the pharmaceutical industry. Since a new
drug usually is more effective than existing products (at least for parts of the
population), society may not materialize the value of the innovation if physicians
exhibit persistence. The presence of switching costs among consumers in a market
has implications also for pricing and market competitiveness (Klemperer 1987).
60
CHAPTER 6: CONCLUSION
A multitude of previous studies in the medical literature have found that
physicians might not change previous practice due to inertia or lack of motivation
to change, irrespective of their knowledge of recent medical advancements. This
behavior is consistent with the non-adherence to clinical guidelines by physicians,
a phenomenon widely recognized by health care organizations and the medical
literature (clinical guidelines are systematically developed to assist practitioners
and patients in their decisions of appropriate health care). This behavior also
poses important questions regarding the decision-making process of health-care
professionals and its similarity to choice in a packaged goods context.
This study investigates one potential similarity between physicians’ drug choice
behavior and consumers’ choices of frequently purchased goods, namely choice
persistence. Specifically, I examine (1) whether physicians’ drug choices might
dependent structurally on the drugs prescribed previously, that is, whether
sequential drug choices exhibit structural state dependence, and (2) whether
physicians’ short- and long-term response to marketing communication efforts
(detailing, outside the office meetings, and symposiums) are affected by this
potential drug choice persistence.
61
I differ from the existing approaches in the literature by proposing that physicians
face search, thinking, and learning costs when deciding which drugs to prescribe
to patients. I further suggest that some physicians might exhibit persistence with a
drug to minimize on such costs. Structural persistence in physician drug choice, if
present, will have important consequences on cost/benefit analyses of marketing
actions. For example, to draw valid inferences on the effectiveness of each
marketing tool it is essential to allow for state dependence (and individual
heterogeneity) and then simulate the dynamic response to the different tools.
I develop an individual-level, two-state dynamic model of physicians’ drug choice
rooted in decision making theory to investigate physicians’ persistence in
prescription choice behavior. I model jointly the probability of a physician being
persistent (or not-persistent) and the drug chosen by the physician for each new
patient visit. I allow for current and lagged effects of marketing communication
on drug choice, and estimate the effect of detailing, outside the office meetings,
and symposiums. I further incorporate competitive marketing communication
efforts, and account for physicians’ unobserved heterogeneity using a random
effects approach estimated via simulated maximum likelihood.
62
My results indicate that physicians’ persistence is mostly a cross-sectional market
characteristic: physicians tend to be of the persistent or non-persistent type and do
not switch back and forth between the two states. In addition, about 66% of
physicians, and about 58% of the prescription occasions, are of the persistent
type, indicating that persistence is a significant phenomenon also in medical
decision making. Hence, structural state-dependence generalizes beyond the more
traditional packaged goods categories which have been the focus of previous
persistence studies. This result further suggests that the strategies decision makers
use to cope with search and thinking costs might easily transfer across decision
types, irrespective of how different such decisions are. Exploring what are the
cost-minimizing decision-making strategies that are easily transferable and
widespread across decision types is an important avenue for future research.
Those physicians not taking into consideration patient welfare or medical
guidelines regarding alternative treatments could then be very persistent
physicians by prescribing always the same drug, regardless of patient symptoms.
In that case δ
i
would be positive and significant. However, being in a persistent
state does not necessarily mean that physicians are not taking into consideration
patient welfare, nor does it mean that physicians are not gathering relevant
information for diagnosis. Being in a persistent state might mean, for example,
63
that physicians have set-up some sort of simple prescribing rules that could be the
most effective for their own patients and at the same time minimize the learning,
search and thinking costs involved in drug decision making. The resulting patterns
of prescribing would lead to persistence but do not necessarily mean physicians
are not taking in consideration the desires and needs of the patient.
