Using virtual environments to unobtrusively measure real-life risk-taking: findings and implications for health communication interventions

PDF

Download

Open document

Flip pages

Summarize

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

USING VIRTUAL ENVIRONMENTS TO UNOBTRUSIVELY MEASURE
REAL-LIFE RISK-TAKING: FINDINGS AND IMPLICATIONS
FOR HEALTH COMMUNICATION INTERVENTIONS

by

Carlos Gustavo Godoy
_________________________________________________________________________

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(COMMUNICATION)

December 2007

Copyright 2007 Carlos Gustavo Godoy

ACKNOWLEDGEMENTS
Writing a dissertation is like creating a piece of fine art. First, it requires an
artist, who has the creative vision and tenacity to want to produce something
meaningful that will live on. Second, it requires sufficient training, support, and
mentorship. And lastly and most importantly, it requires an audience who can
appreciate the meaning and beauty of the work. If you are reading this, then I was
lucky enough to find my audience.
I’d like to thank my alma mater, the University of Southern California’s
Annenberg School for Communication for providing such a diverse and open
community of scholars to learn from. Thanks goes out especially to: Tom Hollihan
for providing early guidance; Marita Sturken and Doug Thomas for their valuable
feedback in the development of my working paper on virtual community; Peter
Vorderer and Ute Ritterfield, for their early influence on my thinking about video
game theory; Sandra Ball-Rokeach for reminding me to think about all levels of
analysis; Steve Ross for introducing me to the historical antecedents of entertainment
as popular culture; Peter Monge for giving me my first taste of statistics; Rand
Wilcox for teaching me even more robust non-parametric methods; Larry Gross for
always believing in me; And of course, Anne-Marie Campian for shepherding me
through the bureaucracy.
I am grateful to Michael Cody for first introducing me to the health
communication literature. I remember our early discussions and how fascinating it
was to learn about entertainment education projects being developed around the
world. It was around that time my friend Ali Bryant mentioned that I might be

ii

interested in some of the research that Lynn Miller was doing on interactive
interventions. I stumbled onto one of the research teams early grant development
meetings and soon found myself involved in creating our very own entertainment
education research study here at Annenberg and the rest is, as they say, history.
Working on a research team over an extended period of time teaches you
many things: the value of collaboration, the courage to try new things, and the insight
that only comes from having a diversity of opinions. It also provides excellent
mentorship. To that end I’d like to thank Robert Appleby for his daily guidance,
friendship, and reminders to come up for air whenever it looked like I would sink into
the data. I am grateful to Stephen Read for always providing the voice of reason in
our marathon research meetings -that only a social-cognitive researcher could
provide. And last but certainly not least I’d like to thank Lynn Miller, who has truly
been an inspiring mentor and tireless dissertation chair. Anyone who works with her
will quickly come to realize what a remarkable person she is. I can’t thank her enough
for helping me through my many manuscript revisions. Being a researcher has many
exciting ‘aha’ moments, but there is also a lot of tough and boring work that precedes
them and I was fortunate to have an advisor who showed me how to appreciate both.
I am grateful to my friends who stuck with me through it all. I am grateful to
Jason Escalante, a true art connoisseur, and an outside influence. I am grateful to
Stephen Bloom and Jimmy Wiser, unique influences in their own ways. I am grateful
to John Christensen, a colleague and collaborator on things great and small. I’m
grateful to Elisia Cohen, Michelle Schumate, and Alison Bryant -three smart cookies
who always got me thinking.

iii

I’d like to thank my parents for valuing my academic pursuits and fostering
my love of learning –this dissertation is dedicated to you both. I am grateful to my
mom, Barbara Godoy, who was always there for me and a great teacher. I am grateful
to my dad, Ignacio Godoy, who I consider to be a research scientist at heart. He
always understood the value in sacrificing short term gains for long term goals.
Finally, thanks to my grandmothers: Victoria Godoy and Rosa Gonzalez Guerra
Reinhardt. They are two very special ladies who overcame a lot and never gave up on
the American dream for their children. What a joy it is to be able to share this special
moment with them and ‘you’ now.

iv

TABLE OF CONTENTS

ACKNOWLEDGMENTS…………………………………………………….........ii
LIST OF TABLES....………………………………………………………….......vii
LIST OF FIGURES……………………………………………………................viii
ABBREVIATIONS…..…..………………………………………………...….......ix
ABSTRACT.…………………………………………………………………..…...x
INTRODUCTION………………………..……………………………………….. 1
CHAPTER 1: PARALLELS BETWEEN VIRTUAL AND REAL BEHAVIOR.. 2
Entertainment and play…………………………………………………………… 4
Children’s play and real-life correspondence…………………………………...... 4
Virtual gaming environments and real-life correspondence……………………… 6
Ecological validity in experiments and interventions…………………………….. 9
Real life experimental studies and virtual-life experimental studies…………...… 9

CHAPTER 2: THEORIES OF LEARNING: KEY TO SOLVE………………... 12
Observational learning………………………………………………………….... 12
Social cognitive learning theory……………………………………………….… 15
Affect and engagement in learning…………………………………………….… 16
Interactive learning…………………………………………………………….… 21
Interactive interventions and risk-taking…………………………………….…... 24
H
1
: Interactive video choices will be related to past risk-taking behavior...…….. 27
H
2
: Interactive video choices will predict future risk-taking behavior….….......... 27

CHAPTER 3: THE THEORY OF PLANNED BEHAVIOR –THE CURRENT
‘GOLD STANDARD FOR ASSESSING RISK-TAKING COMPARED TO
VIRTUAL RISK-TAKING……….……………………………………………... 30
H
3
: Theory of planned behavior should, consistent with the literature,
significantly predict past risk-taking behavior at Time 1 (postdictive)……..…… 36
H
4
: Theory of planned behavior should consistent with the literature
significantly predict future risk-taking behavior (UAI) at Time 2 (predictive)..… 36
The best predictor of future risk-taking behavior is past risk-taking behavior…... 37

v

CHAPTER 4: SHORTCOMINGS OF PRIOR MEDIATIONAL MODELS IN
PREDICTING RISK-TAKING……..………….………………………………...40
H
5
: Past sexual risk-taking behavior should, consistent with the literature,
significantly predict future sexual risk-taking behavior……………...…………...44
H
6
: Interactive video choices will predict past risk-taking behavior (UAI) at
Time 1, better than the traditionally used Theory of planned behavior for the
IAV condition but not for the Yoked condition (postdictive)…………...………. 44
H
7
: Interactive video choices will predict past risk-taking behavior (UAI) at
Time 1, better than the traditionally used Theory of Planned behavior for the
IAV condition but not for the Yoked condition (predictive)………………..…….44

CHAPTER 5: TOWARDS A MORE ROBUST PREDICTIVE MODEL:
COMBINING BOTH EXPLICIT COGNITIVE MEASURES AND
UNOBTRUSIVE EXPERIENTIAL BEHAVIORAL MEASURES….…….…. 45
H
8
: The Theory of Planned Behavior coupled with virtual risk-taking will
account for a majority of the variance of future risk-taking mediating the effect
of past behavior in the IAV condition but not in the Yoked condition.…..……... 48

CHAPTER 6: METHOD………………………………………………………… 49
Procedure....……………………………………………………………………… 50
CHAPTER 7: RESULTS……………………………………………………….... 62
Post-hoc analysis: Self-Efficacy as suppressor variable…………………………. 95
H
9
: Suppressor effect of Self-Efficacy should disappear if the ‘over-confident’
group is excluded from the analysis……………………...………………………. 96

CHAPTER 8: DISCUSSION…………………………………………………..... 103
Fit with and advances beyond past research……………………………………... 105
What processes facilitate the link between virtual and real-life behavior?............ 107
Social presence and behavioral realism………………………………………….. 107
Making the self salient…………………………………………………………… 108
Situational awareness and Theory of Mind……………………………………… 110
How can these findings lead to better interventions?............................................. 112
Theory of planned behavior as a predictor of past and future behavior…………. 116
The best predictor of past behavior is virtual behavior…………………………... 117
Does tapping into embodied cognitions guide behavior?....................................... 117
Theory of planned behavior and virtual risk: equal predictors of future
behavior?..........…………………………………………….…………………….. 120
Interactivity matters in linking the behavior to the experience…………………... 121
Towards a mediational model of risk-taking…………………………………….. 125

CONCLUSION……………………………………………………………..……. 129
Limitations and avenues for further inquiry……………………………………... 131

BIBLIOGRAPHY………………………………………………………………... 134

vi

LIST OF TABLES

Table 1: Baseline: Items Comprising Intentions to use Condoms
Scale and Factor Loadings…………….……..…….……………………....…57

Table 2: Post-Intervention: Items Comprising Intentions to Use
Condoms Scale and Factor Loadings…………………………………..…......58

Table 3: Baseline: Items Comprising Self-Efficacy Scale and
Factor Loadings………………………………………………....………...….60

Table 4: Post-Intervention: Items Comprising Self-Efficacy
Scale and Factor Loadings………………………………..…………..............61

Table 5: Simultaneous Regression: Theory of Planned Behavior
Predicting Past Behavior (past 90 days)………………………………….…..65

Table 6: Simultaneous Regression: Theory of Planned Behavior
Predicting Future Behavior at Time 2 (next 90 days)………………………..68

Table 7: Simultaneous Regression: Past risk-taking behavior
Predicting Future risk-taking behavior……………………………………….72

Table 8: Simultaneous Regression: Virtual Risk Predicting
Past Behavior (past 90 days)…………………………………………………76

Table 9: Simultaneous Regression: Virtual Risk Predicting
Future Behavior (next 90 days)…................. ………………….………….....80

Table 10: Hierarchical Regression: IAV Choices & Theory of Planned
Behavior Predicting Future Behavior mediating Past Behavior……………..85

Table 11: Post-Hoc Hierarchical Regression Suppressor Effect
Demonstration: IAV Choices & Theory of Planned Behavior
Predicting to Future Behavior minus the “overconfident group”…………….91

Table 12: Post-hoc analysis- removing overconfident group increases
the predictive ability of virtual risk-taking…………………………………...99

Table 13: Summary of Results……………………………………………….102

vii

LIST OF FIGURES

FIGURE 1: Somatic Marker Hypothesis…....…….……………………....…16

FIGURE 2: Mediational Model of Risk-Taking…..…………………..…......127

viii

ABBREVIATIONS

ELM Elaboration Likelihood Model
IAV Interactive Video
CEST Cognitive Experiential Self Theory
MACS Motivation, Affect, & Cognition States
MSM Men who have sex with men
SCL Social Cognitive Learning
SOLVE Socially Optimized Learning in Virtual Environments
TPB Theory of Planned Behavior
TBDM Traditional Behavioral Decision-Making Models

ix

ABSTRACT
Would virtual behaviors be predictive of past and future behaviors? Does the
relationship between an individual’s virtual and subsequent real life behavior depend on
active interactive decision-making or is mere passive observation of another’s choices
enough to predict subsequent real-life behavior? Would virtual decisions (and
interventions tied to them) predict behavior 3 months subsequently above and beyond
traditional cognitive variables (e.g., intent, self-efficacy)? These questions can be
addressed by taking a SOLVE (Socially Optimized Learning in Virtual Environments)
approach. To address hypotheses related to these questions, 151 men who have sex with
men (MSM) filled out initial baseline measures of their past risk-taking behavior (e.g.,
use of alcohol, drugs, sexual history) and other measures (e.g., demographics, traditional
self-report predictors) and were then randomly assigned to an interactive video (IAV-
SOLVE) condition or a non-interactive yoked control condition (choices of another
MSM). MSM in the Interactive HIV Prevention Video condition made a range of
behavioral choices (e.g., drink alcohol, take drugs, or take sexual risks) on a virtual date.
The choices were electronically recorded. MSM assigned to the yoked condition
passively observed the choices that had been made by another MSM. Participants in both
conditions then answered immediate post-measures (e.g., traditional psychosocial health
risk predictor variables). After three months, participants filled out a follow-up survey
where risk-taking behavior was again re-assessed.
The first set of analyses (Study Set 1) involved only those MSM in the IAV
condition and addressed two hypotheses. Consistent with hypothesis 1, a series of Chi
square analyses showed that virtual risk-taking was significantly related to past behavior,

x

and consistent with hypothesis 2, it was predictive of future risk-taking. Study Set 2
examined the extent to which virtual decisions accounted for more variance in past or
future risk-behavior. To assess the role of interactivity (actually making choices), these
hypotheses involved examining these links for participants in both the IAV (where virtual
decisions were expected to predict behavior) and Yoked conditions (where simply
watching another’s responses were not expected to predict behavior). Simultaneous and
hierarchical multiple regression analyses revealed that the TPB significantly accounted
for past risk (consistent with H
3
) and significantly predicted future risk (consistent with
H
4
) in the IAV and Yoked conditions. The predictive role of past behavior in predicting
to future behavior (H
5
) was also confirmed. However, virtual risk-taking also accounted
for more unique variance in predicting future risk-taking behavior than the TPB did in the
IAV but not in the Yoked condition (consistent with H
6
& H
7
). In addition, (consistent
with H
8
) the two sets of constructs combined (virtual risk and the TPB) fully mediated
the relationship between past and future risk-taking behavior in the IAV but not in the
yoked condition. These findings suggest that virtual environments using a SOLVE
approach may allow us to unobtrusively identify those most at risk in order to tailor more
effective interactive interventions to reduce future risk-taking.

xi

INTRODUCTION
The impetus for this research began with four broad sets of questions.
1

First, what can choices made within a virtual environment -based on scenarios
likely to be encountered in the real world- tell us about past and future real-life
decision-making and behavior? Second, are we are able to account for significant
variance in future real-life risk behaviors using traditional cognitive measures found
in the Theory of Planned Behavior (e.g. intentions, self-efficacy)? Third, do virtual
decisions account for unique variance in real-life risk-taking beyond traditional
more cognitive self-report predictors (e.g., intent, self-efficacy)? And finally, might
these traditional and virtual predictors in combination mediate the variance
traditionally associated with “past behavior” when individuals actively make
interactive choices instead of passively observing another’s choices? Following an
extensive review of the literature as delineated below eight specific research
hypotheses were derived.

1

1
The research described was supported by Grant Number 1 R01AI052756-01A1 from the
NIAID to Lynn Carol Miller (PI), P. Robert Appleby (Co-PI) and Stephen J. Read (Co-PI). Its
contents are solely the responsibility of the author and do not necessarily represent the official
views of the NIAID.

CHAPTER 1:
PARALLELS BETWEEN VIRTUAL AND REAL BEHAVIOR
“The contemporary theory of communication removes all doubts –we use media not only
for the playing of video games but also during many of our life activities. We cannot, or rather
choose not to, live without television, telephones, and email anymore….That is why the
dissemination of new ways of thinking which makes the real and the virtual worlds equal, is only
a matter of time.”
(Filiciak, 2003, p.101)

Virtual behavior is behavior guided by a real person within a representational
space. It can be performed within environments that are as complicated as a three
dimensional haptic holodeck or as simple as a ‘choose your own adventure’ book.
The key component for virtual behavior is the ability of the representational space to
allow interactivity. “Interactivity generally refers to the processes of communication
that take on some of the characteristics of interpersonal communication” (DeFleur &
Ball-Rokeach, 1989, p.341) and is often “a measure of a media’s potential ability to
let the user exert an influence on the content and/or form of the mediated
communication” (Jensen, 1998, p.201). Virtual behavior could thus be said to be a
kind of simulated behavior in which humans interact with mediated representations of
real humans (e.g. avatars, movie characters) or computer agents that are not human,
such that their choices and behaviors impact what happens next.
How would you respond if you were asked to engage in a virtual interactive
experience? Would you engage in “parallel play,” exhibiting a pattern of behavior
paralleling what you would do in real-life? Or, would you –in the safety of a virtual

2

environment, experiment with novel and risky behaviors-- engage in “divergent play”
you wouldn’t normally choose in real life? Or, would your virtual behavior be
completely unrelated to any of your real-life behavior? From a communication
perspective, whether individuals’ virtual behavior relates to their real-life behavior,
and if so, whether they engage in “parallel” or “divergent” play has potential health
intervention implications. If individuals generally exhibit a “parallel play” response to
virtual environments that are designed to approximate real life, then virtual
environments might be used as an unobtrusive measure of what a participant is willing
to do or would do in a given context, thereby predicting future behavior in the real
world when participants are presented with a similar challenge or threat.
Unfortunately, little research has investigated how individuals’ decisions and
behaviors made in a simulated or virtual environment map onto, and/or are diagnostic
of, their past or future real-world choices. Nevertheless, below we consider research
in two domains, entertainment and experimental contexts that are most relevant to the
links between virtual/simulated behavior and real-life behavior. Where research is
available, it generally supports the link between simulated/virtual behavior and similar
behavior in everyday life.

3

Entertainment and Play
Why might we think that how people respond in one setting would be
consistent with how they would respond when faced with a similar setting?
Personality psychologists have long argued that adults tend to possess and exhibit
clearly established, experientially-based, preferred behavioral patterns which remain
stable over time (Costa & McCrae, 1994). The behavior preferences and responses
that individuals have to situations can vary within a population yet be singularly
enduring (Mroczek, Almeida, Spiro, & Pafford, 2006). These enduring preferences do
not emerge sui generis, rather they develop through interaction early on, at the most
basic of levels, children’s play.

Children’s play and real-life correspondence. Researchers theorize that the
existence of play behavior in animals may be linked to the evolution of higher
intelligence -meaning that only those animals which rely on social interaction and
possess a wide range of loco motor skills, require non-functional rehearsal skills to
master their full repertoire of social behaviors (Bekoff & Byers, 1998). Thus it is no
wonder that only mammals and birds play (Brittanica, 2007). These animal species
exhibit high intelligence and tend to have a longer length of juvenile development
which has also been linked to the propensity for play. Juvenile play serves as practice
for adult behavior. In fact, “youth probably exists for the sake of play” (Groos, 2005,
p.75).

4

Play can be considered a form of simulated experience that carries functional
significance in terms of learning how to relate in the real world. In the developmental
literature, researchers have examined doll-play as an early indicator of children’s real
world home environments. Studies demonstrate that doll-play by children can be
diagnostic of many types of behavioral patterns. For example, how a child plays with
a doll can demonstrate proclivities towards violence as well as indicate patterns of
sexual abuse (e.g. the child re-enacts past experiences) (Everson & Boat, 1990). In
addition, it has been found that the higher the realism of the dolls and settings (e.g.
furniture, mock houses, etc…) the more likely you are to elicit play behavior that is
concordant with real life (Phillips, 1945).
This pattern for highly realistic dolls and settings, compared to low realistic
dolls and settings, has been found for both with short term and long term play
(Phillips, 1945). This finding is consistent with the long held similarity-inhibition
perspective, which holds that in play, “fantasy is inhibited to the degree that the play
setting contains features drawn from real life” (Fein, 1981, p. 1108).. Thus
playfulness can be said to emerge organically out of the media form itself (Gadamer,
1989) requiring first a ‘game field’ of thought from which heuristic rules can emerge
and upon which the ‘give’ or playful moment of movement --based on the bounds of
the self imposed structure of the given system-- can occur (Derrida, 1967).
This can be thought of as a ‘region of proximal learning.’ In the developmental
literature on learning and exploratory behavior Metcalfe and Kornell (2005) found that
people are more likely to explore that which they almost know but have not yet

5

‘mastered’ in some respect (i.e. the low hanging fruit). This mastery approach is
consistent with psychoanalytic theory in which play is thought of as a mechanism
whereby children can cope with specific sources of real-life tension. For example,
experimentally induced stress and/or aggression leads to subsequent increases in
aggressive content of play. Hospitalized children tend to prefer play with health
anxiety type toys or games (Brown, et al., 1997; Lieberman, 1999, 2000; Gilmore,
1966).
In addition, the existence of real-life tensions tends to reduce levels of
‘fantastic’ play but rather induces this latter form of ‘mastery’ play (Fineman, 1962,
Gould, 1972). This mastery approach is more likely to be dependent upon
interpersonal processes as children move towards young adulthood. This is because
play tends to develop societal integrative role structures such that “the activity of each
player is tied to the activity of the others and the roles are constituted through the
appropriate reciprocity of the players” (Fein, 1981, p. 1102). Thus in a sample of
young adults who have not yet mastered a specific behavior (i.e. consistent safe-sex
behavior) you would expect their play to be socially integrative, based on real-life
experiences, and reflect the degree to which anxiety is provoked and mastery desired.

Virtual gaming environments and real-life correspondence. Research that has
examined the relationship between the virtual and the real in gaming contexts has not
focused on actual real world behavior. Instead research has focused either on stable
personality characteristics or static social characteristics which may be manifested

6

similarly within the virtual versus the real world. Griebel (2006), in his study of how
people play the Sims2 (Electronic Arts, 2003) video game, found that real life
personality characteristics of players such as neuroticism, openness to experience, and
conscientiousness were correlated with how they made their Sim characters behave.
Sim game players were found to create Sim family environments that mirrored their
own real life. For example, “(p)articipants who came from divorced households were
more likely to make their Sims experience a divorce; 50% of participants from
divorced households chose to make their Sims get divorced, while 0% of participants
who came from married households made their Sims get divorced” (Griebel, 2006, p.
10).
Yee (2007) found similar results when he looked at how real life males
behaved when playing in the virtual world game known as Second Life (Linden Labs,
2005). Male players tended to have their characters stand further away when talking
to other males in the virtual world and were less likely to keep eye contact --behaviors
that are similar to the way that males, in general, behave in the real world. Yee and
Bailenson (2007) also found that people will tend to show behavioral responses that
reflect their avatar. For example, giving someone a taller avatar to represent them
made them more likely to bargain more aggressively than individuals who were
represented by shorter avatars. Also, if given an extremely attractive avatar, a person
would become friendlier and more revealing about themselves. If people conform to
the stereotypes of the virtual bodies given to them, then it appears that the underlying

7

social norms that govern real life are capable of being transferred subconsciously into
those interactive virtual worlds.
Whang and Geunyoung (2004) examined the relationship between real world
lifestyles and virtual lifestyles among a sample of 4,786 online players of Lineage
(NCsoft, 1998) (at the time it was the largest online game where people could assume
identities and play various roles—the game accommodated over 6 million users
worldwide). Upon classification of their real-world lifestyles, their tendencies and
desires were compared to lifestyles in the virtual world. They found that for those
game players who have developed their own distinctive lifestyles (e.g., communality,
independence, striving for social status) in the real world, these lifestyles were also a
strong predictor in explaining behavior patterns and desires in the virtual world. The
differences were mirrored not only with regard to their lifestyles but also with regard
to their socio-economic status within the virtual world, in which, for example, people
of high wealth in the real world attempted to also continue wealth seeking in the
virtual world.
Apart from the game context, other studies have documented that people
oftentimes use new technology as an extension of themselves if given the opportunity.
For example, Wellman, Haase, Witte, and Hampton (2001) found that, contrary to the
prevailing opinion at the time that heavy Internet use would lead to social isolation,
Internet use actually served as an amplifier, allowing already sociable people to be
even more social in real life and vice versa for introverted people, enabling fewer face-
to-face contacts (see also Rheingold, 1998). Thus, there is evidence to suggest that

8

people utilize representational spaces in a similar manner to their real world spaces
regardless of the type of interactive medium.

Ecological Validity In Experiments And Interventions

Real-life” experimental studies and “Virtual-life” experimental studies.
There has also been some work that seeks to replicate real-life experimental studies in
virtual environments. Such work indicates that humans respond similarly to
experiments whether they are with real humans or with virtual others (Blascovich, et
al., 2002). Work by Blascovich and others suggests that at least under some conditions
behavior in classic social psychological studies in the “lab” can be replicated with
“virtual” experiments. For example, using immersive virtual reality technology,
Blascovich and colleagues were able to replicate a study that examined betting norms
based on social desirability/conformity theory.
They created an immersive virtual casino complete with a blackjack table,
gaming chips, seats, slot machines, cards, a dealing shoe, a dealer, etc… Participants
adjusted their bets in conformity with the virtual agents just as they did with human
confederates back when the original study was conducted in 1974 (Blascovich &
Ginsburg, 1974). A significant main effect for betting norm condition and significant
effects of the a priori comparisons of the low, to same, to high, betting norm
conditions, demonstrated the predicted conformity effect in both the virtual and the
real world lab experimental settings.

9

Slater et al. (2006) replicated the famous Milgram study (Milgram, 1963) --
showing that people would administer lethal electric shocks to a stranger at the behest
of an authority figure-- utilizing a virtual environment. Participants responded to the
social situation as if it were real, despite knowing that no real events were taking place
(merely digital representations). The researchers concluded “that humans tend to
respond realistically at subjective, physiological, and behavioral levels in interaction
with virtual characters notwithstanding their cognitive certainty that they are not real”
(Slater, et al., 2006, p. 5).
But, are laboratory studies also simulations? While it is reassuring that
participants respond similarly to real and virtual others in laboratory studies, it is
unclear if lab experiments are really tapping into “real-life” or if they, themselves,
albeit realistic, are simply other examples of simulations of “real-life”? That is, do
studies seeking to replicate “lab studies” in virtual environments demonstrate a link
between virtual simulation of behavior and real-life behavior or a link between one
type of behavioral simulation (non-mediated) and another (mediated)?
Psychologists have been observing the behavioral choices individuals make in
various non-media ‘virtual environments’ for some time (e.g., creating experimental
environments using confederates). Indeed, some of the most fascinating work in
social psychology has been based primarily on observing the ways individuals act
given various types of situations (Zimbardo, 1988; Cialdini, 2001). These many
studies presuppose that behavior conducted in an experimentally ‘constructed’ setting
is generalizable to individuals’ everyday real world behavior.

10

It is difficult to know for sure if subjects are behaving differently in the lab
setting from how they would normally behave in everyday life. This, of course, is the
generally acknowledged problem of “ecological validity” (Chaytor & Edgecombe,
2004). Unfortunately, social psychologists rarely set up their experiments to record a
series of decisions that directly map onto the decisions individuals (have previously)
or will subsequently make in everyday life in order to examine how behaviors in
experimental settings map onto real-life decisions by those same individuals in their
real-lives.

11

CHAPTER 2:

THEORIES OF LEARNING: KEY TO
BEHAVIOR CHANGE IN SOLVE

When individuals change their patterns of decision-making and behavior, they
have learned from the experience. There are many kinds of learning (and theories
about those forms of learning) and it is possible to incorporate them in virtual
environments to socially optimize learning. That is the underlying goal of Socially
Optimized Learning in Virtual Environments or a SOLVE approach (Miller et al.,
2005; Read et al., 2006, Universitywide AIDS Research Program of the University of
California, Grant R96-USC-160).

Observational (social) learning theory
Observational (social) learning theory is a kind of learning in which individuals
observe, remember, and sometimes imitate a behavior that is positively promoted –
usually by a peer mentor or authority figure. A classic example of observational
learning can be found in Bandura’s Bobo doll experiment, in which a child observes an
adult hit a Bobo blow up doll and then the adult leaves the room. When left alone, the
child will imitate the behavior, thinking that it is socially desirable to hit the doll after
seeing an adult authority figure engaging in such behavior (Bandura, 1961). As
reviewed by Rosenthal and Zimmerman (1978), the extent to which individuals change
their behavior (learn) through passive observation alone has been found to be the result
of both observing positive consequences (vicarious reinforcement) and observing
negative consequences (vicarious punishment) for the model’s behavior.

12

Past research on observational learning suggests that a variety of principles can
guide the circumstances under which individuals will change their behavior, however
learning new behaviors does not necessarily always translate into future performance of
the behavior. For example, learning a deviant behavior, such as ‘seeing’ a robbery take
place, in the observational learning context does not necessarily lead to enactment of
the behavior (Burgess & Akers, 1966). Rather, what matters is the degree to which
there is social scaffolding (i.e. a social push or pull that follows from the observed
behavior). The degree of reinforcement, either extrinsic from outside social norms or
intrinsic based on hedonic value, will have a direct effect on whether the observational
learning is enacted in the future. This effect is further tempered by the constraining
forces that either authority figures or outside penalties place on the enactment of such
behaviors.
Sabido in his work on telenovelas demonstrated that in order to get people to
enact observed behaviors, it is critical to show others being rewarded or punished for
their behaviors within the narrative (Singhal, Rogers, & Brown, 1993). He developed a
methodology which highlights storylines that emphasize a natural confrontation
between good role models and bad role models. Drawing on Bandura’s Social
Learning Theory, Sabido designed a narrative structure in which three types of
characters are presented: 1) A group of characters who support the desired behavior
(good role models); 2) A group who rejects the behavior (bad role models); and 3) A
group who lies somewhere in the middle (doubters). Over the course of the narrative
the doubters begin to move, one at a time, toward the side of the good role models, and

13

any doubters that remain at the end of the narrative are seriously punished (tragedy
ensues), making it clear that decisions ‘must’ be made. The bad role models undergo
bad consequences immediately after bad behavior is acted upon. At the end there is an
epilogue which sums up the storyline and provides a reinforcing explanation for what
was just witnessed. A number of telenovelas developed by Sabido, such as
Simplamente Maria and Ven Conmigo have demonstrated the ability of engaging
narratives to heighten awareness leading to potential increases in intentions and self-
efficacy towards promoted behaviors (Singhal, Cody, Rogers, Sabido, 2003).
Observational learning is an important type of learning that should occur when
media is designed as suggested above. Under this framework learning is expected to
occur whether participants are engaged in an interactive narrative or merely observing
the action of other participants in a linear fashion. A large literature has examined how
media can passively influence individuals (e.g., Oettingen, et al., 2006; Gerbner, Gross,
Morgan & Signorelli, 1986) much in line with this observational theory. Other research
however, suggests that interactivity (e.g., in interpersonal and media interactions) may
be used even more advantageously to induce behavior change in individuals (e.g.,
enhance the transfer of learning, reinforce implementation intentions, increase self-
efficacy, teach behavioral and negotiation rehearsal skills, provide applied knowledge)
(Brown, 1997; Brug, 1999; Davies, 2004; Gollwitzer, 1990; Grodal, 2000; Read et al.,
2006; Vorderer, 2000; Vandelanotte, .et .al, 2005).

14

Social Cognitive Learning Theory
Vygotsky’s (1962) social cognitive learning theory (SCL) posits that inducing
behavior change can greatly be enhanced by framing the meaning of the behavior (i.e.
utilizing the social/cultural context in which a guide, whether a parent, teacher, or peer,
can “frame” the meaning of what is happening in the problem-solving situation).
Vygotsky’s social cognitive learning theory has been applied to thinking about
children’s development and the process by which adults scaffold the learning of
children. However, typically the process of framing the meaning of the ongoing
narrative (in for example a telenovela) is rarely done: Sabido’s work is the exception in
the development of prosocial programming.
A key component of Vgotsky’s SCL theory is the “zone of proximal
development” -that is individuals may only be able to perform some behavior (avoid
risk in a given context) with the aid of an interventionist guide initially, and over time,
this pattern of desired response in a given cultural context is eventually internalized
(Kim & Baylor, 2006). The effect of scaffolding in response to the behaviors of the
learner may depend upon the learner’s ‘actual’ behavior, therefore the SCL theory
requires the learner to be engaged in social interaction with the guide (not merely to
passively observe such an interaction). Nevertheless, although this is often difficult in
real life, it is possible in virtual environments to insure that a guide “pops up” to guide
participant decision-making to scaffold safer choices at the “zone of proximal
development.” In fact, in their interactive video work to change social behaviors,

15

Miller, Read, and Appleby (Appleby, et al., 2007; Miller & Read, 2005; Read et al.,
2006) included such virtual peer guides to scaffold behavior change.

Affect and Engagement in Learning
Everyday we make a host of decisions, most of the time automatically without
much deliberation. Compelling evidence suggests that affect plays a critical role in
judgment and decision-making processes that may interact with how the brain encodes
experiential learning (Damasio, 2000; Panksepp, 1998; Rolls, 1999). A dominant
perspective is the somatic marker FIGURE 1: SOMATIC MARKER HYPOTHESIS
hypothesis, which posits that
emotions and feelings are types of
somatic responses that mark a
situation as good or bad (Bechara,
Damasio, & Damasio, 2000;
Damasio, 2000; Damasio, Everitt,
& Bishop, 1996). When the
situation recurs, these affective markers assist decision-making under circumstances of
conflict or uncertainty and may operate either consciously or non-consciously. That is,
anticipatory emotions serve a socially adaptive function insofar as they may
automatically elicit advantageous behavioral responses in social situations (Frijda,
1986; Keltner & Gross, 1999; Levenson, 1994; Oatley & Johnson-Laird, 1996;
Plutchik, 1979).

16

In fact, a series of studies suggests that the brain is wired to do so ----responding
to stimuli and situations in the environment with positive and negative emotions in
order to automatically guide decision-making (e.g. approaching when we are
experiencing positive emotions in response to stimuli, and avoiding when we
experience negative emotions) (Damasio, 1999). The emotions that we experience in
response to stimuli are often the result of past experiential decisions (Bechara, 2000).
Current decisions in novel situations may also come to be associated with
positive or negative emotions, which may alter how we subsequently respond to new
experiences in the future under similar circumstances. Reyna and Farley (2006)
suggest that many risky decisions are affectively based and made fairly automatically.
It takes experience, which young adults often lack, to associate negative outcomes
with risky decisions. The formation of this causal link between risky behavior and
negative affect however, may depend on the decisions being affirmatively made
(making your own mistakes) and the subsequent negative affect being associated with
learning ‘your’ lesson.
There are a multitude of studies that demonstrate how the brain uses
experience to draw associative connections between ‘the feeling of what happened in
the last similar context’ when a decision was made, and how to proceed in the present,
given a past decision-making affective state (Damasio, 1999). The brain has evolved
to make these quick, automatic decisions in everyday life. These distinctions are in
the generalist sense tied to good and bad --with the left hemisphere being largely
associated with positive emotions and the right hemisphere for negative emotions.

17

In terms of experimentally testing these affective states it is important to note
that the human brain does not draw a distinction between “media” and “real life” in
terms of which regions are activated automatically (Reeves & Nass, 1996). These
differences relate to both the experience of good/bad feelings, as well as the
recognition of good/bad feelings (Reeves & Nass, 1996, p.113). What is important
about this lack of an experience/recognition distinction lies in how memory itself
functions when preceding positive or negative emotions as an individual is responding
to stimuli in a decision-making context.
Decety et .al (1997) found that meaningful actions strongly lit up the left
hemisphere, when subjects were asked to think about imitating a certain action
(presumably a positive one –drinking tea), but this was not the case when a passive
observation was shown (simply a tea set with no action). Active decision-making or
even priming with the intent to make a decision is associated with positive and
negative emotions, thus giving priority to these embodied cognitions that are then re-
enacted with little thought under similar circumstances.
We know from an extensive literature that negative emotions play a critical
role in how long-term memories are encoded and stored (Lang, 1995; Lang &
Friestad, 1993; Fridja, 1998)). Negative emotions not only induce deeper processing
but they also have the effect of causing proactive enhancement --experiences that
come after negative events are remembered better than after positive events-- and
retroactive interference --previously recalled information is immediately lost in order
to attend to whatever caused the negative experience (Thorson & Reeves, 1985).

18

Because “the human mind does not make distinctions between real world
negative experiences or virtual negative experiences” (Reeves & Nass, 1998, p. 123)
it would be safe to assume young adults who have learned to associate negative
emotions with risk-taking, may have the ability to steer themselves away from a
potentially risky arousing decision-making point within an interactive environment,
whereas a young adult who associates risk with positive emotions will likely be
steered towards a risky path, regardless of ‘good’ intentions to be safe. Interactivity
matters because it can both assess the current choice state and intervene at the point
risk-taking occurs, perhaps recalibrating the emotional set point that may trigger the
behavior.
Williams (2006) found evidence for this ability when he showed that it is
possible to cultivate virtual perceptions of specific risky situations using virtual
interactive environments, which will then carry forward into similar real world risk-
perceptions within similar specific types of situations. However, the important thing to
note is that these perceptions of dangers only corresponded to events and situations
found in a similar context to that shown in the game world, not to other real-world
crimes. This means that it is not necessarily the specific act that induces risk-
perception but the situational context in which it is likely that a specific act might
occur which has already been similarly encountered within the virtual environment.
This may also explain how novelty plays out in interactive environments.
Turkle (1995) suggests that individuals are willing to explore their identity by
adopting role-play characters that are unlike themselves in the real world. She

19

performed an ethnographic study on text based virtual worlds, such as LambdaMoo, in
which she argued that when individual identity is viewed as a unified whole, “it (is)
easy to recognize and censure deviation from the norm, (but considering multiplicity)
it is easier to recognize diversity and to accept an array of ours’ and others’, making it
easier to play at being someone else (Turkle, 1995, p. 85).
Although this may explain novelty seeking for behaviors or environments ‘not
yet encountered’ in virtual spaces, it does not explain what would happen if these
textual environments were instead richly furbished and visual enough to activate prior
embodied cognitions (i.e. emotionally, experientially based cognitions). That is, if the
environment was not totally alien but rather similar enough to real life and your virtual
character was said to be ‘the real you’ then those embodied cognitions may then be
triggered to actively restrain or guide behavior away from the novel towards the more
predictable.
As shown earlier in the doll role-play example, the closer to reality the props
and environment are, the less likely you are to deviate from reality. The point of the
matter however, is that individuals rarely encounter richly textured interactive
scenarios or situations which are designed to simulate real life. Thus you would only
expect novelty seeking in situations where a person has not encountered the situation
before. This might particularly be the case for young sensation seekers who may not
have yet had the opportunity to engage in certain risks (i.e. drug use). Quite a bit of
research has shown that these young high sensation seekers are the most likely to seek

20

new and exciting experiences and sensations (Donohew, 2000; Palmgreen et al., 1991;
Zimmerman, 1997).
Thus it is important to consider that virtual risk-taking could be diagnostic of
future real life risk over time among this group of young MSM, if they have never
engaged in this type of behavior before (e.g. UAI). Reyna and Farley (2006) have
shown that the absence of prior risk-taking experiences is one of the major factors that
distinguishes risk-taking behavior between adults and adolescents. We know that
adults tend to take less risk than adolescents. In most cases they do so, not out of any
deliberative processing, but rather because they make decisions intuitively, responding
to the gist cues (Reyna & Farley, 2006) that –via experience—they have learned to
associate with risk.

Interactive Learning

In a meta-analysis that examined the influence of active and passive
interventions, Albarracin et al., (2005) found that active interventions (e.g. those that
included role playing problem situations) were associated with higher increases in
condom use above passive approaches. By allowing participants to gain ‘practice’ with
behavioral self-management strategies, active interventions provided consistently
stronger increases in condom use compared to passive interventions. However, in role
playing, individuals don’t often have the opportunity to make risky choices and observe
the consequences of those actions. This is critical because we know that there is a more
automatic route to decision-making which relies on ‘gist’ cues and non-conscious

21

biases that automatically guide behavior. Negative consequences (i.e. emotions, guilt,
or physical harm) that result from risky decisions help to hone experiential biases in
order to steer future behavior away from risk-taking. With socially optimized learning
in virtual environments (SOLVE), it becomes possible to effectively link risky affect
cues to active decision-making processes in real-time with the participant.
In a prior longitudinal design the SOLVE model was tested among young MSM
in a RCT (IAV condition compared to those in a standard of care control group) and it
was found that those in the IAV condition had lower levels of unprotected anal
intercourse (UAI) 90 days post intervention (Appleby, et al., in press; Miller & Read,
2006; Read et al., 2006). In a more recent study, SOLVE-IAV, on an NIAID RO1 was
also found to reduce UAI over time for specific targeted risk populations (e.g, African-
American, Latino, and Caucasian MSM). SOLVE is effective because it is a
personalized intervention: Each MSM’s virtual experience is based on his decisions (to
be risky or not). SOLVE incorporates traditional more cognitive elements of behavioral
interventions that research reviews (e.g., Kelly, 1995) suggest mediate HIV risk for
MSM while simultaneously also addressing the more reactive and more affect-based
route to decision-making (Reyna & Farley, 2006). In SOLVE-IAV, MSM assume the
role of a character on a “virtual date” within an interactive video (IAV) and “make the
decisions” (e.g., anal sex or not; condom use or not). The guides’ messages, challenge
risky often more automatic behavior and they may provoke negative emotions, and
precede a “second chance” to make a safer choice.

22

Albarracin et al.. (2005) found that threat-inducing arguments have no positive
influence under any conditions (either passive or active) on increasing condom use.
This is consistent with Rothman and Solovey’s (1997) model of framing, in which
‘gain’ frames are necessary to instill proactive behavior, whereas a ‘loss’ frame (fear
inducing) is necessary to avoid a certain behavior. This is one of the short-comings of
prior types of interactive interventions, because the emphasis is often placed on
roleplaying positive behaviors (e.g. using a condom) without accounting for the
possibility that chronic risk-taking is sometimes related to a conscious decision to be
unsafe (which doesn’t allow for the inducement of a loss frame at decision
points/exposing the negative consequences of those actions) and sometime related to an
unconscious decision that leads to unsafe behavior (e.g., becoming highly sexually
aroused). The SOLVE model is capable of matching the frame with the behavior based
on the interactive choices the participant makes.
This makes it highly likely that an interventions affect framing will ‘fit’ with the
participants’ behavior (i.e. if they decide, consciously or not, to be risky, a guide will
pop up inducing negative affect and if they decide to use a condom, a guide will reward
them with praise inducing positive affect or by not interrupting the rewarding flow of
the interaction leading to sex). Higgins (2000) suggests that if the motivation for the
behavior matches the motivation of the message given, then the tailored intervention
will be more persuasive.

23

SOLVE incorporates the above features. Unlike with traditional linear narrative,
the SOLVE-IAV model allows for a much more agentic role in the decision-making
process whereby participants in a self-guided narrative are much more likely to feel like
they are there (Biocca, 2002). SOLVE makes it possible to create more realistic
scenarios that are both emotionally and visually engaging such that the situational cues
(just like what would be encountered in real life) will be associated with risk-reduction
behavior. In other words, choices associated with situated cues “(e.g. when I encounter
this cue or activate this emotion [e.g. sexual arousal], I also activate the goal of
protecting myself))” (Read, 2006, p. 6) will lead to reduced risk-taking, because under
past experience a virtual guide intervened to induce a negative consequence (e.g. affect)
at a risky choice within a prior virtual experience (Miller, 1993). SOLVE helps in the
formation of these implementation intentions (i.e. ‘if, then rules’ ) through responsive
scaffolding and framing by virtual peer guides such that learning is enhanced in ways
that blur the boundary between actual experience and virtual experience, whereby
virtual outcomes result in less-risky future real outcomes.
Interactive Interventions And Risk-Taking. Although Vygotsky’s theory
emphasizes the social interaction between the learner and the guide in scaffolding
learning (and the construction of meaning) the learner him or herself does not typically
actively engage in making decisions. This is of course the case as well in observational
learning. As in real life, however, when individuals make decisions in an interactive
environment, they are simultaneously observing behavior AND using their own desires
and intentions to guide their choices.

24

Numerous intervention studies in the HIV/AIDS prevention domain are
designed to assess and change risky sexual behaviors through the use of role plays
(Butler, Miezitis, & Cole, 1980; Magura, Kang, & Shapiro, 1994; Kelly, 1989;
DiClemente & Wingood, 1995; Jemmot & Jemmot, 1992; Cohen, Dent &
MacKinnon, 1991; Des Jarlais, Casriel, Friedman, & Rosenblum, 1992). In these role
plays participants are asked to engage an interventionist (playing themselves or a
potential other partner) as they learn to negotiate safer sex behavior. Thus it is
hypothetically possible that data regarding the links between participants’ decision-
making during simulated role playing and prior or subsequent real-life behavior might
be examined. However, from the published studies, it is not clear if these interactions
were recorded, and if so, ever coded for participant decision-making. In any event, we
know of no published work that has directly linked individual decision-making in such
role-played contexts (which could constitute simulated behavior) to a participant’s
prior or subsequent “real-life” behavior as reported on baseline or follow-up measures.
Instead, the literature we do have (e.g., Diclemente, et al., 2004; Downs, J.S.,
.et al., 2004; Metzler, C.W., .et al., 2000; Scholes, D., .et al., 2003) considers
participants who are randomly assigned to (and therefore exposed to) one type of
intervention versus another and whether those groups differ in subsequent condom
use. Generally, researchers do not, more specifically, examine participants’ decisions
within an intervention and correlate those decisions with individuals’ current
perceptions (e.g., intent to use condoms) or future behavior (e.g., subsequent condom
use). In fact, we are unaware of any work that has systematically examined

25

individuals’ sexual risk-taking decisions in the lab and related them to similar
decisions those individuals have made in past or future similar decision-making
contexts. If decision-making is akin to real life than this would have powerful
implications for how interventions might be designed in the future.
As the above review underscores, there is very little literature indicating how
decisions in virtual environments yield behavioral patterns similar to those in real-life
similar decision-making contexts. This is especially the case if laboratory experiments
are excluded from the definition of “real-life.” In those rare cases where people’s
choices have been examined, the virtual environments used (e.g., violent video games;
phobia reduction environments) are often not similar to those the user would
encounter in real-life (see for example, Carlin, Hoffman & Weghorst, 1997;
Rothbaum, Hodges, Anderson, Price & Smith, 2002).
From the prior discussion on interactive learning, it is clear that virtual
environments may be capable of activating the more automatic affective based
decision-making that may be mostly responsibly for risk-taking among adolescents. If
there is a relationship between individuals’ virtual behaviors and their future real-life
behaviors, then this would provide good evidence for the potential of interactive
virtual environments in affording unobtrusive measures of real-life preferences and
behavioral inclinations.
Although theoretically sound, there is currently no clear consensus about
whether decision-making in realistic virtual environments will map onto or depart
from decision-making in real-life contexts. This is not surprising, considering that

26

rarely do we assess decision-making by the same individuals in both real and virtual
contexts over time. Because there currently is no clear consensus on the issue the
following grand research question is proposed:

If people are allowed to make their own choices in a virtual environment that
is similar to their everyday environment, will their choices in the virtual world
predict their behavior in the real world, and will their behavior in the real
world predict to their choices in the virtual world?

Since formative research was conducted to assess the sexual scripts that MSM
encounter on a ‘real life date’ (Godoy et al., 2005), it became possible to create a set
of realistic interactive videos that afforded a range of choices that might map onto real
life behavior. This allowed for the testing of two hypotheses regarding the mapping of
behavior between the virtual and the real:
H
1
: Interactive Video Choices will be related to past risk-taking behavior
H
2
: Interactive Video Choices will predict future risk-taking behavior
This possible connection between the virtual and the real suggests that
interactive choice-making may allow us to assess risk-taking behavior which could in
turn, change the way health interventions are currently designed. There is literature on
the distinction between learning through enactment of behavior and passive learning.
According to Deci and Ryan (1984), learning through enactment provides intrinsic
motivation, which leads to better performance of complex activities such as
conceptual learning that is then later applied to future situations (McGraw, 1978).
Extrinsic motivation on the other hand is associated with traditional models of

27

learning, such that individuals remember the facts of the situation but are less likely to
interpret or integrate them in future situations (Benware & Deci, 1984).
These motivational processes appear linked to the distinctions discussed earlier
concerning how interactivity may influence how the brain processes, encodes, and
activates memories depending upon whether they are associated with the intent to
imitate or the intent to observe. Affectively tagged memories coupled with stronger
conceptual learning principles made possible through an interactive environment
should provide a far greater measure of future performance mastery than simply
passively receiving a linear yoked media message could. In fact, a number of
interventions in a wide variety of domains have consistently demonstrated the mastery
performance of behaviors learned from virtual environments as well as the ecological
validity of such results (Johnson, et al., 2004; Rizzo, et al., 2004).
Gilbert (1991) found that individuals tend to become so absorbed in health
simulations that they forget the outcomes are defined by the intervention. In other
words, because their minds are so busy processing other aspects of the experience
their natural inclination is to assume the simulation is accurate and therefore are likely
to adopt a ‘go with the flow’ attitude. This is consistent with the theoretical viewpoint
that in real-life adults either engage in risks or resist taking risks not out of any
conscious deliberation or choice but because “they either are able to, or fail to,
intuitively grasp the ‘gist’ of risky situations, retrieve appropriate risk-avoidant values,
and never proceed down the slippery slope of actually contemplating tradeoffs
between risks and benefits (Reyna & Farley, 2006, p. 7).” This non-conscious

28

deliberation, or ‘gist’ framework, has never been evaluated from an interactive media
perspective nor has it been held up to scrutiny in any quantitative measure against
other more widely accepted theories used to assess future risk-taking such as the
Theory of Planned Behavior.

29

CHAPTER 3:

THE THEORY OF PLANNED BEHAVIOR –THE CURRENT GOLD STANDARD
FOR ASSESSING RISK-TAKING COMPARED TO VIRTUAL RISK-TAKING

The Theory of Planned Behavior (TPB) has long been the method of choice to
predict risk-taking behavior (Ajzen & Fishbein, 1985; Albarracin, 2001; Armitage,
2005). The theory holds that individuals are likely to behave in a rational manner and
that risk-taking is a matter of active, perceived volitional control. Self-professed
intentions to behave in a certain way, coupled with self-efficacy, are thus thought to be
the best indicator of how that person will actually behave in the future. Given the role
embodied cognition may play in guiding behavior, the major limitation of this theory
is that it fails to take into account a myriad of other individual, contextual, or
situational influences that may influence risk-taking behavior beyond just having good
intentions. For example, in the context of sexual risk-taking, the reliance on
individual level intention measures may fail to take into account the role that potential
partners may play in encouraging sexual risk-taking –i.e. a willingness to go along- or
whether impulsive actions based on arousal -acting in the heat of the moment-
interfere with decision-making based on established intentions. In other words, the
TPB fails to take into account the possibility that individuals are unable to self-
regulate their behavior effectively (Read, et al., 2006).

30

Although the Theory of Planned Behavior has been shown to have significant
predictive ability in a variety of contexts (e.g. nutrition, exercise, television
consumption), it has been generally well-established that past behavior is the best
predictor of future behavior (Oullette & wood, 1998; Ajzen, 2002, 1991; Schneider &
Shiffrin, 1977). Indeed, this habitual behavior may actually strengthen embodied
cognitions such that a spiral of reinforcement occurs, making old habits hard to break.
If past behavior is such a good predictor, the obvious question then is why not
just ask about a persons past behavior? In the health intervention context, where the
aim is to assess change in future behavior, asking about past behavior is only half the
story. Under the theory of planned behavior, intentions and efficacy measured at post-
intervention should significantly predict to future behavior but studies have
demonstrated only marginal success in predicting immediate short term change let
alone longitudinal change (Sheeran, Conner, & Norman, 2001).
The TPB would make the following prediction: If a person believed that using
a condom would make sex better because it would feel just as good and would keep
them from worrying about getting HIV; And they believed that other people (i.e.
potential partners) were likely to want to use a condom in the given context; And in
addition they believed they were capable of negotiating and following through in
using a condom based on their past experiences (high behavioral control), then this
should predict high intentions and high self-efficacy to use condoms which should
then translate into actual condom use.