Consistent with my predictions, my model results also suggest that older
physicians and physicians who work in smaller practices are more likely to
exhibit persistence. Furthermore, physicians who meet more frequently the
pharmaceutical sales representatives of all competing firms are less likely to
exhibit persistence. This indicates that acceptance to receive sales representatives
from all competitors can serve as a signal of physicians’ lower information search
and thinking cost (e.g., more time available) and of the higher interest in gathering
medical information either directly from pharmaceutical companies or through
other means. Hence, physicians reveal their interest and availability to search for
information to the pharmaceutical by accepting to meet the sales representatives.
Pharmaceutical firms can then use this signal to differentiate between persistent
and non-persistent physicians.
64
Finally, the model estimates provide a rich set of results that could be useful for
improving marketing communications effectiveness. I have established the
responsiveness of persistent and non-persistent physicians to different forms of
marketing communication (I believe this is the first time a study reports on and
compares the individual-level impact of different forms of marketing
communications on drug choice). First, unlike persistent physicians, non-
persistent physicians are responsive to detailing. The significant response to
detailing of the non-persistent physicians is not only short-term but also long-
term. Second, out-of-the office meetings such as lunches and golf have no
influence on physicians’ prescription behavior. Finally, both persistent and non-
persistent physicians respond to symposium meetings. However, the short-term
effects are small and long-term effects are only present for persistent physicians.
This indicates that even though persistent physicians will not be influenced by
detailing, these physicians can still be reached by the companies through the use
of symposium meetings, justifying their use alongside detailing in more effective
marketing communication campaigns. For pharmaceutical companies these results
are highly encouraging because they suggest that pharmaceutical firms can reach
the different types of physician with the most effective tools and that the effects of
some of these actions can be long-lasting.
65
As an initial modeling study of physician persistence there is much I did not do
leading to limitations in my work and avenues for future research. First, I have not
accounted for patient differences in my study because such information was not
available to us. I do allow for physician-level heterogeneity, which will account
for differences of patient pools across physicians, but I cannot account for the
differences across patients seen by one physician. In addition, I have not taken
into account prescribed drug dosages and drug forms (e.g., capsule versus liquid).
Studying jointly persistence in drug choice, treatment dosages, and drug forms
would provide additional insights on physician compliance to medical guidelines.
Finally, even though I take advantage of physician receptivity to detailing to
predict physician persistence, I do not directly model the physician’s decision to
accept detailing meetings (unlike other forms of marketing communication
physicians cannot simply be “exposed” to detailing but must accept to see a
pharmaceutical sales representative; detailing meetings are a result of a joint
decision of pharmaceutical companies and physicians). Modeling the physician
decision to see sales representatives, determining what influences such decisions,
and incorporating it in a model of market response is a worthwhile area for future
research that would require data on rejected meetings.
66
BIBLIOGRAPHY
Abbasi, Kamran and Richar Smith (2003), “No More Free Lunches,” British
Medical Journal, 326, 1155–1156.
Abramson, Charles, Rick L. Andrews, Imran S. Currim and Morgan Jones (2000),
“Parameter Bias from Unobserved Effects in the Multinomial Logit Model of
Consumer Choice,” Journal of Marketing Research, 37(4), 410–426.
Alvanzo Anika H., Gail M. Cohen and Mary Nettleman (2003),“Changing
physician behavior: half-empty or half-full?”,Clinical Governance: An
International Journal, 8(1), 69–78.
Avorn, Jerry, Milton Chen and Robert Hartley (1982), “Scientific Versus
Commercial Source of Influence on the Prescribing Behavior of Physicians”
American Journal of Medicine, 73 (July), 4–8.
Baldwin, David (2003), “Essential considerations when choosing a modern
antidepressant,” International Journal of Psychiatry in Clinical Practice, 7, 3–8.
Bauer, Raymond A. and Lawrence H. Wortzel (1966), “Doctor’s Choice: The
Physician and his Sources of Information,” Journal of Marketing Research, 3(1),
40–46
Bhat, Chandra (2001), “Quasi-random Maximum Simulated Likelihood
Estimation of the Mixed Multinomial Logit Model,” Transportation Research B,
35, 677–693.