31

The TPB also would predict that behavioral control is also capable of predicting
behavior alone apart from intentions. An example would be of a person who only
feels comfortable engaging in sex when under the influence of drugs. Even though the
intentions may be high to use condoms, the person may be physically unable to assert
behavioral control in the given context, and thus perceived behavioral control in the
given context would be a predictor apart from intentions.
The TPB has been shown to be a moderately robust predictor of risk-taking
overall and in the context of condom use in a number of meta-analytic studies. For
example, in a meta-analysis conducted by Armitage & Conner (2001), the TPB
accounted for nearly 27% of actual behavior and perceived behavioral control alone
accounted for an additional 2% of the variance above that of intentions. Not only did
perceived behavioral control contribute an additional 2% in predicting actual
behavior, but it also helped to explain an additional 6% of the variance over and above
that of norms and attitudes in predicting to the formation of actual intentions in the
first place.
Albarracin (2001) found similar results in her meta-analysis of studies that
used the theory of reasoned action and planned behavior to predict condom use. She
found that intentions were correlated with attitudes r=.58 (weighted mean
correlation), intentions were correlated with subjective norms r=.39 and intentions
were correlated with self-efficacy (perceived behavioral control) r=.44. Although
Albarracin ultimately concluded that perceived behavioral control ultimately provided

32

no additional utility in the prediction of condom use, other studies have demonstrated
a slight advantage in including this variable.
For example, Sheeran and Taylor (1999) in their meta-analysis of the same
application of the theory of planned behavior to the prediction of condom use found
that the theory of planned behavior accounted for nearly 42% of the variance, an
additional 5% over and above the Theory of Reasoned Action.
Despite the apparent considerable predictive utility of the theory of Reasoned
Action and Planned behavior in the aforementioned meta-analysis, there is a
surprisingly less consistency when applied specifically to an MSM population. For
example, Kelly and Kalichman (1998), in a cross-sectional study of 297 HIV-negative
MSM, found that only 7% of the variance was accounted for by the TPB, and only and
additional 6% of the variance was accounted for when including perceived behavioral
control (a total of 13% of the variance).
One possible explanation for the discrepancy in predictive utility of the TPB
among MSM compared to other populations may lay in the varying nature of the
relationships that MSM form. For example, Baker (1994) found that among
heterosexual couples, the quality of the relationship, and their commitment to
maintaining their relationship, was directly related to maintaining a monogamous
relationship. However, these same factors were found to be unrelated to whether
MSM engaged in extra-relational sex (deVroome, 2000). Thus MSM may form close
committed relationships, and yet still have outside partners which may mask the

33

homogeneity of a given sample (i.e. single or coupled MSM) and explain the
attenuated utility of the TPB to predict risk-taking among MSM.
Appleby, Miller & Rothspan (1999) found that among a sample of 46 couples
from Los Angeles County, 62% of the sample reported having had sexual relations
with someone other than their current partner during their relationship, and among
those only 30% consistently used condoms when having sex outside their primary
relationship. DeVroome (2000) found that whether MSM chose to use a condom
when with an outside partner was directly dependent on the strength of the relationship
with their steady partner (this finding was significant even controlling for TPB
variables), but it had no bearing on whether outside partners were sought in the first
place.
This inconsistency suggests that the TPB’s capability to predict variance in
risk-taking is more in line with the Kelly and Kalichman study (1998) because MSM
as a sample are different in terms of their willingness to engage in high-risk with non-
primary partners. Indeed, one of the major findings of the Kelly and Kalichman study
was that the reinforcement value of unprotected sex (i.e. pleasure received) and
substance use, accounted for much more variance in predicting risk-taking than
traditional variables such as intentions, attitudes, and perceived vulnerability (Kelly &
Kalichman, 1998). This finding implies that situational valence (amount of arousal,
strength of habit, drug induced impairment) in combination with a larger proportion of
MSM engaging in non-primary UAI relative to heterosexual populations, may

34

contribute more to risk-taking (failing to self-regulate behavior) than intention based
models assume.
Another important factor to consider, which may also impede the prediction of
risk-taking utilizing the TPB is the question of linear causality. Oftentimes, enacting a
behavior can influence subsequent perceptions of the behavior which may in turn
influence future behavioral intentions. Thus, the linearity of the theory may not
necessarily prove to be the most effective measure of future intentions in the
intervention context, because the psychosocial variables may themselves be influenced
by the enactment or ‘rehearsal’ of the behavior.
For example, Festinger (1962) argues that when individuals enact behaviors that
are not consistent with their internal self-concept, then a reappraisal of the underlying
intentions, attitudes, beliefs, and subjective norms will be called forth to help settle the
internal inconsistency that arises. If a person see’s themselves as a risk-taker, but then
subsequently, within the bounds of a guided virtual environment, decides to not be
risky, then subsequent perceptions may in fact be changed. The behavior both
precedes and subsequently influences the future intention. In addition, research has
demonstrated that merely the production of cognitive inconsistency is capable of
producing dissonance-related attitude change (Harmon-Jones, Brehm, Greenberg,
Simon, & Nelson, 1996).

35

Unfortunately, most studies that have attempted to assess the connection
between intentions and efficacy on behavior have been retrospective, rather than
prospective such as in the Kelly and Kalichman (1998) study. The reliance on past
behavior as a substitute for future behavior in the absence of a longitudinal design in
many studies underscores this problem with theoretical linearity. This may also
explain in large part the varying degree to which the TPB is capable of predicting risk.
Indeed, in all the meta-analysis cited earlier, the researchers have combined
postdictive and longitudinal studies. Because past behavior is capable of influencing
future intentions (at least in the near future) it is no wonder the TPB predicts much
more variance in postdictive studies rather than in longitudinal ones.
This current study seeks to address this issue by examining the TPB both in
the context of past behavior -in line with prior met-analysis and research- and in the
context of future behavior to assess the degree to which the TPB adequately predicts
future actual sexual risk-taking. Thus with respect to the TPB the following
hypothesis are:

H
3
: The Theory of Planned Behavior should, consistent with the literature,
significantly predict past risk-taking behavior (UAI) at Time 1 (Postdictive)

H
4
: The Theory of Planned Behavior should, consistent with the literature,
significantly predict future risk-taking behavior (UAI) at Time 2. (Predictive)

36

Although the TPB should significantly predict risk-taking behavior, I have
pointed out a number of potential short comings with the theory both in terms of
predicting future risk-taking behavior generally and specifically when applied to a
sample of MSM. The question then becomes, is there a more comprehensive method
of predicting future risk-taking behavior, and why might there be a more effective
model for predicting behavior change over time
The best predictor of future behavior is past behavior
The answer to this question lies in the old maxim “the best predictor of future
behavior is past behavior.” This view has been widely accepted by a number of health
researchers, social psychologists, and communication scholars (Oullette & Wood,
1998; Aarkts, Verplanken & van Knippenberg, 1998; Ajzen, 2002, Joseph, 1991;
Schneider & Shiffrin, 1977; Ajzen & Fishbein, 1980; Conner & Armitage, 1998;
Albarracin, Johnson, Fishbein, & Meuellerlieile, 2001). Joseph (1991) found that
nearly 17% of the variance of risk-taking UAI was accounted for by past behavior at 6
months follow-up among a sample of 442 MSM, which accounted for far greater
variance than any other psycho-social variables examined (e.g. gay identity, sexual
compulsivity, importance attached to sexual activity).
One argument for the ability of past behavior to predict future behavior is the
influence of habit (Bargh, 1993; Bourdieu, 1977, Brickell, 2006). Oullette and Wood
(1998) have defined habit as the tendency to repeat past behavior in a stable context.
Their research has consistently demonstrated the robust nature of habit in perpetuating
behaviors over time. Ajzen disputes the utility of the concept of habit, arguing first

37

that past behavior is not a valid indicator of habit strength and second that habit is
merely routinized action which is entirely consistent with the TPB, because it is
guided by spontaneously activated attitudes and intentions (Ajzen, 2001). For Ajzen
then, the distinction between a habit and an intention is a meaningless difference so
long as the context of the behavior remains unchanged. Azjen believes that habits are
merely activated routinized intentions -an argument similar to the dual processing
elaboration likelihood theory promulgated by Petty and Cacioppo (1986) (i.e. behavior
can be guided by conscious deliberation OR by reliance on heuristics as a way of
reducing cognitive load) (see also Slomman, 2003).
This is an interesting tropological argument. However, when you dig into the
underlying logic of this perspective there lays a fundamental flaw in the assumption
that context, novel or otherwise, is the only factor that mediates the activation of
spontaneous or new intentions. If when context remains stable, why is it then that the
TPB does not predict future risk-taking as well as past behavior? Ajzen’s explanation
for this inconsistency is that intentions must have either changed between the time
asked and the behavior measured later on, or that the intentions were simply not strong
enough to begin with (Ajzen, 2001).
This actually belies the point, because the TPB does NOT do a great job at
measuring future risk-taking very well as the amount of time between the measured
intention and the behavior itself increases (Ajzen & Fishbein, 1980). However, it has
been found that intentions are sometimes stabilized when performance of the behavior
in the past occurs regularly (Kashima, Gallois, & McCamish, 1992). Thus, there are

38

three issues: one is that intentions are so contextually based that it is difficult to
measure intentions with regard to a specific behavior, because a host of influences
may influence the context of the act, and therefore the specific intention applied;
Second, as Ajzen argues, habits are notoriously difficult to measure in any meaningful
way given that the number of acts performed in the past does not necessarily a habit
make; and thirdly, habits and intentions are not mutually exclusive dimensions
because they rely on one another’s effects over time.

39

CHAPTER 4:

SHORTCOMINGS OF PRIOR MEDIATIONAL MODELS
IN PREDICTING RISK-TAKING

The TPB is a mediational model which predicts that the effect of past behavior
on intentions and subsequent behavior is mediated by intentions and self-efficacy.
However, there have been a number of studies that have reported independent effects
for past behavior beyond the explanatory value the TPB provides, suggesting that the
cognitive measures do not fully capture the decision-making processes implicit in the
engagement of actual behavior. For example Godin, Valois, and Lepage (1993) found
that in a longitudinal study (8 & 9 month follow ups) of exercise behavior among
pregnant women, none of the TPB variables were associated with predicting actual
future behavior. One possible explanation could be due to hormonal changes during
pregnancy that may influence affective states among these women, which are not
currently accounted for under the TPB.
Similarly, in a study of alcohol consumption by students, Conner, Warren,
Close, and Sparks (1999) found that measures of past behavior consistently
contributed to predictions of future behavior above and beyond contributions from
intentions and perceived behavioral control. This suggests that young people may
engage in drinking or drug use as a means of self-medication to alleviate negative
affective states. Indeed, because alcohol is a depressant, this can often create a
downward spiral leading towards alcoholism. In addition, drinking is often used as a
social lubricant especially among young people who are learning to fit in to new

40

environments they may not be familiar or comfortable with. Thus motivations to
engage in drinking may not even be cognitively clear or self-evident when an
individual initially decides to engage in such behavior which can then subsequently
lead down the path towards impaired cognitive decision-making and/or more
automatic habitual risk-taking (Halkitis & Parsons, 2002; Frosh et al., 1996).
Ajzen (1991) has argued that the perceived behavioral control construct within
the TPB should mediate the effect of past behavior on future behavior because
repetition of behavior should lead to increased perceptions of control, however, as the
example of habitual drinking demonstrates, this is not always the case. –A lot of
alcoholics say “I can quit whenever I want to”--. A number of researchers have begun
to examine how additional constructs could be taken into account to better explain
how non-cognitive processes may play a mediational role in the formation of
intentions and/or behavior.
Is there an alternative, superior method that might be used to either augment or
supersede the TPB’s ability to predict future risk-taking? One possible method is
through the tracking of virtual behavioral actions at the time of the intervention. In
other words, allow the individual to make decisions on a virtual encounter that
simulates what that person might do in real life, as a measure of their ‘actual’ risk-
taking in the future. A virtual environment assessment may actually do a better job at
predicting risk because it takes into account both the willingness of the individual to
assent to risk –when resistance to condom use by a partner is encountered—as well as

41

the tendency to act in contradiction to prior intentions under a sexually ‘hot’ arousal
state.
Interactive virtual environments hold forth the possibility that future risk-
taking can be assessed both in terms of the context and in the strength of the
habit/intention, at the time of the act. A number of studies have demonstrated the
proclivity for individuals to treat virtual spaces, consistent with their real world
experiences (Godoy, 2006, Biocca, 1995; Lee, 2006, Fogg, 2001, Bailenson, 2003;
Blascovitch, 2002). In addition, behavior in a virtual environment should be able to
take into account the degree to which outside factors might influence the follow-
through of intentions as well as the strategic ability for individuals to successfully self-
regulate their behavior. For example, some MSM may have good intentions to stay
safe and do so, but others, in the heat of the moment, with a partner who doesn’t want
to use a condom may, even though the situation may not involve the development of
new intentions, assent to taking sexual risks. Arousal may work in the same manner
as heuristics do within the ELM theory, but instead of taking a short-cut towards
routinized intentions, it takes a short-cut through intentions towards routine sexual
gratification.
Another reason virtual environments may do a better job at predicting future
risk-taking, is the degree to which individuals are able to accurately assess their
perceived behavioral control in a given situation. Ajzen terms this concept belief
veridicality, which is the extent to which individuals accurately represent their ‘actual
control’ over a given situation (Ajzen, 2001). Another way to think of this concept is

42

that individuals who may perceive themselves capable of (or efficacious) when
successfully initiating, negotiating, and using safer sex (and refusing to engage in
unsafe sex) may be succumbing to biases in “affective forecasting” in which under
“cold” emotional state (e.g. not sexually aroused) they fail to take into account “hot”
emotional states (e.g., sexual desire) and how this state will impact their self-efficacy
(see for example, Gilbert, Gill, & Wilson, 2002; Loewenstein, O’Donoghue, & Rabin,
2003) This may produce a false sense of security in entering a decision-making path
(e.g., going to a club; taking drugs or alcohol before sex) that will be far riskier than
they anticipate. Virtual environments may be able to capture what this self reported
data cannot.
This is the crucial difference between a linear narrative and an interactive
narrative. Only with an IAV are you able to get at the issue of interactivity. In the
observational learning context you may be able to only model, or influence the
rational components involved in decision-making. The affective influences that are
involved in actual decision-making are likely to show up only within an interactive
environment -because like in real life, you make actual decisions based on ‘gist’ cues.
In order to test this, then you would need to compare identical interventions, one with
interactivity (IAV) the other a yoked control (choices predetermined).
Those who are safe within the virtual environment may be laying the
groundwork for much more specific intentions to be safe, which may be substantially
different from the generic intention to always use condoms (i.e. the very rehearsal of
the behavior is allowing the creation of implementation intentions that may prove

43

valuable in maintaining risk-free behavior when faced with similar situational cues).
Ajzen (2001) makes this same argument (stressing the importance of specific
implementation intentions, citing Gollwitzer, 2000) when he tries to explain the
apparent difficulty researchers face when trying to predict change in future risk-taking
using the TPB in the health intervention context.
Thus, given the fact that past behavior has been found to be best predictor of
future behavior and that virtual environments should be able to acts a proxy for future
behavior (capturing all of those intangible situational, contextual, and affective
responses that may interfere with self-regulation of behavior based upon pure
cognitive rationality), the following hypothesis are presented:
H
5
: Past sexual risk-taking behavior should, consistent with the literature,
significantly predict future sexual risk-taking behavior.

H
6
: Interactive Video Choices should predict past risk-taking behavior (UAI)
at Time 1, better than the traditionally used Theory of Planned Behavior for
the IAV condition but not for the Yoked Condition (Postdictive)

H
7
: Interactive Video Choices should predict future risk-taking behavior
(UAI) at Time 2, better than the traditionally used Theory of Planned
Behavior for the IAV condition but not for the Yoked Condition (Predictive)

44

CHAPTER 5:

TOWARDS A MORE ROBUST PREDICTIVE MODEL:
COMBINING BOTH EXPLICIT COGNITIVE MEASURES
AND UNOBTRUSIVE EXPERIENTAL BEHAVIORAL MEASURES

Epstein has proposed a theory known as Cognitive-Experiential Self Theory
(CEST) which is based on the notion that decision-making relies on two conceptual
systems that operate in parallel on a given task. He argues that there is both an
experiential system, which is affectively based --designed to for rapid processing of
information and decision-making-- and a rational system –one that is cognitive and
based on more deliberative reasoning--. The experiential system works by encoding
reality through tagging of affective states onto images and narratives (Slovic,
Finucane, Peters, & MacGregor, 2004).
In a classic example of how the experiential system operates to guide behavior
Denes-Raj and Epstein (1994) found that people would rather draw from a large jar
with more total winning beans, than from a small jar with less winning beans, but
greater odds of winning relative to the number of total beans. This occurred despite the
fact that they were told the odds were better with the smaller jar. People ‘felt’ that they
had a better chance because they could see more winning beans.
Slovic (2001) found that in the case of smoking, young people are particularly
susceptible to responding to the affective impulses of the moment and acting
experientially. This leads adolescents down a path, where visceral emotions a la
addiction can take hold. This is the same pattern that risky sexual behavior holds as

45

well for young people. If you lack the experience to properly valuate the costs and
benefits over the long term, then the experiential system can fail early on, thus
allowing the individual to be captured by the experience of subsequent visceral
emotions –creating a risky feedback loop.
The experiential system is based on what you know and see and feel, rather
than on abstract concepts such as probabilities (e.g. a 1 in 100 chance). The less
abstract the perception of the risk, the more likely the experiential system will hold
sway. For example, it was found that a judge will be more likely to grant parole if it is
argued that there is only a 10% chance of recidivism, instead of saying only 10 out of
100 people convicted of this similar crime will commit it again. The risk is assessed
according to how immediate the affect tagged object is perceived.
Evidence that this system also operates through media representations also is
found in a number of media studies which demonstrate that highly publicized murders
or kidnappings increase the perceptions of risk to being a victim of such crimes, even
though statistically this is not the case (Coleman, 1993). The crimes that are
publicized are highly affectively charged. Affect matters because it influences the
need to attend to a given situation. People will oftentimes stop to help an individual in
need, but will not feel the same urgency if a large group of people are in need of help.
The facelessness of a group of people in need dampens the affective influence.
Behavior therefore, is guided by risk perceptions but these in turn are shaped
by both cognitive and non-cognitive factors. What is most interesting is that it has
been shown that even cognitive factors are shaped by affective states as well. In a

46

process known as the ‘affect heuristic’ researchers have demonstrated that affective
views of a given behavior or thing are capable of guiding perceptions of risk based on
a subjective affective weighing of the benefits. That is, how one feels about it
influences how one ultimately considers it.
For example, if you encounter a highly attractive partner, this should lead to
more positive affect towards that partner (i.e. a halo effect), which would then in turn
lead to decreased perceived risk in being sexually unsafe. This effect has been
confirmed in a number of decision-making studies (Finucane et al., 2000). The same
would hold true for the converse, if you have an unattractive mate who wants to be
unsafe, you would have a negative affective state to begin with and then would also
perceive the risk of being unsafe to be ‘even riskier’. Thus there is an inverse effect
between perceived risks and benefits.
This effect was found to increase even more under time pressure experiments.
In the sexual scenario, judgments about whether to be safe or not often occur under
similar conditions, requiring snap decisions, in which this affect heuristic can fail to
protect the individual. Thus we would assume that a mediational model that accounts
for these effects by including both an experiential measure as well as a cognitive
measure, should account for a wider spectrum of variance in sexual risk-taking that
occurs among individuals.

47

Because the residual effect of prior on later behavior is said to be due to the
constellation of factors that include situational, contextual, and affective responses, as
well as self-professed intentions and volitional control, than the TPB coupled with
behavior in a virtual environment should account for a majority of the variance in the
IAV condition but not for the Yoked, making past behavior no longer a significant
predictor of future risk-taking in the health intervention context for the IAV, but still
maintaining its residual effect for the Yoked. Thus the following hypothesis:

H
8
: Variables from the Theory of Planned Behavior (i.e., intentions, self-efficacy)
coupled with behavior in a virtual environment should fully mediate the link
between past and future real-life risk-taking. It should do so in the IAV condition
but not for the Yoked condition.

48

CHAPTER 6:
METHOD
Participants
Eligibility
Because this was an intervention designed for younger African-American,
Latino, and Caucasian MSM at risk for contracting HIV through unprotected anal
intercourse, those eligible to participate in the longitudinal study were men who had
engaged in receptive or insertive unprotected anal intercourse (UAI) with another man
in the past 90 days, were 18 to 30 years old, HIV-negative, had never used injection
drugs, would be in the Los Angeles area for the next 3 months, and were either
African American, Latino, or Caucasian. MSM were recruited randomly at street
intercepts, bar/dance clubs, over the internet, through gay print media, and gay pride
events.
Sub Sample 1: Hypotheses 1 & 2
Sub sample 1 consisted of 76 subjects at time one (IAV condition). Thirty two
were Latino (42.1%), thirteen were African American (17.1%), and thirty one were
Caucasian (40.8%). Their Mean age was 24 with 50% above or below that and the
mean income between 30 and 35,000. At the return visit 90 days later our sample
consisted of 50 subjects. 21 were Latino (42%), 7 were African American (14%), and
22 were Caucasian (44%). Their mean age was 24 with 50% above or below that and
the mean income was between $30,000 and $35,000.

49

Sub Sample2: Hypothesis 3-8
Sub sample 2 consisted of 151 subjects at time one and 97 subjects at Time 2
(three months later). At time 1: sixty one were Latino (40.4%), twenty seven were
African American (17.1%), and sixty two were Caucasian (41.1%). Their Mean age
was 24 with 50% above or below that and the mean income between 30 and 35,000.
At the return visit 90 days later our sample consisted of 97 subjects. Thirty seven
were Latino (38.1%), 15 were African American (15.5%), and 44 were Caucasian
(45.4%). Their mean age was 24 with 50% above or below that and the mean income
was between $30,000 and $35,000.

Procedure
Development of the interactive video. In order to create the IAV, we first
conducted exploratory research to identify the narrative structures of typical MSM
sexual scenarios. Pilot studies were conducted which allowed us to identify key
features of sexual scenarios that might precede risky behavioral choices (e.g. drug use
and UAI). One-on-one interviews and Community Advisory Boards (CAB) informed
the content of the intervention in terms of setting, characters, language use, and
storyline. These CAB’s were of two types. Leadership CAB’s were composed of
representatives from community-based organizations, the entertainment industry, and
local businesses frequented by MSM. Youth CAB’s were composed of young MSM,
all of whom met the study’s eligibility criteria.

50

This research served as the foundation for three separate scripts, each of which
was tailored to a different ethnicity. In an attempt to create interventions that would
speak to their intended audiences, the narrative portion of each script was written by a
professional MSM screenwriter that was matched on ethnicity. Our research team
then wrote and integrated content that was informed by relevant social science
literatures. Prior to production, the script was approved by the Community Advisory
Board’s. The CAB provided feedback regarding the realism of the intervention and its
sensitivity to the MSM community. Realism and sensitivity to the community were
continually addressed throughout the production process. The director, producers, and
many other key personnel were members of the MSM community and represented
diverse ethnic and racial backgrounds.
We produced three separate, yet factually equivalent, versions of the Virtual
Sex Project interactive video (IAV). Each intervention was tailored to be culturally
sensitive to the specific ethnicities targeted in our study: African American, Latino, or
Caucasian. The IAV was programmed such that a prior randomly assigned
participant’s choices could be logged and then re-entered as a yoked linear version for
those randomly assigned to the Yoked condition.
Participant procedure. Upon arrival, participants were introduced to the
researchers and underwent the informed consent process. Participants then either
viewed an interactive video or a yoked video. Reports of sexual behavior were
collected and various measures (e.g., past behavior, self-efficacy, intentions, affect)
were assessed pre- and post-intervention. These measures were collected once more

51

when participants returned for the second session three months later. During the 12-
week period between Sessions One and Two, participants completed weekly phone-in
(or email) sexual behavioral assessments. Participants received $50.00 for each of
session and $5.00 for each weekly phone-in completed. Additionally, they received
$10.00 for each eligible participant referred to the study with a maximum of three
referrals. Thus, participants completing all phases of the study received a total of
$190.00.
Interactive Video Scripts and Guides
Guides. Before the narrative begins, two “guide” characters appear on-screen
and introduce themselves. One guide character is older, more authoritative, and more
knowledgeable of safer sex practices than the other. The other guide is younger and
provides more of a humorous approach to the content. After providing brief
instructions on how to use the video, the guides encourage the user to assumed the role
of the main character and become immersed in the virtual environment, interacting
with other characters and making decisions as the story progresses. The guides remind
the user that they “will be there when it counts, and when things start to heat up”
offering obligatory and non-obligatory advice at critical decision points throughout the
narrative. Throughout the interaction, the guide characters are available for advice at
the choice of the viewer, but also will “pop up” automatically, interrupting the user’s
decisional choice when that choice is risky (e.g., when the user chooses drugs,
unprotected anal sex, etc.). Those in the yoked condition watched this same video but

52

with the choices pre-programmed based on what a randomly assigned prior participant
in the IAV condition chose.
Story line and guide messages within the interaction. When the story begins,
the user is asked to choose where he would like to meet a potential sexual partner. At
this decision point, two options are available: at a dance club or via the internet.
Initial piloting data suggested that these are the two most common settings in which
young MSM meet a sexual hookup, regardless of ethnic background. Once the user
makes a decision, the two characters meet and engage in small talk as they get to know
each other better. Soon thereafter, it becomes obvious that both characters are
attracted to each other and are interested in pursuing a sexual encounter.
Their conversation, however, is interrupted by a peripheral character that offers
the user drugs (i.e. crystal methamphetamine). At this decision point, three options are
available to the user: accept the drugs, decline the drugs, or ask for advice. If the user
accepts the drugs, the guide characters pop-up and provide a mandatory intervention in
which they warn of the risks associated with crystal methamphetamine use in a sexual
context. If the user decides to decline the drugs, the narrative continues and the main
character models successful refusal skills. If the user chooses to ask for advice, the
guide characters appear and offer guidance before encouraging the user to go back and
make a choice.
As the narrative progresses, the main character agrees to go to his partner’s
apartment. While there, he is offered an alcoholic beverage and, again, must make a
decision. Three options are available at this choice point: accept the drink, decline the

53

drink, or ask for advice. Subsequently, they continue to get to know each other better
and exchange flirtatious pleasantries. The guide characters then encourage the user to
initiate a conversation about safer sex “before things get too hot and heavy.” The user
must decide how he would like to bring up the topic of safe sex; by discussing it
directly, by talking about their HIV status, or by talking about their past
relationships/sexual history. If either of the latter two options are chosen, the user
views the appropriate scene and is additionally obliged to view the “direct” discussion
of safer sex in an attempt to reiterate the value of such a frank dialogue. The story
continues, ultimately allowing the users to have virtual intercourse by selecting what
type of sex and what positions they would like.

Dependent Measures

Subjects were asked if they had engaged in any of the following behaviors
(drank alcohol, took crystal methamphetamine, was a top or a bottom, and had
unprotected anal intercourse) in the past 90 days, and were asked the same questions
90 days later. These were coded as binary measures, with 1 yes, and 0 no for each
behavior. Sexual position preference and UAI risk was also binary coded. Those who
engaged in anal sex as an insertive partner at least 60% of the time was coded 1, top,
and those who did not were coded as 0, bottom. A dichotomized median split was
used to binary code sexual risk (amount of UAI) as (0) low (1) high at both Time 1
and Time 2 for Sub sample 1.

54

For Sub sample 2 Sexual risk-taking in the IAV and YOKED was dummy
coded on two binary variables: The type of sex (e.g. Anal sex 1, other type of sex –0)
and whether they used a condom or not (e.g. chose to not use a condom 1, chose to use
a condom 0).

UAI Non-Primary at Time 1 (the past 90 days
Participants self reported at baseline the total number of times that they had
UAI with a non-primary partner both as the insertive or receptive partner in the past 90
days at Time 1.

UAI Non-Primary at Time 2 (the past 90 days)
Participants returned three months after Time 1 and self reported the total
number of times that they had UAI with a non-primary partner both as the insertive or
receptive partner in the past 90 days at Time 2.

Intentions to Use Condoms
The intentions to use condoms scale was designed to measure the
participants present intentions regarding condom use. The scale was administered
both before viewing of the video and then subsequent. This scale was adapted from
Kelly, et al. (1995) and comprises three items: 1) During the next three months you
will be direct with your partner and agree that you are going to use condoms before
things get to hot and heavy. 2) During the next three months, if you are going to have

55

anal sex, you will use condoms each and every time you have sex. 3) During the next
three months, you will talk about safe sex before things get to hot and heavy. These
intention items are designed to map onto the same components that would define safer
sex behavior (e.g. the action, target, context, and time).
Using principal component analysis, all items factor loaded above .75 at
baseline and with a reliability α=.763 at baseline (see Table 1). At post, all items factor
loaded above .80, with a reliability α=.819 (see Table 2).

56

Table 1:
Baseline Items Comprising Intentions to Use Condoms and Factor Loadings

Reliability Statistics
.763 3
Cronbach's
Alpha N of Items

Component Matrix
a
.887
.826
.765
You will be direct with your
partner and agree that you
are going to use condoms
before things get too hot and
heavy.
You will talk about safe sex
before things get too hot and
heavy.
If you are going to have anal
sex, you will use a condom
each and every time.
1
Componen
t
Extraction Method: Principal Component Analysis.
1 components extracted.
a.

57

Table 2:
Post-Intervention Items Comprising Intentions to Use Condoms and Factor Loadings

Reliability Statistics
.819 3
Cronbach's
Alpha N of Items

Component Matrix
a
.889
.883
.803
Post-Intent: You will talk
about safe sex before things
get too hot and heavy.
Post-Intent: You will be
direct with your partner and
agree that you are going to
use condoms before things
get too hot and heavy.
Post-Intent: If you are going
to have anal sex you will use
a condom each and every
time you have sex.
1
Componen
t
Extraction Method: Principal Component Analysis.
1 components extracted.
a.

58

Self-Efficacy To Use Condoms. The self-efficacy scale was designed to
measure the participant’s present sense of self-efficacy regarding condom use. The
scale was administered both before viewing of the interactive video and then
subsequent. This scale was adapted from Kelly, et al. (1995) and comprises three
items: 1) How confident are you that you can bring up the issue of condoms to your
sexual partner. 2) How confident are you that you can convince your partner to use
condoms even if he says he doesn’t want to. 3) How confident are you that you would
leave the situation if he refused to use condoms? These items are designed to map
onto the range of situations that self-efficacy would matter most in each of these three
situations that are likely to arise when initiating and negotiating safer sex. Using
principal component analysis, all self-efficacy items factor loaded above .80, with a
reliability α=.771 at baseline (see Table 3). At post, all items factor loaded above .74,
with a reliability α=.701 (see Table 4)

59

Table 3: Baseline Items Comprising Self-Efficacy Scale and Factor Loadings

Reliability Statistics
.771 3
Cronbach's
Alpha N of Items

Component Matrix
a
.855
.839
.813
How confident are you that
you could bring up the issue
of condoms to your sexual
partner?
How confident are you that
you would leave the situation
if you sexual partner refused
to use a condom?
How confident are you that
you can convince your
partner that the two of you
should use a condom or have
safe sex?
1
Componen
t
Extraction Method: Principal Component Analysis.
1 components extracted.
a.

60

Table 4: Post-Intervention Items Comprising Self-Efficacy Scale and Factor Loadings

Reliability Statistics
.701 3
Cronbach's
Alpha N of Items

Component Matrix
a
.848
.821
.746
Post-Efficacy: How
confident are you that you
can bring up the issue of
condoms to your sexual
partner?
Post-Efficacy: How
confident are you that you
would leave the situation if
your sexual partner refused
to use a condom?
Post-Efficacy: How
confident are you that you
can convince your partner
that the two of you should
use a condom or have safe
sex?
1
Componen
t
Extraction Method: Principal Component Analysis.
1 components extracted.
a.

61

CHAPTER 7:
RESULTS
Study Set 1
Results H
1
: Interactive Video Choices will be related to past risk-taking behavior
The first hypothesis of the current work is that interactive video choices will be
related to past real-life behavior. To assess this, a series of Chi square analyses were
performed. As indicated in Table 5, there was a significant relationship between
subjects’ past behavior and the subsequent virtual behaviors (choices made in the
IAV) that subjects. Those who drank alcohol in the past 90 days were likely to choose
to drink alcohol in the virtual environment X
2
(1, N= 84) = 4.677, p = .031 (see Table
5). Similarly, those who took methamphetamine in the last 90 days, were also likely to
chose to take methamphetamine in the virtual environment, X
2
(1, N= 84) = 54.497, p
=.000 (see Table 5). Of those who chose to have anal sex in the IAV, position
preference was consistent with their real world preferences. Those who were more a
top (at least 60% of the time the insertive partner, in the past 90 days) chose to be a
top in the IAV, and those who preferred to be a bottom (at least 60% of the time the
receptive partner in the past 90 days) chose to be a bottom in the IAV, X
2
(1, N= 54) =
9.481, p = .002. Those who chose to have risky anal sex (no condom) in the virtual

62

environment
2
were more likely to have engaged in risky anal sex in the last 90 days,
X
2
(1, N= 84) = 5.997, p = .014 (see Table 5).
Results H
2
: Interactive Video Choices will predict to future risk-taking behavior
The second hypothesis tested is that interactive video choices predict future
risk-taking behavior. As indicated in table 5, a series of Chi square analyses
demonstrated a significant relationship between subjects’ virtual behaviors and the
behaviors they were likely to engage in 90 days into the future. Those who drank
alcohol in the interactive video were likely to drink alcohol in the next 90 days, X
2
(1,
N= 51) = 7.686, p =.006 (see Table 5). Unfortunately, those who chose to take
methamphetamine in the virtual environment did not return for their visit two follow-
up if that had taken methamphetamine in real life in the past 90 days at baseline, so we
were unable to perform Chi square on this variable (see Table 5). There may have
been some type of reactance due to the fact that this was the one interactive feature
that was not allowed to follow through after the choice was made (i.e. using
methamphetamine was not shown in the IAV if it was picked).

63

2
Because subjects could choose from a range of sexual activities, not all subjects chose to engage
in anal sex thus limiting the sample size for these Chi squares.

Of those who chose to have anal sex in the IAV, position preference choice in
the interactive video was predictive of their future preference for the next 90 days, X
2

(1, N= 29) = 13.426, p =.000. Those who chose to have risky anal sex in the virtual
environment were more likely to have engaged in risky anal sex in the last 90 days at
time 2, X
2
(1, N= 51) = 6.872, p = .009 (see Table 5).

64

Table 5:

Percentage of MSM Engaging or Not in Real Life Risk Behavior in Past 90 Days at
Baseline and Subsequent Follow-up Who Engaged in Virtual Risk Behavior or Not in
the IAV Condition

% Choosing Virtual Risk Within Those Reporting
Presence or Absence of Behavior in Real-life

Alcohol Use

Methamphetamine Use

UAI
Count Count as %
of Row Total
Count Count as %
of Row Total
Count Count as %
of Row Total
Baseline
(90 Days Pre-IAV)

No or Low 3 42.9% 1 1.3% 1 2.2%
Yes or High 61 79.2% 7 77.8% 7 17.9%
Follow-up
(90 Days Post-IAV)

No or Low 1 20% 3 6% 1 3.0%
Yes or High 36 78.3% *
3
* 5 27.8%

3
Three subjects who never took methamphetamine at Time 1 (past 90 days at baseline), chose to
take it virtually, but did not take it in real life at Time 2. This suggests that there may be some
‘divergent play’ for high-risk behaviors that could potentially provide experiental learning to
avoid these high risk behaviors in real life.

65

Study Set 2

For the second set of analysis encompassing Hypothesis 3-8, parametric
tools were used. Before testing each hypothesis, we first performed data cleaning and
transformation steps (following the procedures laid out in Tabachnick & Fidell, 2005),
followed by the appropriate analysis given the hypothesis.

Results H
3
: Theory of Planned Behavior Will Significantly Predict Past Risk-Taking
Behavior (postdictive)

A simultaneous regression was performed between number of unprotected sex
acts with non-primary partners at baseline (Time 1–past 90 days) as the dependent
variable and intentions to use condoms and self-efficacy at baseline as independent
variables. Analysis was performed using SPSS regression and SPSS explore for
evaluation of assumptions. The sample consisted of 151 men who had UAI at least
once with a non-primary partner at Time 1(the Yoked & IAV conditions combined).
Results of evaluation of assumptions led to transformations of the variables to
reduce skewness, eliminate outliers, and improve the normality, linearity, and
homoscedasticity of variables. A log 10, reflect transformation was used on the
measures of intentions to use condoms and self-efficacy to reduce severe negative
skew. A log 10 transformation was used on the dependent variable to reduce severe
positive skew.
With the use of a p<.001 criterion for Mahlanobis distance, with 2 degrees of
freedom, no outliers were found. No missing data was present, n=151.

66

67
Table 6 displays the correlations between the variables, the unstandardized
regression coefficients (B) and intercept, the standardized regression coefficients (B),
the semi-partial correlations, R
2
, and adjusted R
2
. R for regression in the IAV
condition was significantly different from zero. F (2,148) = 4.420, p=.014, with R
2
at
.056 and 95% confidence limits from (.370 to .624). The adjusted R
2
value of .044
indicates that 4.4% of the variance of past UAI risk is predicted by the Theory of
Planned Behavior. One regression coefficient, intentions to use condoms, differed
significantly from zero. The confidence limits for the intentions to use condoms
variable was .085 to .424. Self-efficacy appears to be acting as a suppressor variable,
but it alone contributes nothing to the model (Beta = -.101).
There is no problem with multi-collinearity. Outliers disappeared once the
variables were transformed, and a confirmation using plots and Mahlanobis distance
confirmed this. The skew divided by the standard error of the skew for all variables,
fell below a z-score deviation of 2.03, which is within the limits of assumed normality.
The residuals appear to follow the regression line rather well from an observation of
the plot of the observed and expected residuals.

68

Table 6: Theory of planned behavior predicting to past behavior

Correlations
1.000 .219 .002
.219 1.000 .398
.002 .398 1.000
. .003 .489
.003 . .000
.489 .000 .
151 151 151
151 151 151
151 151 151
UAI past 90 days (baseline)
Intentions Baseline
Efficacy Baseline
UAI past 90 days (baseline)
Intentions Baseline
Efficacy Baseline
UAI past 90 days (baseline)
Intentions Baseline
Efficacy Baseline
Pearson Correlation
Sig. (1-tailed)
N
UAI past
90 days
(baseline)
Intentions
Baseline
Efficacy
Baseline

69

Model Summary
.237
a
.056 .044 .32053 .056 4.420 2 148 .014
Model
1
R R Square
Adjusted R
Square
Std. Error of
the Estimate
R Square
Change F Change df1 df2
Sig. F
Change
Change Statistics
Predictors: (Constant), Efficacy Baseline, Intentions Baseline
a.

Table 6: Theory of planned behavior predicting to past behavior (continued)

Coefficients
a
.497 .064 7.758 .000 .370 .624
.255 .086 .259 2.973 .003 .085 .424 .219 .237 .237 .841 1.188
-.083 .071 -.101 -1.158 .249 -.224 .058 .002 -.095 -.092 .841 1.188
(Constant)
Intentions Baseline
Efficacy Baseline
Model
1
B Std. Error
Unstandardized
Coefficients
Beta
Standardized
Coefficients
t Sig.
Lower
Bound
Upper
Bound
95% Confidence Interval
for B
Zero-order Partial Part
Correlations
Tolerance VIF
Collinearity Statistics
Dependent Variable: UAI past 90 days (baseline)
a.
ANOVA
b
.908 2 .454 4.420 .014
a
15.206 148 .103
16.114 150
Regression
Residual
Total
Model
1
Sum of
Squares df
Mean
Square F Sig.
Predictors: (Constant), Efficacy Baseline, Intentions Baseline
a.
Dependent Variable: UAI past 90 days (baseline)
b.

Results H
4
: The Theory of Planned Behavior Will Significantly Predict Future
Risk-Taking Behavior

A simultaneous regression was performed between number of unprotected
sex acts with non-primary partners at baseline (Time 1 –past 90 days) as the
dependent variable and intentions to use condoms at post-intervention and self-
efficacy at post-intervention as independent variables. Analysis was performed
using SPSS regression and SPSS explore for evaluation of assumptions. The
sample consisted of 97 men who returned for three month follow-up and had UAI
at least once with a non-primary partner at baseline.
Results of evaluation of assumptions led to transformations of the variables
to reduce skewness, eliminate outliers, and improve the normality, linearity, and
homoscedasticity of residuals. A log 10, reflect transformation was used on the
measures of intentions to use condoms and self-efficacy to reduce severe negative
skew. A log 10 transformation was used on the dependent variable to reduce
severe positive skew.
With the use of a p<.001 criterion for Mahlanobis distance, with 2 degrees
of freedom, no outliers were found. No missing data was present, n=97.
Table 7 displays the correlations between the variables, the unstandardized
regression coefficients (B) and intercept, the standardized regression coefficients
(B), the semi-partial correlations, R
2
, and adjusted R
2
. R for the regression was
significantly different from zero, F (2,94) = 7.805, p=.001, with R
2
at .142 and
95% confidence limits from (-.035 to .312). The adjusted R
2
value of .124

70

71
indicates that 12.4% of the variance of future UAI risk is predicted by the Theory
of Planned Behavior. One regression coefficient, intentions to use condoms,
differed significantly from zero. The confidence limits for the intentions to use
condoms variable was .249 to .805. Self-efficacy appears to be acting as a
suppressor variable, but it alone contributes nothing to the model (Beta = -.075)
There is no problem with multi-collinearity. Outliers disappeared once the
variables were transformed, and a confirmation using plots and Mahlanobis
distance confirmed this. The skew divided by the standard error of the skew for all
variables, fell below a z-score deviation of 2.03, which is within the limits of
assumed normality. The residuals appear to follow the regression line rather well
from an observation of the plot of the observed and expected residuals.

72

Table 7: Theory of planned behavior predicting to future behavior
Correlations
1.000 .372 .114
.372 1.000 .464
.114 .464 1.000
. .000 .134
.000 . .000
.134 .000 .
97 97 97
97 97 97
97 97 97
UAI past 90 days
(follow-up)
Intentions Post-Intervention
Efficacy Post-Intervention
UAI past 90 days
(follow-up)
Intentions Post-Intervention
Efficacy Post-Intervention
UAI past 90 days
(follow-up)
Intentions Post-Intervention
Efficacy Post-Intervention
Pearson Correlation
Sig. (1-tailed)
N
UAI past 90
days
(follow-up)
Intentions
Post-
Intervention
Efficacy
Post-
Intervention

73

Table 7: Theory of planned behavior predicting to future behavior (continued)
Coefficients
a
.139 .087 1.586 .116 -.035 .312
.527 .140 .406 3.768 .000 .249 .805 .372 .362 .360 .785 1.274
-.088 .126 -.075 -.694 .489 -.339 .163 .114 -.071 -.066 .785 1.274
(Constant)
Intentions Post-Interventi
Efficacy Post-Intervention
Model
1
B Std. Error
Unstandardized
Coefficients
Beta
Standardized
Coefficients
t Sig.
Lower
Bound
Upper
Bound
95% Confidence Interval
for B
Zero-order Partial Part
Correlations
Tolerance VIF
Collinearity Statistics
Dependent Variable: UAI past 90 days (follow-up)
a.
Model Summary
.377
a
.142 .124 .40623 .142 7.805 2 94 .001
Model
1
R R Square
Adjusted R
Square
Std. Error of
the Estimate
R Square
Change F Change df1 df2
Sig. F
Change
Change Statistics
Predictors: (Constant), Efficacy Post-Intervention, Intentions Post-Intervention
a.

ANOVA
b
2.576 2 1.288 7.805 .001
a
15.512 94 .165
18.088 96
Regression
Residual
Total
Model
1
Sum of
Squares df
Mean
Square Sig. F
Predictors: (Constant), Efficacy Post-Intervention, Intentions Post-Intervention
a.
Dependent Variable: UAI past 90 days (follow-up)
b.

Results H
5
: Past sexual risk-taking behavior will significantly predict future
sexual risk-taking behavior

A simultaneous regression was performed between number of unprotected
sex acts with non-primary partners at follow-up (Time 2 –past 90 days) as the
dependent variable and number of unprotected sex acts with non-primary partners
at baseline as independent variables. Analysis was performed using SPSS
regression and SPSS explore for evaluation of assumptions. The sample consisted
of 97 men who returned for three month follow-up and had UAI at least once with
a non-primary partner at baseline.
Results of evaluation of assumptions led to transformations of the variables
to reduce skewness, eliminate outliers, and improve the normality, linearity, and
homoscedasticity of residuals. A log 10, reflect transformation was used on the
measures of intentions to use condoms and self-efficacy to reduce severe negative
skew. A log 10 transformation was used on the dependent variable to reduce
severe positive skew.
With the use of a p<.001 criterion for Mahlanobis distance, with 2 degrees
of freedom, no outliers were found. No missing data was present, n=97.
Table 8 displays the correlations between the variables, the unstandardized
regression coefficients (B) and intercept, the standardized regression coefficients
(B), the semi-partial correlations, R
2
, and adjusted R
2
. R for the regression was
significantly different from zero, F (1,95) = 24.997, p=.000, with R
2
at .208. The

74

75
adjusted R
2
value of .200 indicates that 20% of the variance of future UAI risk is
predicted by past behavior. The regression coefficient, UAI with a non-primary
partner at baseline, differed significantly from zero. The confidence limits, for this
variable was .360 to .835.
There is no problem with multi-collinearity. Outliers disappeared once the
variables were transformed, and a confirmation using plots and Mahlanobis
distance confirmed this. The skew divided by the standard error of the skew for all
variables, fell below a z-score deviation of 2.03, which is within the limits of
assumed normality. The residuals appear to follow the regression line rather well
from an observation of the plot of the observed and expected residuals.

76

Table 8: Past sexual risk-taking behavior significantly predicts future sexual risk-taking behavior

Correlations
1.000 .456
.456 1.000
. .000
.000 .
97 97
97 97
UAI past 90 days
(follow-up)
UAI past 90 days (baseline)
UAI past 90 days
(follow-up)
UAI past 90 days (baseline)
UAI past 90 days
(follow-up)
UAI past 90 days (baseline)
Pearson Correlation
Sig. (1-tailed)
N
UAI past 90
days
(follow-up)
UAI past
90 days
(baseline)

77
Coefficients
a
-.007 .084 -.078 .938 -.173 .160
.598 .120 .456 5.000 .000 .360 .835 .456 .456 .456 1.000 1.000
(Constant)
UAI past 90 days (baseline)
Model
1
B Std. Error
Unstandardized
Coefficients
Beta
Standardized
Coefficients
t Sig.
Lower
Bound
Upper
Bound
95% Confidence Interval
for B
Zero-order Partial Part
Correlations
Tolerance VIF
Collinearity Statistics
Dependent Variable: UAI past 90 days (follow-up)
a.

Model Summary
.456
a
.208 .200 .38825 .208 24.997 1 95 .000
Model
1
R R Square
Adjusted R
Square
Std. Error of
the Estimate
R Square
Change F Change df1 df2
Sig. F
Change
Change Statistics
Predictors: (Constant), UAI past 90 days (baseline)
a.

Table 8: Past sexual risk-taking behavior significantly predicts future sexual risk-taking behavior (continued)

ANOVA
b
3.768 1 3.768 24.997 .000
a
14.320 95 .151
18.088 96
Regression
Residual
Total
Model
1
Sum of
Squares df
Mean
Square F Sig.
Predictors: (Constant), UAI past 90 days (baseline)
a.
Dependent Variable: UAI past 90 days (follow-up)
b.

Results H
6
: Virtual risk-taking will significantly predict past sexual risk-taking in
the IAV but not in the Yoked Condition.

A simultaneous regression was performed for both the IAV and Yoked
conditions, between number of unprotected sex acts with non-primary partners at
baseline (Time 1 –past 90 days) as the dependent variable and virtual sex risk-
taking (dummy coded –anal sex/other sex and condom/no condom) as independent
variables. Analysis was performed using SPSS regression and SPSS explore for
evaluation of assumptions. The samples consisted of 84 and 67 men who had UAI
at least once with a non-primary partner at Time 1 in the IAV and Yoked
conditions respectively.
Results of evaluation of assumptions led to transformations of the variables
to reduce skewness, eliminate outliers, and improve the normality, linearity, and
homoscedasticity of residuals. A log 10 transformation was used on the dependent
variable to reduce severe positive skew.
With the use of a p<.001 criterion for Mahlanobis distance, with 3 degrees
of freedom, no outliers were found. No missing data was present, n=84 in the IAV
condition and n=67 in the Yoked condition.
Table 9 displays the correlations between the variables, the unstandardized
regression coefficients (B) and intercept, the standardized regression coefficients
(B), the semi-partial correlations, R
2
, and adjusted R
2
. R for regression in the IAV
condition was significantly different from zero, F (2,81) = 14.412, p=.000, with R
2

at .262 and 95% confidence limits from (.463 to .639). The adjusted R
2
value of

78

79
.244 indicates that 26.2% of the variance of past UAI risk is predicted by virtual
risk-taking. One regression coefficient, condom/no condom, differed significantly
from zero. The confidence limits for the condom/no condom variable was .390 to
.860.
In the Yoked condition, the R for regression was not significantly different
from zero, F (2,64)= .139, p=.870, with R
2
at .004 indicating that less than .04%
of the variance of past UAI risk is predicted by Yoked virtual sex choices.
There is no problem with multi-collinearity. Outliers disappeared once the
variables were transformed, and a confirmation using plots and Mahlanobis
distance confirmed this. The skew divided by the standard error of the skew for all
variables, fell below a z-score deviation of 2.03, which is within the limits of
assumed normality. The residuals appear to follow the regression line rather well
from an observation of the plot of the observed and expected residuals.

80

Table 9: Virtual risk-taking predict past risk-taking better than the theory of planned behavior in the IAV but not in the Yoked
Condition
Correlations
1.000 -.084 .511
-.084 1.000 -.223
.511 -.223 1.000
. .224 .000
.224 . .021
.000 .021 .
84 84 84
84 84 84
84 84 84
1.000 .015 -.066
.015 1.000 -.170
-.066 -.170 1.000
. .453 .299
.453 . .084
.299 .084 .
67 67 67
67 67 67
67 67 67
UAI past 90 days (baseline)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
Pearson Correlation
Sig. (1-tailed)
N
Pearson Correlation
Sig. (1-tailed)
N
Condition?
IAV
Yoked
UAI past
90 days
(baseline)
dvd level of
sexual risk
other safer
types of
sex/anal sex
dvd level of
sexual risk
condom/no
condom

81

Table 9: Virtual risk-taking predict past risk-taking better than the theory of planned behavior in the IAV but not in the Yoked
condition (continued)

Model Summary
.512
a
.262 .244 .30951 .262 14.412 2 81 .000
.066
a
.004 -.027 .29496 .004 .139 2 64 .870
Model
1
1
Condition?
IAV
Yoked
R R Square
Adjusted R
Square
Std. Error of
the Estimate
R Square
Change F Change df1 df2
Sig. F
Change
Change Statistics
Predictors: (Constant), dvd level of sexual risk condom/no condom, dvd level of sexual risk other safer types of sex/anal sex
a.