Brand, Ron and Pankaj Kumar (2003), “Detailing Gets Personal,” Pharmaceutical
Executive, 23, 66.
67
Brownstone, David and Kenneth Train (1999) “Forecasting New Product
Penetration with Flexible Substitution Patterns,” Journal of Econometrics, 89,
109–129.
Cabana, Michael D., Cynthia S. Rand, Neil R. Powe, Albert W. Wu, Modena H.
Wilson, Paul-André C. Abboud and Haya R. Rubin (1999), “Why don't
physicians follow clinical practice guidelines? A framework for improvement,”
Journal of American Medical Association, 282, 1458–65.
Caudill T. S., M. S. Johnson, E. C. Rich and W. P. McKinney (1996),
“Physicians, pharmaceutical sales representatives, and the cost of prescribing,”
Arch Fam Med., 5, 201–206
Christakis, Dimitri A., Frederick J. Zimmerman, Jeffrey A. Wright, Michelle M.
Garrison, Frederick P. Rivara, and Robert L. Davis (2001) , “A Randomized
Controlled Trial of Point-of-Care Evidence to Improve the Antibiotic Prescribing
Practices for Otitis Media in Children,” Pediatrics, 107, 15–18.
Choudhry, Niteesh K., Robert H. Fletcher and Stephen B. Soumerai, (2005)”
Systematic Review: The Relationship between Clinical Experience and Quality of
Health Care,” American College of Physicians, 142(4), 260–273.
Coleman, James S., Elihu Katz, and Herbert Menzel (1966), “Medical Innovation:
A Diffusion Study,” Indianapolis, IN: The Bobbs-Merrill Company.
Connelly, Donald P., Eugene C. Rich, S.P. Curley, and J.T. Kelley (1990),
“Knowledge Resource Preferences of Family Physicians,” Journal of Family
Practice, 30 (3), 353–359.
Coscelli, Andrea (2000), “The importance of doctors’ and patients’ preferences in
the prescription decision,” Journal of Industrial Economics, 48(3), 349–369.
68
------- and Mathew Shum (2004), “An Empirical Model of Learning and
Spillovers in New Drug Entry,” Journal of Econometrics, 112 (October), 213–
246.
Crawford, Greg and Mathew Shum (2004), “Uncertainty and Learning in
Pharmaceutical Demand,” forthcoming in Econometrica.
Currie, Gillian R. and Sangin Park (2002), “The Effects of Advertising and
Consumption Experience on the Demand for Antidepressant Drugs,” Working
Paper, University of Calgary, Calgary, Alberta, Canada.
Dholakia, Ruby Roy and Brian Sternthal (1977), “Highly Credible Sources:
Persuasive Facilitators or Persuasive Liabilities?” Journal of Consumer Research,
3 (March), 223–232.
Eagley, Alice H. and Shelly Chaiken (1975), “An Attribution Analysis of the
Effect of Communicator Characteristics on the Opinion Change: The Case of
Communicator Attractiveness,” Journal of Personality and Social Psychology, 32
(July): 136–144.
Freiman, MP ( 1985), “The rate of adoption of new procedures among physicians.
The impact of specialty and practice characteristics,” Medical Care, 23, 939–45.
Gorman, PN. (1995), “Information needs of physicians,” J Am Soc Inf Sci, 46,
729–36.
Gönül, Fusun, Franklin Carter, Elina Petrova and Kannan Srinivasan (2001),
“Promotion of Prescription Drugs and its Impact on Physicians’ Choice
Behavior,” Journal of Marketing, 65, pp. 79–90.
Hall, William (2001), “The Ethical Dilemma of Accepting Gifts from Drug
Makers,” ACP-ASIM Observer by the American College of Physicians-American
Society of Internal Medicine.
69
Heckman, James J.(1981), “Heterogeneity and state dependence”, S. Rosen, ed.
Studies in Labor Markets. University of Chicago Press, Chicago, IL, 91–140.
Hellerstein, Judith K.(1998), “The Importance of the Physician in the Generic
versus Trade-Name Prescription Decision,”The RAND Journal of Economics,
29(1), 108–136.