Coefficients
a
.551 .044 12.464 .000 .463 .639
.024 .074 .032 .327 .744 -.123 .172 -.084 .036 .031 .950 1.052
.625 .118 .518 5.297 .000 .390 .860 .511 .507 .505 .950 1.052
.619 .046 13.600 .000 .528 .710
.002 .079 .003 .028 .978 -.155 .160 .015 .003 .003 .971 1.030
-.079 .154 -.065 -.515 .609 -.388 .229 -.066 -.064 -.064 .971 1.030
(Constant)
dvd level of sexual risk ot
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
(Constant)
dvd level of sexual risk ot
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
Model
1
1
Condition?
IAV
Yoked
B Std. Error
Unstandardized
Coefficients
Beta
Standardized
Coefficients
t Sig.
Lower
Bound
Upper
Bound
95% Confidence Interval
for B
Zero-order Partial Part
Correlations
Tolerance VIF
Collinearity Statistics
Dependent Variable: UAI past 90 days (baseline)
a.

82

Table 9: Virtual risk-taking predict past risk-taking better than the theory of planned behavior in the IAV but not in the Yoked
condition (continued)

ANOVA
b
2.761 2 1.381 14.412 .000
a
7.760 81 .096
10.521 83
.024 2 .012 .139 .870
a
5.568 64 .087
5.592 66
Regression
Residual
Total
Regression
Residual
Total
Model
1
1
Condition?
IAV
Yoked
Sum of
Squares df
Mean
Square F Sig.
Predictors: (Constant), dvd level of sexual risk condom/no condom, dvd level of sexual risk other safer
types of sex/anal sex
a.
Dependent Variable: UAI past 90 days (baseline)
b.

Results H
7
: Virtual risk-taking will significantly predict future sexual risk-taking in
the IAV but not in the Yoked Condition.

A simultaneous regression was performed for both the IAV and Yoked
conditions, between number of unprotected sex acts with non-primary partners at
follow-up (Time 2 –past 90 days) as the dependent variable and virtual sex risk-
taking (dummy coded –anal sex/other sex and condom/no condom) as independent
variables. Analysis was performed using SPSS regression and SPSS explore for
evaluation of assumptions. The samples consisted of 51 and 46 men who had UAI
at least once with a non-primary partner at Time 1 in the IAV and Yoked
conditions respectively.
Results of evaluation of assumptions led to transformations of the variables
to reduce skewness, eliminate outliers, and improve the normality, linearity, and
homoscedasticity of residuals. A log 10 transformation was used on the dependent
variable to reduce severe positive skew.
With the use of a p<.001 criterion for Mahlanobis distance, with 3 degrees
of freedom, no outliers were found. No missing data was present, n=51 in the IAV
condition and n=46 in the Yoked condition.
Table 10 displays the correlations between the variables, the
unstandardized regression coefficients (B) and intercept, the standardized
regression coefficients (B), the semi-partial correlations, R
2
, and adjusted R
2
. R for
regression in the IAV condition was significantly different from zero, F (2,48) =
4.439, p=.017, with R
2
at .156 and 95% confidence limits from (.181 to .520). The

83

84
adjusted R
2
value of .121 indicates that 12.1% of the variance of future UAI risk-
taking is predicted by virtual risk-taking. One regression coefficient, condom/no
condom, differed significantly from zero. The confidence limits for the
condom/no condom variable was .174 to .993.
In the Yoked condition, the R for regression was not significantly different
from zero, F (2,43)= 1.810, p=.176, with R
2
at .078 and an adjusted R
2
of .035
indicating that 3.5% of the variance of future UAI risk is predicted by Yoked
virtual sex choices. There is no problem with multi-collinearity. Outliers
disappeared once the variables were transformed, and a confirmation using plots
and Mahlanobis distance confirmed this. The skew divided by the standard error
of the skew for all variables, fell below a z-score deviation of 2.03, which is within
the limits of assumed normality. The residuals appear to follow the regression line
rather well from an observation of the plot of the observed and expected residuals.

85

Table 10: Virtual risk-taking predicts future risk-taking better than the theory of planned behavior in the
IAV but not in the Yoked condition
Correlations
1.000 -.108 .395
-.108 1.000 -.247
.395 -.247 1.000
. .226 .002
.226 . .040
.002 .040 .
51 51 51
51 51 51
51 51 51
1.000 .236 -.180
.236 1.000 -.141
-.180 -.141 1.000
. .057 .116
.057 . .175
.116 .175 .
46 46 46
46 46 46
46 46 46
UAI past 90 days
(follow-up)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days
(follow-up)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days
(follow-up)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days
(follow-up)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days
(follow-up)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days
(follow-up)
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
Pearson Correlation
Sig. (1-tailed)
N
Pearson Correlation
Sig. (1-tailed)
N
Condition?
IAV
Yoked
UAI past 90
days
(follow-up)
dvd level of
sexual risk
other safer
types of
sex/anal sex
dvd level of
sexual risk
condom/no
condom

86

Table 10: Virtual risk-taking predicts future risk-taking better than the theory of planned behavior in the IAV but not in the Yoked
condition (continued)
Model Summary
.395
a
.156 .121 .45401 .156 4.439 2 48 .017
.279
a
.078 .035 .36101 .078 1.810 2 43 .176
Model
1
1
Condition?
IAV
Yoked
R R Square
Adjusted R
Square
Std. Error of
the Estimate
R Square
Change F Change df1 df2
Sig. F
Change
Change Statistics
Predictors: (Constant), dvd level of sexual risk condom/no condom, dvd level of sexual risk other safer types of sex/anal sex
a.

Coefficients
a
.351 .084 4.161 .000 .181 .520
-.011 .141 -.011 -.080 .936 -.296 .273 -.108 -.012 -.011 .939 1.065
.584 .204 .392 2.866 .006 .174 .993 .395 .382 .380 .939 1.065
.267 .066 4.047 .000 .134 .400
.170 .117 .215 1.453 .154 -.066 .405 .236 .216 .213 .980 1.020
-.267 .264 -.150 -1.012 .317 -.798 .265 -.180 -.152 -.148 .980 1.020
(Constant)
dvd level of sexual risk othe
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
(Constant)
dvd level of sexual risk othe
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
Model
1
1
Condition?
IAV
Yoked
B Std. Error
Unstandardized
Coefficients
Beta
Standardized
Coefficients
t Sig.
Lower
Bound
Upper
Bound
95% Confidence Interval
for B
Zero-order Partial Part
Correlations
Tolerance VIF
Collinearity Statistics
Dependent Variable: UAI past 90 days (follow-up)
a.

87

Table 10: Virtual risk-taking predicts future risk-taking better than the theory of planned behavior in the IAV but not in the Yoked
condition (continued)

ANOVA
b
1.830 2 .915 4.439 .017
a
9.894 48 .206
11.724 50
.472 2 .236 1.810 .176
a
5.604 43 .130
6.076 45
Regression
Residual
Total
Regression
Residual
Total
Model
1
1
Condition?
IAV
Yoked
Sum of
Squares df
Mean
Square F Sig.
Predictors: (Constant), dvd level of sexual risk condom/no condom, dvd level of sexual risk other safer
types of sex/anal sex
a.
Dependent Variable: UAI past 90 days (follow-up)
b.

Results H
8
: The theory of planned behavior combined with virtual risk-taking will
account for the majority of the variance, mediating the residual effect of past risk-
taking behavior

A stepwise hierarchical regression was performed for both the IAV and
Yoked conditions, between number of unprotected sex acts with non-primary
partners at follow-up (Time 2- past 90 days) as the dependent variable and on the
first step 1) intentions to use condoms at baseline, self-efficacy at baseline; on the
second step 2) virtual risk-taking (dummy coded- anal sex or other safer kind of
sex, and condom/no condom); on the third step 3) past risk-taking behavior -as
independent variables. Analysis was performed using SPSS regression and SPSS
explore for evaluation of assumptions. The sample consisted of 51 and 46 men
who had UAI at least once with a non-primary partner at Time 1 in the IAV and
Yoked conditions respectively.
Results of evaluation of assumptions led to transformations of the variables
to reduce skewness, eliminate outliers, and improve the normality, linearity, and
homoscedasticity of residuals. A log 10, reflect transformation was used on the
measures of intentions to use condoms and self-efficacy to reduce severe negative
skew. A log 10 transformation was used on the dependent variable to reduce
severe positive skew.
With the use of a p<.001 criterion for Mahlanobis distance, with 3 degrees of
freedom, no outliers were found. No missing data was present, n=51 in the IAV
and n=46 in the Yoked condition.

88

Table 11 displays the correlations between the variables, the unstandardized
regression coefficients (B) and intercept, the standardized regression coefficients
(B), the semi-partial correlations, R
2
, and adjusted R
2
. R for regression in the IAV
condition was significantly different from zero, F (5,45) = 3.167, p=.010, with R
2

on the first step at .118 (TPB), R
2
on the second step at .234, R
2
on the third step
.260. The adjusted R
2
value of .082 indicates that 8.2% of the variance in
unprotected anal sex at Time 2 is predicted by the TPB. The adjusted R
2
value of
.168 on the second step indicates that virtual risk-taking accounts for an additional
8.6% of the variance beyond the TPB. The adjusted R
2
value of .178 on the third
step indicates only an additional 1% of the variance in future risk-taking is
accounted for by past behavior. Two regression coefficients, whether the person
had safe or unsafe UAI in the IAV & intentions to use condoms, both differed
significantly from zero. The confidence limits for the condom/no condom variable
was .037 to .886. The confidence limits for intentions to use condoms was .037 to
1.000.
The 4 variables in combination on the first two steps in the IAV condition
accounted for nearly 16.8% of the variance. It appears that the first dummy coded
variable (other safer sex or anal sex) contributes nothing to the model (Beta =
.074). There is no problem with multi-collinearity. Outliers disappeared once the
variables were transformed, and a confirmation using plots and Mahlanobis
distance confirmed this. The skew divided by the standard error of the skew for all
variables, fell below a z-score deviation of 2.03, which is within the limits of

89

90
assumed normality. The residuals appear to follow the regression line rather well
from an observation of the plot of the observed and expected residuals.
R for regression in the Yoked condition was significantly different from zero,
F (5,40) = 5.048, p=.001, with R
2
on the first step at .191 (TPB), R
2
on the second
step at .230, R
2
on the third step .387. The adjusted R
2
value of .154 indicates that
15.4% of the variance in unprotected anal sex at Time 2 is predicted by the TPB.
The adjusted R square value of .155 on the second step indicates that virtual risk-
taking accounts for an additional .001% of the variance beyond the TPB. The
adjusted R square value of .310 on the third step indicates that an additional 16.5%
of the variance in future risk-taking is accounted for by past behavior. One
regression coefficient, intentions to use condoms differed significantly from zero
on the first step. The confidence limits for intentions to use condoms was .154 to
.791

91

Table 11: The theory of planned behavior combined with virtual risk-taking accounts for the majority of
the variance, making the residual effect of past-risk taking behavior no longer significant

Correlations
1.000 .322 .065 -.108 .395 .415
.322 1.000 .521 .264 .249 .403
.065 .521 1.000 .208 .163 -.046
-.108 .264 .208 1.000 -.247 -.037
.395 .249 .163 -.247 1.000 .452
.415 .403 -.046 -.037 .452 1.000
. .011 .326 .226 .002 .001
.011 . .000 .031 .039 .002
.326 .000 . .071 .127 .374
.226 .031 .071 . .040 .398
.002 .039 .127 .040 . .000
.001 .002 .374 .398 .000 .
51 51 51 51 51 51
51 51 51 51 51 51
51 51 51 51 51 51
51 51 51 51 51 51
51 51 51 51 51 51
51 51 51 51 51 51
1.000 .437 .152 .236 -.180 .541
.437 1.000 .397 .140 -.161 .404
.152 .397 1.000 -.084 -.004 .061
.236 .140 -.084 1.000 -.141 .099
-.180 -.161 -.004 -.141 1.000 -.056
.541 .404 .061 .099 -.056 1.000
. .001 .157 .057 .116 .000
.001 . .003 .177 .143 .003
.157 .003 . .289 .488 .344
.057 .177 .289 . .175 .256
.116 .143 .488 .175 . .356
.000 .003 .344 .256 .356 .
46 46 46 46 46 46
46 46 46 46 46 46
46 46 46 46 46 46
46 46 46 46 46 46
46 46 46 46 46 46
46 46 46 46 46 46
UAI past 90 days
(follow-up)
Intentions Post-Intervention
Efficacy Post-Intervention
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
UAI past 90 days
(follow-up)
Intentions Post-Intervention
Efficacy Post-Intervention
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
UAI past 90 days
(follow-up)
Intentions Post-Intervention
Efficacy Post-Intervention
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
UAI past 90 days
(follow-up)
Intentions Post-Intervention
Efficacy Post-Intervention
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
UAI past 90 days
(follow-up)
Intentions Post-Intervention
Efficacy Post-Intervention
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
UAI past 90 days
(follow-up)
Intentions Post-Intervention
Efficacy Post-Intervention
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
Pearson Correlation
Sig. (1-tailed)
N
Pearson Correlation
Sig. (1-tailed)
N
Condition?
IAV
Yoked
UAI past 90
days
(follow-up)
Intentions
Post-
Intervention
Efficacy
Post-
Intervention
dvd level of
sexual risk
other safer
types of
sex/anal sex
dvd level of
sexual risk
condom/no
condom
UAI past
90 days
(baseline)

92

Table 11: The theory of planned behavior combined with virtual risk-taking accounts for the majority of
the variance, making the residual effect of past-risk taking behavior no longer significant (continued)

Model Summary
.344
a
.118 .082 .46405 .118 3.222 2 48 .049
.484
b
.234 .168 .44179 .116 3.479 2 46 .039
.510
c
.260 .178 .43899 .026 1.590 1 45 .214
.438
a
.191 .154 .33801 .191 5.090 2 43 .010
.480
b
.230 .155 .33774 .039 1.035 2 41 .364
.622
c
.387 .310 .30517 .157 10.218 1 40 .003
Model
1
2
3
1
2
3
Condition?
IAV
Yoked
R R Square
Adjusted R
Square
Std. Error of
the Estimate
R Square
Change F Change df1 df2
Sig. F
Change
Change Statistics
Predictors: (Constant), Efficacy Post-Intervention, Intentions Post-Intervention
a.
Predictors: (Constant), Efficacy Post-Intervention, Intentions Post-Intervention, dvd level of sexual risk condom/no condom, dvd level of sexual risk
other safer types of sex/anal sex
b.
Predictors: (Constant), Efficacy Post-Intervention, Intentions Post-Intervention, dvd level of sexual risk condom/no condom, dvd level of sexual risk
other safer types of sex/anal sex, UAI past 90 days (baseline)
c.

93

Table 11: The theory of planned behavior combined with virtual risk-taking accounts for the majority of the variance,
making the residual effect of past-risk taking behavior no longer significant (continued)
ANOVA
d
1.387 2 .694 3.222 .049
a
10.336 48 .215
11.724 50
2.745 4 .686 3.517 .014
b
8.978 46 .195
11.724 50
3.052 5 .610 3.167 .016
c
8.672 45 .193
11.724 50
1.163 2 .582 5.090 .010
a
4.913 43 .114
6.076 45
1.399 4 .350 3.067 .027
b
4.677 41 .114
6.076 45
2.351 5 .470 5.048 .001
c
3.725 40 .093
6.076 45
Regression
Residual
Total
Regression
Residual
Total
Regression
Residual
Total
Regression
Residual
Total
Regression
Residual
Total
Regression
Residual
Total
Model
1
2
3
1
2
3
Condition?
IAV
Yoked
Sum of
Squares df
Mean
Square F Sig.
Predictors: (Constant), Efficacy Post-Intervention, Intentions Post-Intervention
a.
Predictors: (Constant), Efficacy Post-Intervention, Intentions Post-Intervention, dvd level of sexual
risk condom/no condom, dvd level of sexual risk other safer types of sex/anal sex
b.
Predictors: (Constant), Efficacy Post-Intervention, Intentions Post-Intervention, dvd level of sexual
risk condom/no condom, dvd level of sexual risk other safer types of sex/anal sex, UAI past 90 days
(baseline)
c.
Dependent Variable: UAI past 90 days (follow-up)
d.

94

Co fficients
a
e
.209 .146 1.430 .159 -.085 .503
.594 .238 .396 2.493 .016 .115 1.072 .322 .339 .338 .729 1.373
-.190 .213 -.142 -.892 .377 -.618 .238 .065 -.128 -.121 .729 1.373
.231 .139 1.654 .105 -.050 .511
.518 .239 .346 2.166 .036 .037 1.000 .322 .304 .279 .654 1.530
-.197 .204 -.147 -.965 .340 -.607 .214 .065 -.141 -.124 .719 1.390
-.095 .148 -.092 -.644 .523 -.392 .202 -.108 -.095 -.083 .817 1.224
.461 .211 .310 2.187 .034 .037 .886 .395 .307 .282 .828 1.207
.098 .174 .567 .574 -.251 .448
.363 .268 .242 1.359 .181 -.175 .902 .322 .199 .174 .516 1.937
-.094 .218 -.070 -.430 .669 -.534 .346 .065 -.064 -.055 .619 1.615
-.096 .147 -.093 -.655 .516 -.391 .199 -.108 -.097 -.084 .817 1.224
.341 .230 .229 1.481 .145 -.123 .805 .395 .216 .190 .686 1.457
.264 .210 .208 1.261 .214 -.158 .687 .415 .185 .162 .607 1.648
.086 .100 .855 .397 -.116 .287
.472 .158 .447 2.993 .005 .154 .791 .437 .415 .410 .842 1.187
-.025 .145 -.026 -.174 .863 -.318 .268 .152 -.026 -.024 .842 1.187
.061 .107 .573 .570 -.155 .278
.417 .162 .395 2.568 .014 .089 .745 .437 .372 .352 .794 1.260
.008 .147 .008 .056 .956 -.289 .305 .152 .009 .008 .820 1.219
.133 .111 .168 1.194 .239 -.092 .358 .236 .183 .164 .945 1.058
-.165 .250 -.093 -.663 .511 -.669 .339 -.180 -.103 -.091 .958 1.044
-.218 .130 -1.673 .102 -.482 .045
.211 .160 .200 1.314 .196 -.113 .535 .437 .203 .163 .665 1.504
.057 .134 .059 .425 .673 -.213 .327 .152 .067 .053 .810 1.235
.123 .101 .155 1.220 .230 -.081 .326 .236 .189 .151 .944 1.059
-.181 .226 -.101 -.801 .428 -.636 .275 -.180 -.126 -.099 .957 1.045
.589 .184 .436 3.197 .003 .217 .962 .541 .451 .396 .824 1.213
(Constant)
Intentions Post-Intervention
Efficacy Post-Intervention
(Constant)
Intentions Post-Intervention
Efficacy Post-Intervention
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
(Constant)
Intentions Post-Intervention
Efficacy Post-Intervention
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
(Constant)
Intentions Post-Intervention
Efficacy Post-Intervention
(Constant)
Intentions Post-Intervention
Efficacy Post-Intervention
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
(Constant)
Intentions Post-Intervention
Efficacy Post-Intervention
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
Model
1
2
3
1
2
3
Condition?
IAV
Yoked
B Std. Error
Unstandardized
Coefficients
Beta
Standardized
Coefficients
t Sig.
Lower
Bound
Upper
Bound
95% Confidence Interval
for B
Zero-order Partial Part
Correlations
Tolerance VIF
Collinearity Statistics
Dependent Variable: UAI past 90 days (follow-up) a.

Table 11: The theory of planned behavior combined with virtual risk-taking accounts for the majority of
the variance, making the residual effect of past-risk taking behavior no longer significant (continued)

Post-hoc Analysis

Based on the partial (r= -.128) and semi-partial (r= -.121) correlations in
H
7
, self-efficacy appears to be acting as a suppressor variable. In other words,
self-efficacy measures do not account for any significant unique variance but
rather may be helping to account for additional error variance among intentions.
This finding is consistent with prior research by Rye, Fisher and Fisher (2001)
which found that perceived behavioral control does not contribute significantly to
the prediction of any range of safer sex behaviors among MSM over and above the
predictive power of intentions. Young MSM who attempt to negotiate safer sex
online but then lack the requisite skills to subsequently follow through on their
behavioral intentions may be have overestimated their perceived self-efficacy.
In the IAV sample subset,, efficacy measures actually predicted greater, not
reduced UAI risk-taking. Such suppressor effects suggest that many individuals
who perceive themselves capable of (or efficacious) successfully initiating,
negotiating, and using safer sex (and refusing to engage in unsafe sex) may be
succumbing to biases in “affective forecasting” in which under “cold” emotional
state (e.g. not sexually aroused) they fail to take into account “hot” emotional
states (e.g., sexual desire) and how this state will impact their self-efficacy (see for
example, Gilbert, Gill, & Wilson, 2002; Loewenstein, O’Donoghue, & Rabin,
2003) .

95

This may produce a false sense of security in entering a decision-making
path (e.g., going to a club; taking drugs or alcohol before sex) that will be far
riskier than they anticipate, which the interactive video is able to capture while self
reported data cannot. In order to test whether this is true or not the following
hypothesis is proposed:
H
9
: Removing the overconfident individuals (highly self-efficacious) will
eliminate the suppressor effect of self-efficacy found in predicting to risk-
taking behavior.

In order to test this hypothesis, a post-hoc analysis was performed after
removing those participants who gave the highest self-efficacy scores (top 1%).
On a histogram, this group stands out as a small bump on an otherwise normal
distribution.
A simultaneous regression was performed between number of unprotected
sex acts with non-primary partners at baseline (TIME 1) as the dependent variable
and intentions to use condoms and self-efficacy at baseline, and DVD choice
(dummy coded- anal sex or other safer kind of sex, and protected or unprotected
anal sex) as independent variables. Those individuals who claimed to be highly
self-efficacious, i.e. the top 1% of those giving the highest possible ratings were
excluded, as a possibly overconfident group. Analysis was performed using SPSS
regression and SPSS explore for evaluation of assumptions. The sample consisted
of 37 men who had UAI at least once with a non-primary partner at Time 1.
Results of evaluation of assumptions led to transformations of the variables
to reduce skewness, eliminate outliers, and improve the normality, linearity, and

96

homoscedasticity of residuals. A log 10, reflect transformation was used on the
measures of intentions to use condoms and self-efficacy to reduce severe negative
skew. A log 10 transformation was used on the dependent variable to reduce
severe positive skew.
With the use of a p<.001 criterion for Mahlanobis distance, with 4degrees
of freedom, no outliers were found. No missing data was present, n=37.
Table 12 displays the correlations between the variables, the
unstandardized regression coefficients (B) and intercept, the standardized
regression coefficients (B), the semi-partial correlations, R
2
, and adjusted R
2
. R for
regression was significantly different from zero, F (4,32) = 5.602, p=.002, with R
2

at .148 for the TPB at step 1 and R
2
at .412 at step two. The adjusted R
2
value of
.338 indicates that 24% of the variance in unprotected anal sex at Time 1 is
predicted by virtual risk taking variables. An additional 9.8% of the variance is
predicted by the Theory of Planned Behavior, intentions to use condoms and self-
efficacy. Two regression coefficients, whether the person had safe or unsafe UAI
in the IAV & intentions to use condoms, both differed significantly from zero.
The confidence limits for the condom/no condom variable was .247 to .912. The
confidence limits for the intentions to use condoms (post-intervention) was .008 to
.896.
The 4 variables in combination accounted for nearly 26.4% of the variance,
this accounting of the variance is higher than the 20% accounted for by just
regressing past behavior onto future behavior as shown in Table 7. There is no

97

98
problem with multi-collinearity. Outliers disappeared once the variables were
transformed, and a confirmation using plots and Mahlanobis distance confirmed
this. The skew divided by the standard error of the skew for all variables, fell
below a z-score deviation of 2.03, which is within the limits of assumed normality.
The residuals appear to follow the regression line rather well from an observation
of the plot of the observed and expected residuals.
The disappearance of the suppressor effect suggests that it may be possible
to actively train individuals to better assess their self-efficacy by actively
rehearsing, and experiencing decision-making under ‘hot emotional’ states. The
bias in affective forecasting may be due to the visceral nature sexual arousal can
evoke in people. Visceral feelings are feeling that have direct, hedonic impacts on
people’s behavior (e.g. hunger, thirst, sexual desire, pain, drug addiction). The
strong feelings that are evoked ‘in the heat of the moment’ are difficult to recall in
as vivid a sense as the actual moment they occur and are also not easily anticipated
(Lowenstein & Schkade, 1999; Lowenstein, 2001).
Because the IAV may be able to capture the difference in self-efficacy that
individuals might not recognize when not in a ‘hot state’, it may also, through
increased choices and scenarios allow them to realistically assess their self-
efficacy in a ‘hot state’ over time and therefore make better judgments about
whether to initially place themselves in situations that they may not realistically be
able to negotiate effectively under.

99

Table 12: Post-hoc analysis- removing overconfident group increases the predictive ability of virtual risk taking

Correlations
1.000 .385 .203 -.192 .587
.385 1.000 .497 -.048 .255
.203 .497 1.000 .119 .289
-.192 -.048 .119 1.000 -.257
.587 .255 .289 -.257 1.000
. .009 .114 .127 .000
.009 . .001 .389 .064
.114 .001 . .242 .041
.127 .389 .242 . .062
.000 .064 .041 .062 .
37 37 37 37 37
37 37 37 37 37
37 37 37 37 37
37 37 37 37 37
37 37 37 37 37
UAI past 90 days (baseline)
Intentions Baseline
Efficacy Baseline
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
Intentions Baseline
Efficacy Baseline
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
UAI past 90 days (baseline)
Intentions Baseline
Efficacy Baseline
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
Pearson Correlation
Sig. (1-tailed)
N
UAI past
90 days
(baseline)
Intentions
Baseline
Efficacy
Baseline
dvd level of
sexual risk
other safer
types of
sex/anal sex
dvd level of
sexual risk
condom/no
condom

100

Table 12: Post-hoc analysis- removing overconfident group increases the predictive ability of virtual risk taking (continued)
Model Summary
.385
a
.148 .098 .35936 .148 2.959 2 34 .065
.642
b
.412 .338 .30781 .264 7.171 2 32 .003
Model
1
2
R R Square
Adjusted R
Square
Std. Error of
the Estimate
R Square
Change F Change df1 df2
Sig. F
Change
Change Statistics
Predictors: (Constant), Efficacy Baseline, Intentions Baseline
a.
Predictors: (Constant), Efficacy Baseline, Intentions Baseline, dvd level of sexual risk other safer types of sex/anal sex, dvd level of
sexual risk condom/no condom
b.

Coefficients
a
.231 .297 .778 .442 -.372 .834
.452 .219 .377 2.067 .046 .008 .896 .385 .334 .327 .753 1.328
.030 .351 .016 .086 .932 -.683 .744 .203 .015 .014 .753 1.328
.437 .261 1.678 .103 -.094 .968
.353 .190 .294 1.858 .072 -.034 .739 .385 .312 .252 .734 1.363
-.179 .315 -.093 -.568 .574 -.820 .462 .203 -.100 -.077 .688 1.453
-.025 .118 -.031 -.213 .833 -.265 .215 -.192 -.038 -.029 .886 1.128
.580 .163 .531 3.550 .001 .247 .912 .587 .532 .481 .822 1.217
(Constant)
Intentions Baseline
Efficacy Baseline
(Constant)
Intentions Baseline
Efficacy Baseline
dvd level of sexual risk other
safer types of sex/anal sex
dvd level of sexual risk
condom/no condom
Model
1
2
B Std. Error
Unstandardized
Coefficients
Beta
Standardized
Coefficients
t Sig.
Lower
Bound
Upper
Bound
95% Confidence Interval
for B
Zero-order Partial Part
Correlations
Tolerance VIF
Collinearity Statistics
Dependent Variable: UAI past 90 days (baseline)
a.

101

Table 12: Post-hoc analysis- removing overconfident group increases the predictive ability of virtual risk taking (continued)

ANOVA
c
.764 2 .382 2.959 .065
a
4.391 34 .129
5.155 36
2.123 4 .531 5.602 .002
b
3.032 32 .095
5.155 36
Regression
Residual
Total
Regression
Residual
Total
Model
1
2
Sum of
Squares df
Mean
Square F Sig.
Predictors: (Constant), Efficacy Baseline, Intentions Baseline
a.
Predictors: (Constant), Efficacy Baseline, Intentions Baseline, dvd level of sexual risk
other safer types of sex/anal sex, dvd level of sexual risk condom/no condom
b.
Dependent Variable: UAI past 90 days (baseline)
c.

102

Hypothesis

Supported

H
1
: Interactive Video Choices will be related to past risk-taking behavior
Yes
X
2
p<.05 for alcohol use, drug use,
& sexual risk-taking

H
2
: Interactive Video Choices will predict future risk-taking behavior

Yes
X
2
p<.05 for alcohol use, drug use,
& sexual risk-taking

H
3
: Theory of Planned Behavior Will Significantly Predict Past Risk-Taking Behavior (postdictive)
Yes
Adjusted R
2
=.044, p<.05

H
4
: The Theory of Planned Behavior Will Significantly Predict Future Risk-Taking Behavior

Yes
Adjusted R
2
=.124, p<.05

H
5
: Past sexual risk-taking behavior will significantly predict future sexual risk-taking behavior

Yes
Adjusted R
2
=.200, p<.05

H
6
: Virtual risk-taking will significantly predict past sexual risk-taking in the IAV but not in the Yoked Condition.

Yes
IAV Adjusted R
2
= .244, p<.05
Yoked: Adjusted R
2
= .027, p>.05

H
7
: Virtual risk-taking will significantly predict future sexual risk-taking in the IAV but not in the Yoked Condition.

Yes
IAV Adjusted R
2
= .121, p<.05
Yoked Adjusted R
2
= .035, p>.05

H
8
: The theory of planned behavior combined with virtual risk-taking will account for the majority of the variance,
mediating the role of past risk-taking behavior in the IAV but not in the Yoked condition.

Yes
TPB/IAV Adjusted R
2
= .168, p<.05
Past behavior Adjusted R
2
=.026 , p>.05
additional variance
TPB/Yoked Adjusted R
2
=.155, p<.05
Past behavior Adjusted R
2
=.157, p<.05
additional variance
Post-Hoc: Self-Efficacy acts as a suppressor variable in the IAV condition Yes

Table 13: Summary of Results

CHAPTER 8:
DISCUSSION

In the current work, four main questions led to eight major hypotheses.
These questions were: First, what can choices made within a virtual environment -
based on scenarios likely to be encountered in the real world- tell us about past
and future real-life decision-making and behavior? Second, are we are able to
account for significant variance in future real-life risk behaviors using traditional
cognitive measures found in the Theory of Planned Behavior (e.g. intentions, self-
efficacy)? Third, do virtual decisions account for unique variance in real-life risk-
taking beyond traditional more cognitive self-report predictors (e.g., intent, self-
efficacy)? And finally, might these traditional and virtual predictors in
combination mediate the variance traditionally associated with “past behavior”
when individuals actively make interactive choices instead of passively observing
another’s choices?
Below, we consider each of these questions in turn, discussing the
implications of the findings associated with each of the relevant hypotheses.

103

How are Choices Within Virtual Environments
Related to Real-world Decision-making and Behavior?

Review of Findings

The research question posed above led to two hypotheses. The first
hypothesis asked whether virtual risk-taking behavior (based on scenarios similar
to what they might encounter in their real-life) was related to past real-life risk-
taking. In the current work, MSM were significantly more likely to make virtual
behavioral choices (i.e., having unsafe sex, using drugs, using alcohol) in the IAV
consistent with their respective past real-life behavior. For example, MSM who
had unsafe sex as a receptive partner in the past 90 days, were significantly more
likely to also do so when given the opportunity, within the virtual environment.
Those who used alcohol or methamphetamine in the past 90 days were more
likely to want to use alcohol and use methamphetamine respectively in the IAV.
The second hypothesis, utilizing study set one, went a step further by also
asking whether risky choices made in a virtual environment predict behavior 3
months later into the future. This longitudinal study found that risk-taking
behavior in a virtual environment was indeed capable of significantly predicting
real-life risk-taking into the future (e.g., those who engaged in unsafe sex or drank
alcohol in the IAV were significantly likely to do so into the future).

104

Fit With and Advances Beyond Past Research
These findings are consistent with past research, mentioned earlier, that
has shown a link between individual difference variables and virtual behaviors
within interactive environments (Bailenson, 2007; Yee, 2007; Griebel, 2006;
Whang & Geunyoung, 2004; Wellman, 2001; Lieberman, 2000). Prior to this
current study, however, the literature has only been able to effectively link
individual difference variables with virtual behavior, not virtual behavior with
either prior or subsequent actual behavior (for virtual rehabilitation lab based
training exceptions see Rizzo, 2001). Thus, this is the first study to demonstrate
the relationship between people's decisions in realistic virtual situations and their
past or future real-life decisions.
Establishing this link between virtual and real life behavior in terms of
active decision-making has three major implications. First, it confirms the
ecological validity of using SOLVE to examine risky decision making for
individual MSM. Ecological validity is usually examined by noting that the
patterns found in the lab are also found in comparable samples in comparable
real-world phenomena (Yee, 2007; Bailenson, 2007; Blascovich, 2002; Rizzo,
2004). But, that work rarely is able to directly link an individual’s chosen
behavior in the experimental setting to an individuals’ chosen behavior in similar
contexts in the real world over time (for magnetic imaging neuroscience
exceptions see Chaytor et al., 2004; Weber, 2006). Being able to readily do so,
and the potential to examine ecological validity more broadly, is a major
methodological advance of the SOLVE approach.

105

Second, the current work suggests the potential value of virtual
environments for unobtrusively diagnosing, predicting, and understanding the
circumstances under which real-life risk-taking might take place. Furthermore, if
we knew that the link between the virtual and the real was generalizable, then we
could attempt to assess behavior across multiple contexts at different points in
time, in order to determine with finer granularity the degree to which the social
(e.g., willingness), the physiological (e.g., arousal state), or personal (e.g.,
personality) cause certain individuals to take risks.
As Miller et al. (2007) have argued, changing behavior at the level of the
individual would then be easier with these more tailored risk-taking profiles
which would be a great advance above traditional methods, that seek to change
behavior based on theories that have been confirmed to work on a macro-level
(e.g., experimental comparisons of groups of subjects). Standard interventions,
although effective, don’t always work for everyone. What if we could examine
risk on the micro and individual level?
Thirdly, in the case of virtual environments, if risk-taking behavior is
found to be congruent with behavior in real-life, than this should mean that the
environment created contains enough realism and salience such that the subject
doesn’t ‘feel’ like they are in an experimental setting. Gilbert (1991) argues that
people believe in the ideas they comprehend as quickly and automatically as they
believe in the objects they see. This suggests that it might be far easier to alter
mental models in virtual environments where congruent behavior is elicited. Thus
in the health intervention context, there is much to be gained if we can maximize

106

this link between the virtual and the real, because it might mean that participants
may be less likely to question the various aspects of the intervention itself (or
experience reactance). An important question for further research is how can we,
as researchers, maximize this link between the virtual and the real?

What Processes Facilitate the Link between Virtual and Real-life Behavior?
In the current work, the focus was on exploring the link between virtual
and real-life decision-making and behavior. The goal of future research is to
examine why we see this link. One promising explanation that has been offered
(Biocca, 1997, Lenggenhager, et al., 2007, Metzinger, 2003) is that interactive
environments induce a sense of self
4
which may lead people to treat their virtual
self “as if” it is their real self. Because of this inducement, people may rely on
their real-life experiences to guide their behavior in virtual environments. In other
words, they chose to treat their virtual self “as if” it is their real self (Biocca,
1997), and just like in real-life, they look to their past experiences to guide their
future behavior (Bandura, 2000).
Social presence and behavioral realism. Blascovich (2001) argues that
this sense of self is most likely to be induced when two factors are present: social
presence and behavioral realism. He found that these factors along with self-

107

4
Biocca argues that three ‘bodies’ are present in the virtual world: the actual body, the virtual body,
and the body schema, or the user’s mental model of self. When you direct a graphic representation
of yourself or are asked to make choices within a virtual environment, the representational state
evokes mental models of your own self.

relevance and interactive responsiveness, are likely to influence how people
behave in concordance with how they would between virtual and real life within
his virtual lab based experimental settings. Virtual and real-life behavior may be
especially likely to be congruent when the actions and situations are realistic and
probable (Biocca, 1997). In the present case, self-relevance is likely to be
extremely high in the IAV, given that we conducted formative research to create
realistic interactive videos (Godoy, et. al. 2005) and that the sample was
preselected for some baseline level of risk-taking (at least one instance of UAI
with a non-primary partner in the past 90 days).
Another possible way to increase social presence, which is consistent with
Blascovich’s method to evoke a stronger sense of self, might be to make the
virtual self (e.g., avatar or character) more (vs. less) similar to the real self. For
example, Bailensen found that embodied agents that were capable of mimicking
the behavior of the real person (e.g., moving their head exactly the same way,
nodding in sync, similar posture) were more persuasive and likable. This effect
occurred despite the fact that the subtle differences made it impossible to
explicitly detect the mimicry. The subjects seemed to identify with them on an
unconscious level (Bailensen & Yee, 2005). Blascovich et al. (2001) also found
similar results when morphing the participants face onto their avatar. He was able
to increase the effect of the laboratory study through this method.
Making the self salient. Other factors could also increase the likelihood of
evoking a “sense of self” in virtual environments. One possibility is by making
the self more (vs. less) salient. Duval and Wicklund (1972), for example, were

108

able to enhance self-focused attention using mirrors and cameras. They found
that when self-focused, individuals tend to think more about their traits, attitudes,
and feelings, and behave more consistently in regard to them.
Other researchers have made the self more salient through other methods.
Klinger et al (2006) demonstrated that social phobias can be treated by making
the public self the focus of attention using virtual reality environments. In one of
their experiments, for example, the researchers created a dinner party, in which
every character would stare at the client no matter which way he or she walked or
sat. This made clients extremely self-conscious, but they eventually became
desensitized to their own self-consciousness over time.
Bailenson (2007) found that when individuals are exposed to a mirror in a
1
st
person perspective virtual environment, such that their reflected avatar is of a
different ethnicity than their ‘real’ self, it subsequently influences their level of
racial prejudice on subsequent post-test items. Similarly, Meyers and Biocca
(1992) found that implicit representation of the self can be influenced by media,
such that distortions in body schema can result from exposure to implicit
representations of the self that emphasize ideal body shapes.
Lenggenhager et al. (2007) also found that it is possible to manipulate this
bodily self-consciousness. They found that a virtual body placed 10 feet in front
of an individual in a VR space caused the person to later misinterpret their true
position outside of where they were standing in the real world. This suggests that
behaviors engaged in -- and to some extent manipulated in virtual environments --

109

are capable of influencing cognitive representations of the self in the real world.
These intriguing possibilities should be explored in future research.

Situational awareness and Theory of Mind. Making sense of a situation,
critical to acting appropriately, may also make the self more salient. In the act of
sense-making we consider not only ourselves and our affective states but we also
consider other people and what their “theory of mind “may be. So for example,
with regard to the mirror study earlier, one interpretation for the finding that it
induces self-awareness, could be that it is simply showing us another person
(albeit a reflection) to which we are automatically primed to consider that other’s
”theory of mind”
5
. Everyday, as social actors, we seemingly automatically
adjust how we interact with other people, on the basis of what we think they are
thinking, feeling, and what their goals might be (e.g., their Theory of Mind)
(Gallese, 1998, Read & Miller, 1998).
Decision-making in interactive environments it seems, may not only
induce self-awareness, but require it. Interactive virtual environments require the
subject to constantly, although perhaps not consciously, make relational
observations bridging their real-life experiences and a host of factors, namely the
realism (that which is within the realm of probability), salience (how relevant is

110

5
"Theory of Mind" has more commonly been used to refer to a specific cognitive capacity: the
ability to attribute mental states—beliefs, intents, desires, pretending, knowledge, etc.—to
oneself and others and to understand that others have beliefs, desires and intentions that are
different from one's own (Gallese & Goldman, 1998).

this to my daily experiences), arousal (what am I feeling?), and degree of virtual
presence that is being re-created virtually (sense of being there, 1
st
person, 3
rd

person) (Lee, 2004) in order to make sense and act accordingly in the
environment.
These relational observations are consistent with current theories that have
focused on human behavioral choices as a form of psychology of action
(Gollwitzer, 1996). It is believed that people are capable of reflecting back on
their own past and present psychological state in order to make choices to regulate
such states into the future. This psychology of action relies on the concept of the
self –what motivates the self, how does the self act in a given situation, and how
does one define him/herself through the act.
From this vantage point, the glue it seems, that helps us to maintain
behavioral consistency between the virtual and the real is our concept of the ‘self
‘which relies on embodied cognition to continually drive our psychology of
action. The main idea underlying embodied cognition is that cognitive
representations and operations are fundamentally grounded in their physical
context (Gallagher, 2005; Hayles, 1998). Rather than relying solely on amodal
abstractions that exist independently of their physical instantiation, cognition
relies heavily on the brain’s modality-specific systems (e.g., mirror neurons) and
on actual bodily states. One intuitive example of how this operates can be seen in
the case of empathy. We understand another person’s emotional state by mentally
“re-creating” the person’s feelings in ourselves (Jackson, Meltzoff, & Decety,
2005).

111

Our continuing experiences with our own sense of self (e.g., embodied
cognitions) not only helps us to interpret our own actions (given the affective
states we experience, see for example, Anderson, 2005) but it may also help us to
understand and ultimately guide our behavior, in regard to another’s actions (e.g.,
interpreting another person’s Theory of Mind, given their affective state).
Because the SOLVE-IAV’s were designed to incorporate realistic situations that
allow the user to experience the emotional/arousal states that are likely to occur
when risky decisions might happen. The IAV’s thus may have been able to induce
or tap into decision-making under a state of embodied cognition (i.e., the gut
feeling you would have in real-life to the given situation).

How Can these Findings Lead To Better Interventions?
As an intervention approach, SOLVE may afford the opportunity to assess
and diagnose the circumstances under which risk-taking decisions are likely to
occur, and intervene at these critical junctures. SOLVE is premised on the belief
that realistic rehearsals and interventions that act within ‘hot’ affective states are
most likely to elicit behavior change that will then translate into similar behavior
in the real world. The current finding, that participants make interactive decisions
in the IAV that are consistent with their real-life decisions, advances the
development of better interventions because it demonstrates the ecological

112

validity of emotionally rich and interactive tailored environments in diagnosing
and predicting risk-taking decisions.
In the future, with multiple SOLVE based encounters, researchers may be
able to better diagnose what specific aspects of the situation are likely to cause
risk-taking, such that intervention messages can be even more finely tailored to
intervene more effectively at each of these junctures. Communication research
suggest a myriad of ways in which virtual environments could be modified to
further enhance SOLVE based intervention. For example, researchers have found
that increasing identification with characters over time can strengthen effects from
exposure (Donohew, 2000; Palmgreen et al., 1991; Zimmerman, 1997).
6

Because it is a realistic tailored environment, SOLVE may be better suited
for some individuals compared to other individuals. For example, high sensations
seekers have been shown to be particularly high-risk in terms of UAI (Donohew
et al., 1998). Interactive environments may be particularly well suited for and
diagnostic of these high sensation-seekers’ future behavior because of the
increased likelihood of attending to the arousing aspects of the intervention. As a
group, they may be more impulsive, but it may be difficult to capture or identify
these more impulsive risky choices in traditional one-on-one or linear media
based interventions: An interactive environment that socially situates a sexual

113

6
It has been suggested that the need for novelty and sensation may be biologically rooted
because it may have provided a vital means for survival and adaptability in the past and its
present day effects produce a reinforcement value through dopamine pathways –possibly acting
in a similar manner as a drug reward might. (Vaccarino, Schiff & Glickman, 1989; Zuckerman,
1994)

scenario is more likely to capture this impulsive tendency to choose risky
behavior.
7

Real-virtual choice consistency, which this study was able to document,
appears to be occurring because it is a well designed virtual environment
simulating ‘real life’ increasing the likelihood that the embodied cognitions of the
participants are being activated. This is a critical finding in the context of health
interventions because by tracking virtual risk-taking, interventionists will be able
to assess effectively in what contexts and with what ‘triggers’ MSM will chose to
engage in sexual risks. This may also someday serve as a diagnostic tool in lieu
of traditional self-report measures and as we will see in research questions two
and three, it may also capture additional variance in risk-taking under ‘hot’
emotional states, not easily assessed with paper and pencil measures.

Are virtual decisions better able to predict to future risky real-life behavior
beyond traditional cognitive measures (e.g., intentions, self-efficacy)?

Review of Findings
The research question posed above led to hypotheses H
3
through H
7
. H
3

asked whether traditional cognitive measures found in the theory of planned
behavior (e.g., intentions, self-efficacy) were capable of predicting past sexual
risk-taking behavior. Utilizing study sets one and two (combining both pre-test
measures of a yoked and IAV condition), it was found that the TPB was a

114

7
Many researchers have argued that both sensation seeking and impulsivity are complementary
components of decision-making and thus the combination of the two may have particularly
deleterious interactive effects by increasing risk-taking (Donohew, 2000).

significant predictor, accounting for 4% of the variance in past risk-taking
behavior. H
4
asked whether these same predictor variables taken at post-
intervention (post-test) were capable of predicting future sexual risk-taking
behavior (3 months out). It was found that the TPB was a significant predictor,
accounting for 12.4% of the variance in future risk-taking behavior.
H
5
examined the best predictor of future sexual risk-taking behavior, past
behavior, in order to see how it compares to either cognitive or virtual risk
measures. It was found that past behavior was a significant predictor, accounting
for 20% of future sexual risk-taking behavior. H
6
looked at whether virtual sexual
risk-taking was capable of predicting past sexual risk-taking better than the TPB,
utilizing a yoked condition (MSM were randomly assigned to view a
linear/noninteractive version based on the logged choices made by another MSM
in the IAV condition ) as a control. It was found that virtual sexual risk-taking
was a significant predictor, accounting for 24% of the variance in past sexual risk-
taking behavior while the yoked control was non-significant.
H
7
looked at whether virtual sexual risk-taking was capable of predicting
future sexual risk-taking better than the TPB, also utilizing a yoked condition as a
control. It was found that virtual sexual risk-taking was a significant predictor,
accounting for 12.1% of the variance in future risk-taking behavior, a number
nearly equal to the TPB, with the yoked control again being non-significant.

115

TPB as a predictor of past and future behavior.
Since a number of meta-analysis have combined predictive and postdictive
measures (Armitage & Conner, 2001, Albarracin et al., 2005) in assessing the
utility of the TPB, both H
3
and H
4
were proposed in order to determine the
predictive validity of each measure in the health intervention context. Because H
3

consisted of measures taken at pre-intervention and H
4
consisted in identical
measures taken immediately post-intervention, we would assume theoretically
that there would be no difference between the two items in terms of their
predictive utility. However, it was surprising to find that in the pre-intervention
context much less utility was gained when using the TPB to assess past sexual
risk-taking behavior. This implies that when people ‘know’ that they are in a
longitudinal health intervention study, their answers are likely to be forward
looking, meaning they may tend to give more biased answers (e.g., a Hawthorne
effect) or their intentions might actually have shifted by mere virtue of the fact
that they agreed to be in a health research study.
In terms of predicting future risk-taking behavior, the TPB accounted for
12.1% of the variance in future risk-taking behavior. This is one of the first
studies to demonstrate that the TPB can be used to significantly account for
variance in future risk-taking behavior longitudinally, in the post-intervention
context among young MSM. This is also the first study to demonstrate how in the
health intervention context, the TPB differs significantly between pre and post test
measures in assessing risk-taking. This poses the question for future research,
how and in what ways can we minimize the discrepancy between forward looking

116

pre-test measures and subsequent post-test measures in terms of past and future
actual behavior. This strikes me as a serious issue, because it implies that the
efficacy of many interventions in terms of ‘change’ and ‘strength of change’ in
these behavioral measures could be potentially masked by this forward looking
phenomena

The best predictor of past behavior is virtual behavior
One surprising finding is that contrary to the TPB pre-test measures,
virtual sexual risk-taking actually did a superb job at predicting past sexual risk-
taking behavior-(with virtual sexual risk-taking accounting for 24.4% of the
variance). Why might this be the case? One explanation could be that virtual risk-
taking is tapping into the more automatic affect based processes (embodied
cognitions) involved in real life decision-making, thereby serving as an
unobtrusive measure of actual decision-making. We find support for this
argument based on the findings earlier in H
1
and H
2
, that there is a link between
the virtual and the real. The question is how might this virtual behavior be
tapping into these more automatic decision-making processes?
Does tapping into embodied cognitions guide behavior? If you will recall
in the discussion earlier, we theorized that interactivity may be acting to make the
self more salient, thereby inducing a state of embodied cognition, but how exactly
is it that embodied cognitions operate to automatically guide human behavior?
Cognition is embodied in the sense that how we use our bodies ‘reminds us’ and
‘prioritizes’ our cognitions. Damasio (1999, p.44) gives a telling example of an

117

experiment he conducted on a man who sustained damage to his brain such that
he had no short term memory. The man could no longer remember what happened
for more than a few hours. The man was exposed to two confederates over time.
One confederate would give him a tedious task and the other would give him a
pleasurable task. Although the man could not remember what happened from one
day to the next, he nevertheless developed an aversion to the confederate who
supplied the tedious tasks. This ‘gut feeling’ or emotional response, guides
human behavior at the most primitive levels of the brain. The primitive mind it
seems is capable of linking felt experience with behavior at an unconscious level.
Other research supports this proposition that memories of experiences are
encoded and activated to guide behavior. Pugnetti et al. (1998) performed a study
in which spatial recall was tested between those who were healthy and those that
had MS, when exposed to either a virtual environment or a yoked condition.
Although the MS group did worse than the healthy group in object and spatial
layout recall, performance did not decrease as cognitive requests increased in the
virtual environment –contrary to divided attention task tests in a standard lab
setting.
They reasoned that passive observation is simply a monitoring task –
giving no real incentive to process the spatial properties of a particular
environment (such as riding in a car as a passenger), whereas subjects within an
interactive environment must behave as if they are really inside. In doing so, they
utilize the special properties of the virtual world, ‘implicitly memorizing them as
behavioral representations that include visuospatial, temporal and sensor motor

118

components” (Pugnetti et al., 1998, p. 428). Thus MS patients didn’t have to
actually walk through the space to get the same visual processing effect. The
involvement of the self again seems to be an important element in activating these
embodied cognitions which significantly demonstrates the difference between
passive and active observation in the degree to which embodied cognitions are
activated.
Embodied knowledge, that is, the reasons why we act in certain situations
may not even be entirely formalized even in the subjects own mind but are the
result of a multitude of factors (past experiences, outcomes, emotions elicited, and
learned results). Habit is a type of embodied knowledge, that our bodies know and
understand (Hayles, 1999). IAV’s may be tapping into this embodied cognition
construct.
Taking risks under this state of embodied cognition may indicate a failure
on the part of the individual to self-regulate their risk-taking behavior. Recent
work by deWit & Adam (2006) has examined the construct of behavioral
willingness with regard to self-regulation. They define willingness as the non-
intentional processes in social reaction to opportunities that can sometimes lead to
risk-taking. They argue that behavioral willingness encompasses a multitude of
factors such as: sensation seeking, social influence, affect, condom fatigue,
optimistic beliefs about HIV therapies, etc… The virtual risk-taking measure is
very compatible with this willingness concept because it appears that embodied
cognition is capable of capturing many of these non-intentional social influences
and reactions to unfolding situations.