Hlatky MA, Cotugno H, O’Connor C, Mark DB, Pryor DB and Califf RM (1988),
“Adoption of thrombolytic therapy in the management of acute myocardial
infarction,” American Journal of Cardiology, 61, 510–4.
Hodges, Brian (1995), “Interactions with the pharamceutical industry: experiences
and attitudes of psychiatric residents, interns and clerks”, Canadian Medical
Association Journal, 153, 553–559
Jeuland, Abel P. (1979), "Brand Choice Inertia As One Aspect of the Notion of
Brand Loyalty." Management Science, 25(7), 671–82.
Kamakura, A.Wagner, and Gary J. Russell (1989), “A Probabilistic Choice
Model for
Market Segmentation and Elasticity Structure,” Journal of Marketing Research,
26(2), 379–90.
Keane, Micheal K.(1997), “Modeling heterogeneity and state dependence in
consumer choice behavior. Journal of Business and Economic Statistics, 15(3),
310–327.
Klemperer, Paul (1987), “Markets with Consumer Switching Costs,” The
Quarterly Journal of Economics, 102, 375–394.
Laine, Christine, and David S. Weinberg (1999), “How Can Physicians Keep up-
to-date,” Annu.Rev.Med., 50, 99–110.
70
Lexchin, Joel (1989), “Doctors and detailers: Therapeutic education or
pharmaceutical promotion,” International Journal of Health Services, 19, 663–
679.
Lichstein, Peter R., Robert C. Turner, and Kevin O’Brien (1992), “Impact of
Pharmaceutical Company Representatives on Internal Medicine Residency
Programs: a Survey of Residency Program Directors,” Archives of Internal
Medicine, 152 (5), 1009-1013.
Lilien, Gary L., Ambar Rao and Shlomo Kalish (1981), “Bayesian Estimation and
Control of Detailing Effort in a Repeat Purchase Environment,” Management
Science, 27, 493–507.
Lurie N., E. C. Rich, D. E. Simpson et al. (1990), “Pharmaceutical representatives
in academic medical centers: interaction with faculty and housestaff,” Journal of
General Internal Medicine, 5, 240–243.
Manchanda, Puneet, and Pradeep K. Chintagunta, (2004), “Responsiveness of
Physician Prescription Behavior to Salesforce Effort: An Individual Level
Analysis,” Marketing Letters, 15 (2–3), 129–145.
Manchanda, Puneet, Peter E. Rossi and Pradeep K. Chintagunta (2004),
“Response Modeling with Non-Random Marketing Mix Variables,” Journal of
Marketing Research, 41 (November), 467–478.
Medical Marketing and Media (2000), “Pharmaceutical company outlay tops $8
billion in the first half 2000,” December, 2000.
McAlister, Leigh (1982), "A Dynamic Attribute Satiation Model of Variety-
Seeking Behavior," Journal of Consumer Research, 9 (September), 141–150.
71
McFadden, Daniel.,(1978), “Modeling the Choice of Residential Location,” in A.
K. et al. eds., Spatial Interaction Theory and Residential Location, North Holland
Pub. Co.
McFadden, Daniel and Kenneth Train (2000), “Mixed MNL Models of Discrete
Response,” Journal of Applied Econometrics, 15(5), 447–470.
McKinley, W. Paul, David L. Schiedermayer, Nicole Lurie, Deborah E. Simpson,
Jesse L. Goldman, Eugene C. Rich (1990), “Attitudes of Internal Medicine
Faculty and Residents toward Professional Interactions with Pharmaceutical Sales
Representatives”, Journal of the American Medical Association, 264 (13), 1693-
1697.
Meyers-Levy, Joan (1988), “The influence of sex roles in judgment,” Journal of
Consumer Research, 14, 522–530.