119

In a similar study using randomly generated computer vignettes with
participants rating the likelihood of their engaging in risks given each vignette,
researchers were able to predict the environmental variables that were most likely
to lead to risk-taking among young MSM (Ross et al., 2004). Virtual risk-taking
is akin to these situational vignettes but also has the added benefit of being even
more engaging and realistic, thus more accurately representing the salience of
embodied cognition in the decision-making process.

TPB and Virtual Risk: equal predictors of future behavior?
One of the interesting things about the finding that virtual risk-taking
predicts to actual future risk-taking, is that it accounts for about as much variance
as the TPB. Two critical questions need to be addressed however. One question
is whether the variance being accounted for by virtual risks and the TPB, is the
same type of variance, or is virtual decision-making tapping into something
different, (e.g more automatic processes relating to embodied cognition, that
account for unique variance beyond the TPB). The second question, given that the
best predictor of past behavior is virtual behavior, is what role does interactivity
play in linking behavior with experience. These critical questions are addressed in
grand research question 3.

120

Do virtual decisions (that may tap into more automatic affective processes)
account for additional unique variance in real-life risk-taking beyond traditional
cognitive measures? And if so, then might the combination of cognitive and affect-
based predictors fully mediate the link between past and future risky behavior?

Review of Findings
H
8
asked whether the variance accounted for in future sexual risk-taking
by virtual risk-taking, overlapped with the variance accounted for by the TPB. A
yoked control was used also to test whether observational learning or interactive
learning was providing the additional predictive utility over and above the TPB.
On the second step of the regression, with the TPB at the first step and virtual risk
taking on the second, it was found that virtual risk-taking was indeed accounting
for significant additional unique variance (8.8%) nearly equal to that of the TPB
(a total of 16.8%). This was not the case with the yoked control, where virtual
risk provided no additional significant unique variance.
Interactivity Matters in linking the behavior to the experience. The
finding that virtual risk-taking provided additional significant unique variance in
the IAV but not at all in the yoked condition brings up an interesting point. Why
doesn’t observational learning alone (found in the yoked condition) also influence
future behavior to a significant degree beyond cognitive predictors. The reason
this is an important question, is that a long body of literature has demonstrated
how observational learning acts as a means to influence future behavior. What
this finding may imply, is that observational learning only acts upon those explicit
cognitive measures, like those found in the TPB, but fails to change behavior

121

based on a more visceral level of decision-making, such as with regard to
affective experiences or embodied cognitions.
Recent work on brain imaging may provide new insight into the processes
by which this distinction between observational learning and interactive learning
effect explicit cognitive and non-cognitive encoding. Anderson et al (2006) found
that when people watched boxing matches, there was activation of the pre-motor
cortex, indicating that viewers were actually attempting to imitate or place
themselves into the action. Similarly, Murray (2006) found that young children
processed violent videos in the area of the brain traditionally associated with
significant traumatic events (despite the fact that the children knew that the movie
violence was not ‘real’). The neurologic evidence suggests that children who
watch a lot of violence may be more likely to be aggressive when triggered
because they have these prior traumatic response scripts stored in their long-term
memory. This certainly makes the case for observational learning to provide
additional unique variance on a non-cognitive level, but yet we do not see that in
the present study.
One determinant that may be the key to explaining why this is not the case
in the present study, has to do with what was discussed earlier in research
question 1, regarding the Theory of Mind. There has been much discussion
regarding this concept, especially the neurophysiologic aspects of how we reflect
on our own behavior, utilizing something known as a mirror neuron. Mirror
neurons are tiny structures in the brain that fire either when we perceive an action
or when we take an action ourselves. For example, mirror neurons fire in

122

experiments with monkeys when they break peanuts in their hands, watch other
monkeys break the peanuts, or even just hear the sound of a peanut breaking. “It’s
almost like the monkey is watching his own action reflected by the mirror”
(Iacoboni, 2007).
Part of the significance of mirror neurons is the understanding that they
may actually play a role in codifying or understanding our own and others’
intentions. In one experiment to test whether mirror neurons are used to simply
recognize actions as opposed to understanding the underlying meaning of the
action (i.e., contextual pros and cons), subjects were shown either context clips
(free of action, a table with a tea setting) or intention clips (a hand getting ready to
grab a cup of tea). Though there was some mirror neuron activity during the
context videos, there was significantly more firing during the drinking intention
clips, suggesting that these neurons might actually codify intentional actions. In
another study, mirror neurons were found to fire at an even more pronounced rate
with an intention to drink from a cup, if the subject was particularly thirsty.
Decety, et al. (1997) conducted a study which examined brain activation
using PET scans under conditions in which individuals were presented with a
media stimuli and instructed to watch the actions

with one of two aims: to be able
to recognize or to imitate them later. Differences in the meaning of the action
(i.e., highly meaningful or self-relevant actions), irrespective of the

strategy used
during observation, lead to different patterns of brain

activity.

Observing with the
‘intent to recognize’ activated memory-encoding

structures. In contrast,
observation with the ‘intent to imitate’ was

associated with activation in the

123

regions involved in the planning and in

the generation of actions. This finding is
consistent with the role ‘mirror neurons’ play in codifying intentions and appears
to be why interactive virtual behavior might be more diagnostic of future behavior
than passive media exposure, and also why embodied cognitions (i.e., spatial
temporal, emotional, arousal) play such a critical role in influencing behavior --
yet are the least recognized by ourselves in terms of conscious deliberative action.
Researchers have found that if you're skilled or habituated to performing
a specific physical activity the part of your brain that controls movement activates
differently than the same part in the brain of someone who's not skilled in that
activity (Calvo-Merino, B., Grezes, J., Glaser, D.E., Passingham, R.E., &
Haggard, P., 2006). This suggests that once the brain has learned a skill, it may
simulate the skill without bodily movement, through simple observation of the
same action. Thus we see evidence for the distinction between planning or
considering an enactment of one’s own behavior and merely recognizing a
behavior.
In the intervention context, showing a behavioral narrative that is not akin
to pre-existing intentions may be memorable but may not influence subsequent
behavior. If the narrative matches the pre-existing intentions, however, it may
actually re-enforce the intention if there is no attempt to intervene in the
experience at the time of the mirror neuron firing. Thus it is crucial to understand
the neurophysiological aspects of learning that are occurring while individuals
participate in interactive simulations in order to determine the motivational,

124

affective, cognitive and social states (MACS) that are attendant with degrees of
learning and behavior change.

Towards a Meditational Model of Risk-Taking
In order to test whether it is sufficient to just be exposed to the choices
oneself or whether it is the interaction of the choice with the message that
influences embodied cognitions apart from explicit cognitive measures, past
behavior was regressed onto steps one and two to determine whether the
combination of virtual risks in either the IAV or Yoked mediates the relationship
between past and future behavior. It was found that only active virtual risk-taking
in combination with the TPB fully mediated the relationship between past and
future behavior, accounting for nearly 16.8% of the variance. The mediational
model was not significant for the yoked condition. No other mediational model
has been able to fully mediate the relationship between past and future sexual
risk-taking among MSM.
This confirms that embodied cognitions do play a unique and significant
role in risk-taking behavior and that only virtual environments that allow active
decision-making will be able to effectively alter or intervene when these
embodied cognitions are activated. This mediational model also demonstrates that
affect does play a critical role in risk-taking, AND that the TPB is not as effective
in accounting for change in these more automatic affective responses that may be
induced when an interactive intervention is utilized to modify real world future
risk-taking behavior.

125

As mentioned in the introduction, people learn from interactivity (e.g.,
play) by behaving in a way that rings true---i.e., the low hanging fruit. We learn
incrementally, oftentimes by observing (because we intend to do it in the future or
wanted to do it but haven’t had the opportunity) but mostly by doing (because we
have been doing it—maybe not that well or too well with the risk of deleterious
consequences). This is why interactivity in virtual environments matter, because
people learn from interactivity better (linking the experience with the behavior)
giving interventionists the opportunity to challenge these more automatic
behaviors and subsequently assess the effectiveness of the intervention. These
virtual environments also hold great promise in changing behavior among those
who have learning difficulties (perhaps caused by drug use) and those with
insufficient experience to begin with. Below is a graphic presentation of this
proposed meditational model:

126

FIGURE 2:

A mediational model of both the rational and
non-rational aspects of sexual risk-taking

This SOLVE based mediational model maps onto the general model of
decision-making based on somatic influence, developed by Bechara and Damasio
(1997). This model is much more robust because it includes both the TPB and virtual
risk-taking. This model accounts for those non-cognitive elements that may also be
playing a role in terms of influencing behavior. These relate to affective or
experiential biases that may subconsciously steer individuals away from or towards,
risk-taking situations. Virtual risk-taking situations are capable of including elements
such as: arousal interfering with rational decision-making and perceived self-efficacy
(e.g., inaccurate affective forecasting), a willingness to engage in risk (e.g., giving in
to partner demands or motives), and individual personality differences (e.g., high
sensation seekers, high attachment avoidance).

127

In this capacity, SOLVE based interventions are able to provide virtual
experiences that are capable of ‘substituting’ for real experiences –emotionally
punctuating the space, and thereby marking the experience for future recall. One
goal for future research, would be to create advanced computational models that are
capable of testing which aspects of the SOLVE based intervention work best and for
whom based on multiple test scenarios. In this way, you could methodologically go
about testing each part of the model (e.g. procedural skills, reasoning strategies,
affect, decision-making processes).

128

CONCLUSION
This research suggests that individuals immersed in a realistic real life
environment will choose to engage in behaviors consistent with their real life
behaviors. This is an important development in the field of health communication
because it will allow us to diagnose risk-takers and create interventions that will
be most effective for them in ‘real time’. If health interventionists knew that
individuals were most likely to use IAV in terms of their past sexual behaviors,
then IAV could be used as a possible means to discover the degree of risk that
participants engage in without having to administer a set of survey questions that
may or may not be answered truthfully or reliably. This is particularly salient for
MSM who are young, shy, inexperienced or struggling with questions regarding
their sexual identity.
IAV, as a simple diagnostic tool for high risk-takers, could serve a
potentially useful role in traditionally hard to reach populations. Kiosks or online
sites could be set up that would invite participants to view a fun virtual hook up
IAV. Those who score above a certain pre-defined risk assessment scale based on
the virtual choices would be invited to participate in a health intervention at a
local CBO or if viewed online, referred to a website specifically designed to help
foster healthy behavior change (i.e. linking those MSM with mentors or peers and
providing self-efficacy enhancement exercises). Virtual environments could be
used as a confirmatory test of the effectiveness of a more comprehensive health

129

intervention or as a confirmation of a new novel intervention itself (such as virtual
environment with responsive agents).
Normally we don’t know if a particular health intervention is effective
until after a longitudinal real-life study is completed, and as past research has
suggested, change in TPB variables immediately following an intervention, do not
account for much variance longitudinally. With virtual technology, we can have
test phases, try an intervention in a virtual environment, and then test again, in
order to see if and how our intervention has adjusted the participant’s behavior,
and as this study has suggested, reliably predict future behavior in a more accurate
way.
With technologies that make it possible to experience virtual realities that
map onto real life experiences, the door is open to considering not only how one
maps onto the other and its utility for diagnostic purposes, but also for learning
how behavior in the virtual environment might alter patterns of behavior in the
real environment. Perhaps fully immersive virtual environments with autonomous
agents could be developed that would allow for a panoply of risk taking scenarios
to be experienced in the virtual environment. This is important from a
developmental perspective because we know that younger MSM 18-30, are much
more likely to engage in risky sex (MacKellar, et al., 2005; Xia, Osmond,
Tholandi, Pollack, Zhou, Ruiz, and Catania, 2006). About 50% of all new HIV
infections in the United States are among those under age 25 (CDC, 2006).
Traditional Behavioral Decision-Making models (TBDM) that underlie
most HIV prevention efforts focus on individuals’ rational deliberation (e.g. the

130

costs and benefits) in making decisions to optimize the decision-maker’s goals:
Although effective, they account for less than 5% of the variance in risk-taking
Reyna & Farley, 2006). This may be because young adults, including MSM,
often make emotional and cognitive decisions that are not conscious or
deliberative ones. For example, young MSM may intend to use condoms, but are
willing to make decisions reactively in response to contextual triggers (e.g. an
attractive partner in the heat of the moment) that they later regret (Gibbons, et al.,
2004). Or, they may make decisions intuitively, responding to the gist cues
(Reyna & Farley, 2006) that –via experience—they have learned to associate with
risk. But, if UAI is not associated with contracting HIV subsequently, “risk
avoidance cues” and accompanying negative affect, may not be learned (Reyna &
Farley, 2006)—until it’s too late.
Existing HIV prevention interventions acknowledge obstacles to safer sex
such as “being in the heat of the moment” and research demonstrates the potential
role of sexual arousal in impacting decision-making (Ariely & Loewenstein,
2005). Nevertheless we typically haven’t simulated such emotion-evoking
contexts to challenge and retrain more automatic affect based responses and to
associatively link risky choices with negative affect. By doing so, young MSM
could derive the benefits of (virtual) experience without the potential real costs
(e.g. HIV).
Limitations and Avenues for Future Inquiry. The issue for future research
is under what circumstances can we expect to cue similarities between real past
behavior and behavior within virtual environments, and where can we find the

131

opportunity for altering behavior within these environments. How fast and in
what ways can we alter these behavior patterns in virtual environments to
influence decision making in the real world? How can we both assess behavior
and also effectively change behavior, and can we tailor virtual environments to
allow us to do both?
Unfortunately IAV technology restricts the number of test-situations (e.g.
venues, interaction partners, type of risk, risk scenarios, etc…). This presents a
series of threats for further optimizing personalized risk-reduction for MSM. In
this current study, IAV technology limited the capability to proceed with more
fine-grained analysis. For example, as Miller et al. (2007) noted it was not
possible to “(1) insure that there was at least one risky test-situation that was
relevant to all MSM, (2) assess whether different MSM exhibited different risk-
taking profiles across various situations, (3) assess whether a given MSM’s
responses to specific types of test-situations predicted similar real-life behavior
and (4) assess whether a given MSM learned to self-regulate his initially risky
choice in a given virtual situation and if so, if that predicted similar risk-reduction
into the future under similar situations.”
The ultimate goal that this line of research speaks to is the possibility of
creating a diagnostic ‘test-bed’ using more complex immersive virtual
environments. By creating more immersive virtual environments that will extend
the range of behavioral choices and scenarios through the use of
gaming/autonomous intelligent agent technology, it will become possible to
advance a cumulative science of optimizing personalized risk-reduction - using a

132

technology-enabled “test-bed”(Miller et al., 2007). In the future, with virtual
technology as our ‘test bed’, we can have pilot test phases –i.e. try an intervention
in a virtual environment, and then test again, in order to see if and how our
intervention has adjusted the participant’s behavior.
This will allow us to discover new opportunities for altering behavior
within virtual environments that will translate into real life behavior changes,
potentially exposing people to fewer of life’s ‘experiential’ serious health risks.
The notion that "you are who you pretend to be" has a mythic resonance rooted in
the Pygmalion story. Virtual environments are powerful tools not only because
they speak to a fundamental human need to believe that we are not limited by our
histories --that we can recreate ourselves-- but also because this self-
transformation can have constructive functional significance in the real world.

133

BIBLIOGRAPHY
Ajzen, I. (2002). Residual effects of past on later behavior: Habituation and reasoned
action perspectives. Personality and Social Psychology Review, 6, 107-122.

Albarracin, D., Gillette, J.C., Earl, A.N., Glasman, L.R., Durantini, M.R., & Ho, M.
(2005). A test of major assumptions about behavior change: A comprehensive
look at the effects of passive and active HIV-prevention interventions since the
beginning of the epidemic. Psychological Bulletin, Vol. 131, No 6., 856-897.

Albarracin, D., Johnson, B.T., Fishbein, M., & Meullerleile, P.A. (2001). Theories of
reasoned action and planned behavior as models of condom use: A meta-analysis.
Psychological Bulletin, 127, 142-161.

Anderson, M.L. (2005). Embodied cognition: a field guide. Artificial Intelligence. 149,
91–130.

Anderson, D. R., Bryant, J., Murray, J. P., Rich, M., Rivkin, M. J., Zillman, D. (2006).
Brain Imaging –An introduction to a new approach to studying media processes
and effects. Media Psychology, 8, 1-6.

Andsager, J.L., Austin, E.W., & Pinkleton, B.E. (2001). Questioning the value of realism:
Young adults’ processing of messages in alcohol-related public service
announcements and advertising. Journal of Communication, 51(1), 121-142.

Angela, B., Fisher, W.A., & Fisher, J.D. (2001). Tests of the mediational role of
preparatory safer sexual behavior in the context of the theory of planned behavior.
Health Psychology, 21(1), 71-80.

Appleby, P.R., Ayala, A., Storholm, E.D., Christensen, J.L., & Miller, L.C. (under
review). Methamphetamine use and high risk sex among men who have sex with
men (MSM): The paradox of disconnecting to connect.

Appleby, P.R., Godoy, C.G.,, Miller, L.C., & Read, S. J. (2007). Increasing healthy
behavior through the use of interactive video technology. In T. Edgar, S. M. Noar,
V.S. Freimuth (Eds.). Communication perspectives for HIV/AIDS in the 21st
century. Mahwah, NJ: Lawrence Erlbaum Associates.

Appleby, P.R., Miller, L.C., & Ayala, A. (2003). Why methamphetamine contributes to
risky sex for MSM. (Tech. Rep.). Los Angeles, California: University of Southern
California, Annenberg School for Communication.

134

Armitage, C.J. & Conner, M. (2001). Efficacy of the theory of planned behavior: A meta-
analytic review. British Journal of Social Psychology, 40, 471-499.

Armitage, C. J. (2005). Can the theory of planned behavior predict to maintenance of
physical activity? Health Psychology, 24, 235-245.

Bailenson, J., Beall, A.C., Loomis, J., Blascovich, J., & Turk, M.C. (2004). Transformed
social interaction: Decoupling representation from behavior and form in
collaborative virtual environments. Presence: Teleoperators and Virtual
Environments, 13, 428-441.

Bailenson, J.N., Blascovich, J., Beall, A.C., Loomis, J.M. (2003). Interpersonal distance
in immersive virtual environments. Personality and Social Psychology Bulletin,
29, 819-834.

Bandura, Albert, Ross, Dorothea, & Ross, Sheila A. (1961). Transmission of aggressions
through imitation of aggressive models. Journal of Abnormal and Social
Psychology, 63, 575-582.

Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory.
Englewood Cliffs, NJ: Prentice-Hall.

Bandura, A. (1997). Self-efficacy: The exercise of control. New York: Freeman.

Bandura, A. (2001). Social cognitive theory of mass communication. Media Psychology,
3, 265-299.

Bargh, J. (1997). The automaticity of everyday life. In R.S. Wyer (Ed.), The automaticity
of everyday life: Advances in social cognition (pp. 1-61). Mahwah, NJ: Lawrence
Earlbaum Associates.

Barkow, J.H., Cosmides, L., & Tooby, J. (Eds.).(1992). The adapted mind. Evolutionary
psychology and the generation of culture. New York: Oxford University Press.

Bechara, A., Damasio, H., Tranel, D., & Damasio, A.R. (1997). Deciding advantageously
before knowing the advantageous strategy. Science, 275, 1293-1295.

Bekoff, M. & Byers, J.A. (1998). Animal Play: Evolutionary, Comparative, &
Evolutionary Perspectives. Cambridge University Press.

Biocca, F., & Delaney, B. (1995). Immersive virtual reality technology. In F. Biocca &
M.R. Levy (Eds.), Communication in the age of virtual reality (pp. 57-124).
Hillsdale, NJ: Lawrence Earlbaum Associates.

135

Biocca, F (1997). The cyborg’s dilemma: Progressive embodiment in virtual
environments. Journal of Computer Mediated Communication, 3(2).

Biocca, F. (2002). The evolution of interactive media: Toward “being there” in nonlinear
narrative worlds. In M.C. Green, J.J. Strage, & T.C. Brock (Eds.), Narrative
impact: Social and cognitive foundations (pp.97-130). Mahwah, NJ: Lawrence
Earlbaum Associates.

Blascovich, J. (2002). Social influence within immersive virtual environments. In: R.
Schroeder (ed.), The social life of avatars. London: Springer-Verlag, pp.127–145.

Blascovich, J., Loomis, J., Beall, A., Swinth, K., Hoyt, C., & Bailenson, J. (2002).
Immersive virtual environment technology as a research tool for social
psychology. Psychological Inquiry, 13, 103-125.

Bourdieu, P. (1977). Outline of a Theory of Practice, translated by Richard Nice.
Cambridge, England: Cambridge University Press.

Brickell, T.A., Chatzisarantis, N. L. & Petty, G. (2006). Using past behavior and
spontaneous implementation intentions to enhance the utility of the theory of
planned behavior in predicting exercise. British Journal of Health Psychology,
Vol 11(2), 249-262.

Play. (2007). In Encyclopedia Britannica. Retrieved September 4, 2007, from
Encyclopedia Online: http://www.britannica.com/eb/article-9060375.

Brown, S. J., Lieberman, D.A., Gemeny, B.A., Fan, Y.C., Wilson, D.M., & Pasta, D.J.
(1997). Educational video game for juvenile diabetes. Results of a controlled trial.
Medical Information, 22(1), 77-89.

Butler, L., Miezitis, S., Friedman, R., & Cole, E. (1980). The effect of two school-based
intervention programs on depressive symptoms in preadolescents. American
Educational Research Journal, 17(1), 111-119.

Calvo-Merino, B., Grezes, J., Glaser, D.E., Passingham, R.E., & Haggard, P. (2006).
Seeing or doing? Influence of visual and motor familiarity in action observation.
Current Biology. 10; 16(19):1905-10.

Carlin, A.S., Hoffman, H., & Weghorst, S. (1997). Virtual reality and tactile
augmentation in the treatment of spider phobia: A case study. Behavior Research
and Therapy, 35(2), 153-158.

136

Chaytor, N. & Schmitter-Edgecombe, M. (2004). The Ecological Validity of
Neuropsychological Tests: A Review of the Literature on Everyday Cognitive
Skills. Neuropsychology Review, Vol. 13, No. 4.

Cialdini, R. B. (2001). Influence: Science and practice (4th ed.). Boston: Allyn & Bacon.

Costa, P.T., Jr., & McCrae, R.R. (1994). Set like plaster? Evidence for the stability of
adult personality. In T.F. Heatherton & J.L. Weinberger (Eds), Can Personality
Change? (pp. 21-40). Washington, DC: American Psychological Association.

Cohen, D., Dent, C., & MacKinnon, D. (1991). Condom skills education and sexually
transmitted disease reinfection. Journal of Sex Research, 28 (1), 139-144.

Coleman, C.L. (1993). The influence of mass media and interpersonal communication on
societal and personal risk judgments. Communication Research, 20, 611-628.

Conner, M., Warren, R., Close, S., & Sparks, P. (1999). Alcohol consumption and the
theory of planned behavior: An examination of the cognitive mediation of past
behaviorid. Journal of Applied Social Psychology, 29(8), 1676-1704.

Damasio, A. (1999). The feeling of what happens: Body and emotion in the making of
consciousness. Harvest Book. Harcourt, Inc. San Diego.

Decety, J., Grezes, J., Costes, N., Perani, D., Jeannerod, M., Procyk, E., et al. (1997).
Brain activity during observation of actions. influence of action content and
subject's strategy. Brain, 120(10), 1763-1777.

DeFleur, M. L. and S. J. Ball-Rokeach (1989) Theories of Mass Communication. New
York: Longman.

Denes-Raj, V., & Epstein, S. (1994). Conflict between intuitive and rational processing:
When people behave against their better judgment. Journal of Personality and
Social Psychology, 66, 819-829.

Derrida, J. (1967). Structure, Sign, and Play in the Discourse of the Human Sciences. In
The Structuralist Controversy: the Language of Criticism and the Sciences of
Man Richard Macksey and Eugino Donato (eds.), Baltimore, 1970.

Des Jarlais, D.C., Casriel, C., Friedman, S.R., & Rosenblum, A. (1992). AIDS and the
transition to illicit drug injection: Results of a randomized trial prevention
program. British Journal of Addiction, 87 (3), 493-498.

De Wit, J.B.F. & Adam, P.C.G. (2006). Sexual risk-taking as a result of social reactivity:
Assessment of non-intentional components of unprotected intercourse in Dutch
MSM. (Progress Report)

137

De Vroome EMM, Stroebe W, Sandfort, T.G.M, De Wit, J.B.F. & van Griensven G.J.P.
(2000) Safer sex in social context: Individualistic and relational determinations of
AIDS Preventive behavior among gay men. Journal of Applied Social
Psychology; 30 2322-40.

DiClemente, R.J., & Wingood, G.M. (1995). A randomized controlled trial of an HIV
sexual risk-reduction intervention for young African-American women. Journal
of the American Medical Association, 274 (16), 1271-1276.

Dijksterhuis, A., Bs, M.W., Nordgren, L.F., & van Baaren, R.B. (2006). On making the
right choice: The deliberation-without-attention effect. Science, 311, 1005-1007.