----- and Brian Sternthal (1991), “Gender Differences in the Use of Message
Cues and Judgments,” Journal of Marketing Research, 28(1), 84–96
Mizik, Natalie and Robert Jacobson (2004), “Are Physicians Easy Marks?:
Quantifying the Effects of Detailing and Sampling on New Prescriptions”,
Management Science, 50(12), 1704–1715
Mottur-Pilson, Christel, Vincenza Snow and Kyle Bartlett, (2001), “Physician
explanations for failing to comply with best practices,” Effective Clinical
Practice, 4, 207–213.
Mukherji, Prokriti, Shantanu Dutta and Surendra Rajiv (2004), “A Structural
Model of Learning for New Products,” working paper, University of Minnesota.
Narayanan, Sridhar, Puneet Manchanda and Pradeep K. Chintagunta (2004),
"Temporal Differences in the Role of Marketing Communication in New Product
Categories," Journal of Marketing Research (forthcoming).
72
Narayanan, Sridhar, Ramarao Desiraju and Pradeep Chintagunta (2004) “Return
on Investment Implications for Pharmaceutical Promotional Expenditures: The
Role of Marketing Mix Interactions,” Journal of Marketing, 68, 4 (October), 90-
105
Nerlove, Marc and Kenneth J. Arrow (1962), “Optimal Advertising Policy under
Dynamic Conditions,” Economica, 29, 129–142.
Parsons, Leonard J. and Piet Vanden Abeele (1981), “Analysis of Sales Call
Effectiveness,” Journal of Marketing Research, 18, 107–113.
Pauker SD, Gorry G, Kassirer J, Schwartz W. (1976), “Towards the simulation of
clinical cognition: taking a present illness by computer,” Am J Med., 60(7), 981–
96
Peay, Marilyn Y. and Edmund R. Peay (1990), “Patterns of Preference for
Information Sources in the Adoption of a New Drugs by Specialists,” Social
Science and Medicine, 31 (4), 467–476.
Petersen, Melody (2002a), “Merck is said to limit perks in marketing to
physicians”, The New York Times (18 January).
Petersen, Meldoy (2002b), “Vermont to require drug makers to disclose payment
to doctors”, The New York Times (13 June).
Phillips, Lawrence S., William T. Branch, Jr., Curtiss B. Cook, Joyce P. Doyle,
Imad M. El-Kebbi, Daniel L. Gallina, Christopher D. Miller, David C. Ziemer,
and Catherine S. Barnes, (2001), “Clinical Inertia,” Ann Intern Med, 135(Nov),
825–834.
Reich, Lisa (2005), “Are GPs in thrall to the drug companies?” Telegraph,
February 23.
73
Seetharaman, P.B. (2004), “Modeling Multiple Sources of State Dependence in
Random Utility Models: A Distributed Lag Approach,” Marketing Science, 23
(2), 263-271.
Seetharaman, P.B., Andrew Ainslie and Pradeep K. Chintagunta (1999),
“Investigating Household State Dependence Effects Across Categories,” Journal
of Marketing Research, 36 (4) 488-500.
Shaughnessy AF, and Slawson DC (1996), “Pharmaceutical Representatives”,
British Medical Journal, 312 (June), 1494.
Shugan, Steve (1980), “The Cost of Thinking”, Journal of Consumer Research,
7(2), 99–111.
Smith, Richard (1996), “Information in Practice: What clinical information do
doctors need?” British Medial Journal, 313 (October),1062–1068.
Spilker, Bert (2002), “The Benefits and Risks of a Pack of M&M’s: A
Pharmaceutical Spokesman Answers His Industry’s Critics,” Health Affairs, 21
(2), 243-244.
Stern, Scott and Manuel Trajtenberg (1998), “Empirical Implications of Physician
Authority in Pharmaceutical Decision Making,” NBER working paper 6851.
Strang D, Gagnon M, Molloy W, Bedard M, Darzins P, Etchells E and Davidson
W. (1996), “National survey on the attitudes of Canadian physicians towards
drug-detailing by pharmaceutical representatives,” Ann R Coll Physicians Surg
Can., 29(8), 474–8.
Train, Kenneth (2003), Discrete Choice Methods with Simulation, Cambridge
University Press, New York.
74
University of Pittsburgh Medical News Center News Bureau
(http://newsbureau.upmc.com/Medsurg3/IcuChestStudy.htm), Accessed April 11,
2005.
Van den Bulte, Christophe and Gary L. Lilien (2001), “Medical Innovation
Revisited: Social Contagion versus Marketing Effort.” American Journal of
Sociology, 106, 1409–35.
Wazana, Ashley (2000), “Physicians and the pharmaceutical industry: is a gift
ever just a gift?” Journal of American Medical Association, 283 (January), 373–
380
Williamson JW, German PS, Weiss R, Skinner EA and Bowes F., (1989), “Health
science information management and continuing education of physicians. A
survey of US primary care practitioners and their opinion leaders,” Ann Intern
Med., 110, 151–60.
Wittink, Dick R. (2002), “Analysis of ROI for pharmaceutical promotions”,
unpublished study conducted for the Association of Medical Publications,
available at www.rappstudy.org.
Wolf, Bruce L. (1998), “Drug Samples: Benefit or Bait?” Journal of the American
Medical Association, 279 (21), 1698-1699
Wyatt J. (1991), “Uses and sources of medical knowledge,” Lancet, 338,1368–72.
Young MJ, Fried LS, Eisenberg J, Hershey J, and Williams S. (1987), “Do
cardiologists have higher thresholds for recommending coronary arteriography
than family physicians?” Health Serv Res., 22 (5), 623–35.
UMI Number: 3238347
3238347
2007
UMI Microform
Copyright
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
ProQuest Information and Learning Company
300 North Zeeb Road
P.O. Box 1346
Ann Arbor, MI 48106-1346
by ProQuest Information and Learning Company.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Multiobjective decision methods in a capital budgeting context: A laboratory experiment
PDF
A critical analysis of goodwill arising from consolidation and the varying effects on financial statements
PDF
The impact of paperless systems and other technological changes upon auditing
PDF
Purposive, goal-oriented knowledge and attitudes: the case of brand extensions
PDF
Elements of kanban production control for dynamic job shops
PDF
Toward a general theory of governmental accounting
PDF
Spatial distribution of neuroendocrine motoneuron pools in the hypothalmic paraventricular nucleus
PDF
The project life cycle and budgeting functions: planning, control, and motivation
PDF
Semiperipheral mobility in the world economy: the experience of South Korea's industrial upgrading
PDF
Initiation mechanism of estrogen neuroprotection pathway
PDF
Internet security and quality-of-service provision via machine-learning theory
PDF
Elevated inflammation in late life: predictors and outcomes
PDF
The importance of using domain knowledge in solving information distillation problems
PDF
Efficient media access and routing in wireless and delay tolerant networks
PDF
Protocol evaluation in the context of dynamic topologies
PDF
It's only temporary?: The reproduction of gender and race inequalities in temporary clerical employment
PDF
Dual devotions: African American clergywomen and work-family dilemmas
PDF
Participation in informal labor markets: evidence from unauthorized Latina immigrants in Los Angeles County
PDF
The economics of tenure: understanding the effects of ethnicity, status and discipline on faculty attitude, workload and productivity
PDF
Transfer pricing policies for multinational corporations
Asset Metadata
Creator
Janakiraman, Ramkumar
(author)
Core Title
The effects of marketing communication on consumers' choice behavior: the case of pharmaceutical industry
School
Graduate School
Degree
Doctor of Philosophy
Degree Program
Business Administration
Degree Conferral Date
2006-08
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
business administration, marketing,OAI-PMH Harvest
Language
English
Contributor
Digitized by ProQuest
(provenance)
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c17-135961
Unique identifier
UC11349409
Identifier
3238347.pdf (filename),usctheses-c17-135961 (legacy record id)
Legacy Identifier
3238347.pdf
Dmrecord
135961
Document Type
Dissertation
Rights
Janakiraman, Ramkumar
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 au...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA
Tags
business administration, marketing