Donohew, L., Zimmerman, R., Cupp, P.S., Novak, S. Colon, S. & Abell, R. (2000).
Sensation seeking, impulsive decision-making, and risky sex: implications for
risk-taking and design of interventions. Personality and Individual Differences
28, 1079-1091.

Epstein, S. (1993). Emotion and Self-Theory, in Handbook of Emotions, ed. Lewis, M. &
Haviland, J.M., New York: Guildford, 313-326.

Everson, M.D. & Boat, B.W. (1990). Sexualized doll play among young children:
implications for the use of anatomical dolls in sexual abuse evaluations. Journal
of American Academy of Child and Adolescent Psychiatry, 29(5):736-42.

Epstein, S. (1994). Integration of the cognitive and the psychodynamic unconscious.
American Psychologist, 49, 705-724.

Fein, G. (1981). Pretend play in childhood: An integrative review. Child Development.
Vol. 52, No. 4, 1095-1118.

Fogg, B.J. (2003). Persuasive Technology: Using computers to change what we think and
do. Morgan Kaufman Publishers, San Francisco, CA.

Filiciak, M. (2003). Hyperidentities. In The Video Game Theory Reader, edited by
M. J. P. Wolf and B. Perron. London: Routledge.

Fisher, W.R. (1987). Human communication as narration: Toward a philosophy of
reason, value, and action. Columbia: University of South Carolina Press.

Frijda, N.H. (1988). The laws of emotion, American Psychologist, 43(5), 349-358.

Frosch, D., Shoptaw, S., Huber, A., Rawson, R.A., & Ling, W. (1996). Sexual HIV risk
among gay and bisexual methamphetamine abusers. Journal of Substance Abuse
Treatment, 13(6), 483-486.

138

Gallagher, S. (2005). How the body shapes the mind. Oxford University Press, Oxford,
U.K.

Gallese, V., & Goldman, A. (1998). Mirror neurons and the simulation theory of mind
reading. Trends in Cognitive Science, Vol. 2, No. 12.493-501.

Gerbner, G., Gross, L., Morgan, M. & Signorielli, N. (1986). Living with Television: The
Dynamics of the Cultivation Process, in J. Bryant & D.Zillman (Eds).
Perspectives on Media Effects, Lawrence Erlbaum Associates, Hillsdale, New
Jersey.

Gerrard, M., Gibbons, F.X., Benthin, A.C., & Hessling, R.M. (1996). A longitudinal
study of the reciprocal nature of risk behaviors and cognitions in adolescents:
What you do shapes what you think and vice versa. Health Pyschology, 15, 344-
354.

Gilbert, D.T. (1991). How mental Systems believe. American Psychologist, 46(2): 107-
109.

Glassner, B. (1999). The culture of fear: Why Americans are afraid of the wrong things.
New York: Basic Books.

Godoy, C., Appleby, P.R., Miller, L.C., & Read, S.J. (2005) Sexual Risk Taking Scripts
Among A Multi-Ethnic Sample of Young MSM. Poster session presented at the
Institute for the Society of Sexually Transmitted Diseases (ISSTDR), Amsterdam,
Netherlands.

Godin, G., Valois, P., & Lepage, L. (1993). The pattern of influence of perceived
behavioral control upon exercising behavior: An application of Ajzen’s
Theory of Planned Behavior. Journal of Behavioral Medicine, 16, 81-102.

Gollwitzer, P.M., & Bargh, J.A. (Eds.) (1996). The psychology of action: Linking
cognition and motivation to behavior. New York: Guilford.

Griebel, T. (2006). Self-portrayal in a simulated life: Projecting personality and values in
the Sims 2. International Journal of Computer Games Research, Vol. 6, No. 1, 1-
17.

Groos, K. (2005). The play of animals: Studies in play and games. (E.L. Baldwin,
Trans.). Kessinger Publishing, LLC.

Halkitis, P. N., & Parsons, J. T. (2002). Recreational drug use and HIV-risk
sexual behavior among men frequenting gay social venues. Journal of Gay and
Lesbian Social Services, 14, 19-39.

139

Harmon-Jones, E., Brehm, J.W., Greenberg, J., Simon, L., & Nelson, D.E. (1996).
Evidence that the production of aversive consequences is not necessary to create
cognitive dissonance. Journal of Personality and Social Psychology, Vol. 70, No.
1, 5-16.

Hayles, K. N. (1999). How we became posthuman: Virtual Bodies in Cybernetics,
Literature and Informatics. Chicago: University of Chicago Press.

Idhe, D. (1991). Instrumental realism: The interface between philosophy of science and
philosophy of technology. Bloomington: Indiana University Press.

Jackson, P.L., Meltzoff, A.N., & Decety, J. (2005). How do we perceive the pain of
others? A window into the neural processes involved in empathy. NeuroImage,
Vol 24, Issue 3, 771-779.

Jemmott, J.B., Jemmott, L.S., & Fong, G.T. (1992). Reductions in HIV risk-associated
sexual behaviors among black male adolescents: Effects of an AIDS prevention
intervention. American Journal of Public Health, 82 (3), 372-377.

Joseph, M.K., Adib, S.M., Joseph, J.G., & Tal M. (1991). Gay identity and risky sexual
behavior related to the AIDS threat. Journal of Community Health, Vol.16, No. 6.

Kant, I. (1987/1929). Immanuel Kant's critique of pure reason. (N.K. Smith, Trans.).
New York: St Martin's Press.

Kelly, J.A. (1994). Sexually transmitted disease prevention approaches that work:
Interventions to reduce risk behavior among individuals, groups, and
communities. Sexually Transmitted Diseases, 21 (2, Suppl.), S73-S75.

Kelly, J.A., Sikkema, K.J., Winett, R.A., Solomon, L.J., Roffman, R.A.,
Heckman, T. G., Stevenson, L.Y., Perry, M.J., Norman, A.D., & Desiderato, L.L.
(1995). Factors predicting continued high HIV risk behavior among gay men in
small cities: Psychological, behavioral, and demographic characteristics related to
unsafe sex. Journal of Consulting and Clinical Psychology, 63, 101-107.

Kim, Y., & Baylor, A.L. (2006). A Social-Cognitive Framework for Pedagogical Agents
as Learning Companions. Educational Technology and Research Development.
54, No. 6: 569-596.

Kiousis, S. (2002) ‘Interactivity: A Concept Explication’ , New Media & Society 4(3):
355-383.

140

Klinger, E., Legeron, P., Roy, S., Chemin, I., Lauer, F., & Nugues, P. (2006) Virtual
reality exposure in the treatment of social phobias. In Cybertherapy: Internet and
virtual reality as assessment and rehabilitation tools for clinical psychology and
neuroscience. G. Riva, C. Botella, Legaran, P. & Optale, G. (Eds) Amsterdam,
IOS Press.

Lang, A. & Friestad, M. (1993). Emotion, hemispheric specialization, and visual and
verbal memory for television messages. Communication Research, 20(5), 647-
670.

Lang, P.J. (1995). The emotion probe: Studies of motivation and attention, American
Psychologist, 50(5), 372-385.

Lauria, R. (1996). Virtual reality: An empirical-metaphysical testbed. Journal of
Computer Mediated-Communication [On-line] 2,(1). Available:
http://www.ascusc.org/jcmc/vol3/issue2/lauria.html

Lee, K.M. (2004). Why presence occurs: Evolutionary psychology, media equation, and
presence. Presence: Teleopters and Virtual Environments, 13, 494-505.

Lenggenhager, B., Tadi, T., Metzinger, T., & Blanke, O. (2007). Video ergo sum:
Manipulating bodily self-consciousness. Science, Vol. 317, 1096-1099.

Leuvan, J. van (1981). Expectancy theory in media and message selection.
Communication Research, 8, 425-434.

Lichtenstein, S., Slovic, P., Fischhoff, B., Layman, M., & Combs, B. (1978). Judged
frequency of lethal events. Journal of experimental psychology: Human learning
and memory, 4, 551-578.

Lieberman, D.A. (2001). Management of Chronic Pediatric Diseases with Interactive
Health Games: Theory and Research Findings. Journal of Ambulatory Care
Management, 24(1), 26-38.

Lieberman, D.A. (2001). Using Interactive Media in Communication Campaigns for
Children and Adolescents. Chapter in R. Rice & C. Atkin (Eds.), Public
Communication Campaigns, 3rd Edition. Newbury Park, CA: Sage Publications,
pp. 373-388.

Lieberman, D.A. (1999). The Researcher's Role in the Design of Children's Media and
Technology. Chapter in A. Druin (Ed.), The Design of Children's Technology.
San Francisco: Morgan Kaufmann Publishers, pp. 73-97.

Locke, J. (1700/1979). An Essay Concerning Human Understanding, Edited with forward
by P.H. Nidditch. Oxford: Oxford University Press.

141

Lowenstein, G.F., & Schkade, D. (1999). Wouldn’t it be nice? Predicting future feelings.
In E. Diener,, N. Schwartz, & D. Kahneman (Eds.), Well-Being: The foundations
of hedonic psychology (pp. 85-105). New York: Russell Sage Foundation.

Magura, S., Kang, S., & Shapiro, J.L. (1994). Outcomes of intensive AIDS education for
male adolescent drug users in jail. Journal of Adolescent Health, 15 (6), 457-463.

Malone, T.W. (1981). Toward a theory of intrinsically motivating instruction. Cognitive
Science, 5, 333-369.

Mann, T., Sherman, D., & Updegraff, J. (2004). Dispositional motivations and message
framing: A test of the congruency hypothesis in college students. Health
Psychology, 23, 330-334.

Matas, L., Arend, R.A., & Sroufe, A. L. (1978). Continuity of adaptation in the second
year: The relationship between quality of attachment and later competence. Child
Development, 49, 547-556.

Matheis, R., Schultheis, M.T., Tiersky, L.A., DeLuca, J., Mills, S.R. & Rizzo, A.A. (in
press). Is learning and memory different in a virtual environment? Journal of
Clinical and Experimental Neuropsychology.

Maturana, H.R. & Varela, F.J. (1980) Autopoiesis and Cognition: The Realization of the
Living. Dordrecht: D. Reidel.

McDonald, D.G., & Kim, H. (2001). When I die, I feel small: Electronic game characters
and the social self. Journal of Broadcasting and Electronic Media. 45, 241-258.

McLaughlin, M. L., Cody, M. J., & Read, S. J. (1992). Explaining one's self to others:
Reason-giving in a social context. Hillsdale, NJ, England: Lawrence Erlbaum
Associates, Inc.

Metzinger, T. (2003). Being no one. The self-model theory of subjectivity. MIT Press,
Cambridge, MA.

Meyers, P., & Biocca, F. (1992). The elastic body image: An experiment on the effect of
advertising and programming on body image distortions in young women.
Journal of Communication, 42, 3, 108-133.

142

Miller, L.C., Appleby, P.R., Read,, S.J., Marsella, S. & Clark, L. (2007). SOLVE-IT:
socially optimized learning in virtual environments utilizing intelligent
technologies. Grant submitted to NIH in consideration of funding, submission
date August, 28
th
,, 2007.

Miller, L.C., & Read, S. J. (2006). Virtual Sex: Creating environments for reducing risky
sex. In S. Cohen, K. Portnoy, D. Rehberger, & C. Thorsen (Eds.) Virtual
Decisions: Digital simulations for teaching reasoning in the social sciences and
humanities. Mahway, N.J.: Lawrence Erlbaum.

Milgram, S. (1963). Behavioral Study of Obedience. Journal of Abnormal and Social
Psychology 67: 371–378.

Monahan, J. L, Murphy, S.T., & Miller, L.C. (1999). When women imbibe: Alcohol and
the illusory control of HIV Risk. Psychology of Women Quarterly, Vol. 23(3),
643-651.

Mroczek, D. K., Almeida, D. M., Spiro, A., & Pafford, C. (2006). Modeling
intraindividual stability and change in personality. Mahwah, NJ, US: Lawrence
Erlbaum Associates Publishers.

Murray, J., Liotti, M., Ingmundon, P.T., Mayberg, H.S., Pu, Y., Zamarripa, F., Yijun, L.,
Woldorff, M.G., Gao, J.H., & Fox, P.T. (2006). Children’s brain activations while
viewing televised violence revealed by fMRI. Media Psychology, Vol. 8, No.1,
25-37.

Murray, J. (1997). Hamlet on the Holodeck: the future of narrative in cyberspace. New
York: Free Press.

Murphy, S. T., Monahan, J. L., & Miller, L.C. (1998). Inference under the influence: The
impact of alcohol and inhibition conflict on women's sexual decision making.
Personality and Social Psychology Bulletin, 24, 517-528.

Norman, P., & Smith, L. (1995). The theory of planned behavior and exercise: An
investigation into the role of prior behavior, behavioral intentions and attitude
variability. European Journal of Social Psychology, 25, 403-415.

Oettingen, G., Grant, H. Smith, P., Skinner, M. & Gollwitzer, P.M. (2006). Non-
conscious goal pursuit: Acting in an explanatory vacuum. Journal of
Experimental Social Psychology. Vol. 42(5), 668-675.

Ouellette, J.A., & Wood, W. (1998). Habit and intention in everyday life: The multiple
processes by which past behavior predicts future behavior. Psychological
Bulletin, 124, 54-74.

143

Palmgreen, P., Donohew, L., Lorch, E., Rogus, M., Helm, D., & Grant, N. (1991).
Sensation seeking, message sensation value, and drug use as mediators of PSA
effectiveness. Health Communication, 3(4), 217-234.

Parker, A.M. & Fischoff, B. (2005). Decision-making competence: External validation
through an individual-differences approach. Journal of Behavioral Decision
Making, 18, 1-27.

Phillips, R. (1945). Doll play as a function of the realism of the materials and the length
of the experimental session. Child Development, 16, 123-143.

Pugnetti, L., Mendozzi, L., Brooks, B.M., Attree, E.A., Barbieri, E., Alpini, D., Motta,
A., & Rose, D.F. (1998). Active versus passive exploration of virtual
environments modulates spatial memory in MS patients: a yoked control study.
Italian Journal of Neurological Science, 19, 424-430.

Rafaeli, S. (1988) ‘Interactivity: From New Media to Communication’, in R. P. Hawkins,
J. M. Wieman and S. Pingree (eds) Advancing Communication Science: Merging
Mass and Interpersonal Processes, pp. 110-134. Newbury Park, CA: Sage.

Read, S. J., Miller, L. C., Appleby, P. R., Nwosu, M. E., Reynaldo, S., Lauren, A. et al.
(2006). Socially Optimized Learning in a Virtual Environment: Reducing Risky
Sexual Behavior Among Men Who Have Sex With Men. Human Communication
Research, 32(1), 1-34.

Reeves, B., & Nass, C. (1996). The media equation: How people treat computers,
televisions, and news media like real people and places. New York: Cambridge
University Press.

Read, S. J., & Miller, L. C. (1998). On the dynamic construction of meaning: An
interactive activation and competition model of social perception. In S. J. Read &
L. C. Miller (Eds.). Connectionist Models of Social Reasoning and Social
Behavior. Mahwah, NJ: Erlbaum.

Read, S. J., Miller, L. C., Appleby, P. R., Nwosu, M. E., Reynaldo, S. Lauren, A., &
Putcha, A. (2006). Socially Optimized Learning in a Virtual Environment:
Reducing Risky Sexual Behavior Among Men Who Have Sex With Men. Human
Communication Research 32 (1), 1-33.

Reingold, H. (2002). Virtual Communities. California: Touchstone Books.

Reyna, V.F. & Brainerd, C.J. (1991). Fuzzy-trace theory and framing effects in choice:
Gist extraction, truncation, and conversion. Journal of Behavioral Decision
Making, 4, 249-262.

144

Reyna, V. F. & Farley, R. (2006). Risk and Rationality in Adolescent Decisión Making:
Implications for Theory, Practice, and Public Policy. Psychological Science in the
Public Interest. Vol 7 No.1.

Rizzo, A.A., Bowerly, T., Shahabi, C., Buckwalter, J.G., Klimchuk, D. & Mitura, R.
(2004). Diagnosing attention disorders in a virtual classroom. Entertainment
Computing, June, 87-89.

Rizzo, A.A., & Kim, G.J. (2005). A SWOT analysis of the field of virtual reality
rehabilitation and therapy. Presence: Teleoperators and Virtual Environments.
14(2), 1-28.

Rokeach, S.B. (1973). From pervasive ambiguity to a definition of the situation.
Sociometry, Vol. 36, No. 3, 378-389.

Rosenthal, T. L., and Zimmerman, B. J. (1978).Social Learning and Cognition,
Academic Press, New York.

Ross, M. W., Henry, D., Freeman, A., Caughy, M., & Dawson, A. G. J. (2004).
Environmental influences on safer sex in young gay men: A situational
presentation approach to measuring influences on sexual health. Archives of
Sexual Behavior, 33(3), 249-257.

Ross, M. W. & McLaws, M.L. (1992). Subjective norms about condoms are better
predictors of use and intention to use than attitudes. Health Education Research,
7, 335-339.

Rothbaum, B.O., Hodges, L., Anderson, P.L., & Smith, S. (2002). 12-month follow-up of
virtual reality exposure therapy for the fear of flying. Journal of Consulting and
Clinical Psychology, 70, 428-432.

Rothman, A.J., & Salovey, P. (1997). Shaping perceptions to motivate healthy behavior:
The role of message framing. Psychological Bulletin, 121, 3-19.

Sabido, M. (2004). The Origins of Entertainment-Education. In A. Singhal, M.J. Cody, E.
M. Rogers & M. Sabido (Eds.), Entertainment-Education and social change:
History, research and practice. LEA’s communication series (pp.61-74).
Mahwah, NJ: Lawrence Earlbaum Associates.

Saladin, M., Brady, K.T., & Graap, K. (2006). A preliminary report on the use of virtual
reality technology to elicit craving and cue reactivity in cocaine dependent
individuals. Addictive Behaviors, Vol 31(10), 1881-1894.

Shapiro, M.A. & Chock, T.M. (2003). Psychological processes in perceiving reality.
Media Psychology, 5(2), 163-198.

145

Sheeran, P., Conner, M., & Norman, P. (2001). Can the theory of planned behavior
explain patterns of health behavior change? Health Psychology. Vol. 20, No. 1,
12-19.

Shoda, Y., & Mischel, W. (1998). Personality as a stable cognitive-affective activation
network: Characteristic patterns of behavior variation emerge from a stable
personality structure. In S. J. Read & L.C. Miller (Eds.). Connectionist models of
social reasoning and social behavior (pp. 175-208). Mahwah, NJ: Lawrence
Earlbaum Associates, Inc.

Sierra, J.J. & Hyman, M.R. (2006). A dual-process model of cheating intentions. Journal
of Marketing Education. 28: 193.

Singer, D.G., & Singer, J.L. (Eds) (2001). Handbook of children and the media.
Thousand Oaks, CA: Sage.

Singhal, A., Rogers, E.M., & Brown, W.J. (1993). Harnessing the potential of
entertainment-education telenovelas. International Communication Gazette. 51,
1-18.
Singhal, A. Cody, M., Rogers, E.M. eds. (2003). Entertainment-education and social
change: history, research, and practice. Lea Communication Series, Lawrence
Earlbaum Associates.

Slater M, Antley A, Davison A, Swapp D, Guger C, et al. (2006) A virtual reprise of the
Stanley Milgram obedience experiments. PLoS ONE 1(1): e39.
doi:10.1371/journal.pone.0000039.

Sloman, S.A. (2002). Two systems of reasoning. In T. Gilovich, D. Griffin, & D.
Kahneman (Eds.), Heuristics and biases (pp. 379-296). New York: Cambridge
University Press.

Slovic, P. (2001). Cigarette smokers: rational actors or rational fools? In P. Slovic (Ed.),
Smoking: Risk, Perception, and policy (pp. 97-124). Thousand Oaks, CA: Sage.

Slovic, P., Finucane, M. L., Peters, E. & MacGregor, D.G. (2004). Risk as analysis and
risk as feelings: Some thoughts about affect, reason, risk, and rationality. Risk
Analysis, Vol. 24. No. 2, 311-322.

Spigel, L. (1992). Installing the television set: Popular discourses on television and
domestic spaces, 1948-1955. In L. Spigel & D. Mann (Eds.), Private Screenings:
Television and the female consumer. Minneapolis: University of Minnesota Press.

Tabachnick, B. G., & Fidell, L. S. (2007). Using Multivariate Statistics (5th ed.). Boston:
Allyn & Bacon.

146

Tamborini, R., Eastin, M., Skalski, P., Lachlan, K., Feduik, T., & Brady, R. (2004).
Violent virtual video games and hostile thoughts. Journal of Broadcasting &
Electronic Media, 48(3), 335-357.

Thorson, E. & Friestad, M. (1985). The effects of emotion on episodic memory for
television commercials, in P. Cafferata & A. Tybout (Eds.), Advances in
consumer psychology (pp. 131.136). Lexington, MA: Lexington.

Thorson, E. & Reeves, B. (1985). Memory effects of over-time measures of viewer liking
and activity during programs and commercials, in R.J. Lutz (Ed.), Advances in
consumer research, Vol. XII. New York: Association for Consumer Research.

Turkle, S. (1995). Life on the screen: Identity in the age of the Internet. New York:
Touchstone.

Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of
choice. Science, 211, 453-458.

Vaccarino, F. J., Schiff, B.B., & Glickman, S.E. (1989). Biological view of
reinforcement. In S. B. Klein, & R.R. Mowrer, Contemporary Learning Theories
(p. 111-142). Hillsdale, NJ: Lawrence Erlbaum.

Vandelanotte, C.V., Bourdeaudhuij, I.D., Sallis, J.F., Spittaels, H., & Brug, J. (2005).
Efficacy of sequential or simultaneous interactive computer-tailored interventions
for increasing physical activity and decreasing fat intake. Annals of Behavioral
Medicine, Vol. 29, No. 2, 138-146.
Verplanken, B., Myrbakk. & Rudi, E. The measurement of habit. In T. Betsch & S.
Haberstroh (Eds.), The routines of decision making. (pp. 231-247). Mahwah, NJ:
Lawrence Erlbaum Associates.

Vorderer, P. (2000). Interactive entertainment and beyond. In D. Zillman & P. Vorderer
(Eds.), Media entertainment. The psychology of its appeal (pp. 21-36). Maywah,
NJ: Lawrence Erlbaum Associates.

Wellman, B., Haase, A.Q., Witte, J., & Hampton, K. (2001). Does the Internet increase,
decrease, or supplement social capital. American Behavioral Scientist. Vol 45,
No. 5, 436-455.

Williams, D. (2006). Virtual cultivation: Online worlds, offline perceptions. Journal of
Communication, 56(1), 69-87.

Whang, L. S.M., & Geunyoung, C. (2004). Lifestyles of virtual world residents: Living in
the on-line game lineage. Cyberpsychology & Behavior, Vol. 7 No. 5: 592-600.

147

Yee, N., Bailenson, J.N., Urbanek, M., Chang, F., Merget, D. (2007). The Unbearable
Likeness of Being Digital: The Persistence of Nonverbal Social Norms in Online
Virtual Environments. CyberPsychology and Behavior. Vol. 10, No. 1: 115-121.

Zimbardo, P. G. (1988). Psychology and life (12th ed.). Glenview, IL: Scott,
Foresman.

Zuccerman, M. (1994). Behavioral expressions and biosocial bases of sensation seeking.
Cambridge, UK: Cambridge University Press.

148

Abstract (if available)

Abstract Would virtual behaviors be predictive of past and future behaviors? Does the relationship between an individual's virtual and subsequent real life behavior depend on active interactive decision-making or is mere passive observation of another's choices enough to predict subsequent real-life behavior? Would virtual decisions (and interventions tied to them) predict behavior 3 months subsequently above and beyond traditional cognitive variables (e.g., intent, self-efficacy)? These questions can be addressed by taking a SOLVE (Socially Optimized Learning in Virtual Environments) approach. To address hypotheses related to these questions, 151 men who have sex with men (MSM) filled out initial baseline measures of their past risk-taking behavior (e.g., use of alcohol, drugs, sexual history) and other measures (e.g., demographics, traditional self-report predictors) and were then randomly assigned to an interactive video (IAV-SOLVE) condition or a non-interactive yoked control condition (choices of another MSM). MSM in the Interactive HIV Prevention Video condition made a range of behavioral choices (e.g., drink alcohol, take drugs, or take sexual risks) on a virtual date. The choices were electronically recorded. MSM assigned to the yoked condition passively observed the choices that had been made by another MSM. Participants in both conditions then answered immediate post-measures (e.g., traditional psychosocial health risk predictor variables). After three months, participants filled out a follow-up survey where risk-taking behavior was again re-assessed.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

When it's good to feel bad: how responses to virtual environments predict real-life sexual risk-reduction

Asset Metadata

Creator Godoy, Carlos Gustavo (author)

Core Title Using virtual environments to unobtrusively measure real-life risk-taking: findings and implications for health communication interventions

School Annenberg School for Communication

Degree Doctor of Philosophy

Degree Program Communication

Publication Date 11/26/2007

Defense Date 10/29/2007

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Bandura,Cody,Damasio,entertainment education,health communication,HIV,interactive,Miller,OAI-PMH Harvest,risk taking,sex,video,virtual

Language English

Advisor Miller, Lynn Carol (committee chair), Cody, Michael J. (committee member), Read, Stephen J. (committee member)

Creator Email cgodoy@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m943

Unique identifier UC1328154

Identifier etd-Godoy-20071126 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-597225 (legacy record id),usctheses-m943 (legacy record id)

Legacy Identifier etd-Godoy-20071126.pdf

Dmrecord 597225

Document Type Dissertation

Rights Godoy, Carlos Gustavo

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu