 

 
 

 

 

 

 

 

 

 

 

 

 

_, » ml. ‘
,.
‘ ...
n ;.
l .
n‘ .
n . ,u ,,
... f
v z .. 1
H u .v.. A
A ‘
o ‘
., 1,1 l ..
“4. ‘n . .
\ . _-.
.. ‘ .‘i, .3

 

 

 

 

 

 

   

 

-. A»
. . .
r .
ﬁ, 2..
. .
.v V ,
. ‘ Z
1 i
.v . V
.
f ,

     

A P}. Lu Nun...

 

.
‘ ,. I
'4 ‘3 \

1)

k

'r .
‘IJ-L‘A’iC
_ ‘ I]

b

‘31

I to

LIBRARIES
MICHIGAN STATE UNIVERSITY
EAST LANSING, MICH 48824-1048

This is to certify that the
dissertation entitled

INTERACTIVE SOCIAL AGENT AND SOCIAL PRESENCE
EFFECTS ON INFORMATION PROCESSING AND PERSUASION

presented by
PAUL SKALSKI

has been accepted towards fulfillment
of the requirements for the

Ph.D. degree in Communication

8\ ’30 ”AQ'E'M‘

 

Major Professor’s Signature

ll 'qugcl

 

Date

MSU is an Afﬁrmative Action/Equal Opportunity Institution

n.---n-.-u-._---n--o-.-u-o-o-o-o----u--n-o-uco--u-n-u-rm:--o-o-o--o-o-o-o—o-o-n-o-a—u-o-n-o-o-o--u-o-c-u-o---.

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 CZICIRC/DateDuetp65-p15

INTERACTIVE SOCIAL AGENT AND SOCIAL PRESENCE EFFECTS ON
INFORMATION PROCESSING AND PERSUASION

By

Paul Skalski

A DISSERTATION
Submitted to Michigan State University
in partial fulﬁllment of the requirements

-for the degree of
DOCTOR OF PHILOSOPHY

Department of Communication

2004

ABSTRACT

INTERACTIVE SOCIAL AGENT AND SOCIAL PRESENCE EFFECTS ON
INFORMATION PROCESSING AND PERSUASION

By

Paul Skalski

This study applied the Heuristic-Systematic Model (HSM) to examine how
interactive social agent technology affects social presence, message involvement, and
information processing leading to persuasion. Two models were tested predicting that
interactive media will increase both social presence and message involvement, which in
turn were expected to affect subsequent information processing styles. Speciﬁcally, social
presence was expected to increase heuristic processing and alter resultant attitudes and
intentions (depending on source attractiveness). Message involvement was expected to
increase systematic processing and lead to more favorable attitudes and intentions toward
a health issue. A two-factor, between-subjects experiment was conducted (N = 125), with
interactivity as the ﬁrst factor (interactive or non-interactive social agent) and source
attractiveness as the second (attractive or unattractive agent). The results of path analyses
suggest that, as expected, interactivity increased social presence, and social presence
related positively to heuristic processing, though this processing did not affect attitude or
intention. Also as expected, message involvement had a positive inﬂuence on systematic
processing, and this type of processing increased behavioral intention toward a blood
pressure checkup. However, the predicted inﬂuence of interactivity on message
involvement was indirect through its inﬂuence on social presence. The results are

discussed in light of research on interactive media, social presence, and the HSM.

Copyright by
PAUL SKALSKI
2004

ACKOWLEDGMENTS

Thanks to everyone who helped with this project, chief among whom stands Ron
Tamborini, my advisor. His keen insight and hard work inspired me to get this done. I
can’t think of a single word that captures my lofty view of Ron, so I’ll create a new one:
wondersplentastic.

Thanks also to my fabulous committee members, Chuck Atkin, Bradley
Greenberg, and Sandi Smith, for their time and guidance on this project, and for giving
me the chance to work with them during my time at MSU. These are experiences I will
always remember fondly.

Finally, thanks a bunch to Mom, for all of her help and support over the years,
and Stacy Fitzpatrick, for encouraging me to not give up during crisis moments (or so

they seemed) and answering the call when help was needed, among other things.

iv

TABLE OF CONTENTS

LIST OF TABLES ................................................................................. vii
LIST OF FIGURES ............................................................................... viii
INTRODUCTION .................................................................................. 1
CHAPTER 1: LITERATURE REVIEW ........................................................ 4
The Heuristic-Systematic Model (HSM) ................................................ 4
Reconsidering the Role of Modality ...................................................... 7
The Role of Social Presence ............................................................... 8
Media Technology and Presence: Prior Work and Unanswered Questions ......... 10
Interactivity and Social Presence ......................................................... 12
Social Agent Technology and Persuasion ................................................ 14
CHAPTER 2: RATIONALE AND MODEL PREDICTIONS ............................... 17
CHAPTER 3: METHODS ........................................................................ 22
Overview .................................................................................... 22
Participants .................................................................................. 22
Procedure .................................................................................... 22
Measures ..................................................................................... 26
CHAPTER 4: RESULTS .......................................................................... 34
Descriptive Statistics ........................................................................ 34
Interactivity Induction Check .............................................................. 37
Descriptive Statistics for Model Variables ............................................... 38
Evaluation of Models ........................................................................ 39
Post-Hoe Analyses .......................................................................... 45
Models Accounting for All Thoughts ..................................................... 51
Models with Recognition Memory Measures ............................................ 56
CHAPTER 5: DISCUSSION ..................................................................... 60
Interactive Media and Social Presence ................................................... 60
The Effect of Social Presence on Heuristic Processing ................................ 61
Message Involvement and Systematic Processing ...................................... 62
The Effect of Systematic Processing on Behavioral Intention ........................ 63
Implications of Revised Models ........................................................... 64
Limitations ................................................................................... 67
Conclusion ................................................................................... 72
REFERENCES ...................................................................................... 74

APPENDICIES ...................................................................................... 79

vi

LIST OF TABLES

Table 1: Descriptive Statistics ..................................................................... 34
Table 2: Zero-Order Correlations between Study Variables .................................. 35
Table 3: Descriptive Statistics for Variables in Attractive-Source Model .................. 39
Table 4: Descriptive Statistics for Variables in Unattractive-Source Model ............... 39

Table 5: Zero-Order Correlations Used to Calculate Parameter Estimates in Model ...... 42

Table 6: Zero-order Correlations Used to Calculate Parameter Estimates in Model. . 43

vii

Figure 1:

Figure 2:

Figure 3:

Figure 4:

Figure 5:

Figure 6:

Figure 7:

Figure 8:

Figure 9:

LIST OF FIGURES

Expected relationships for attractive and unattractive source ..................... 16
Results for attractive source ............................................................ 42
Results for unattractive source ......................................................... 43
Results for revised attractive-source model .......................................... 46
Results for revised unattractive-source model ....................................... 49
Results for attractive-source model with additional paths ......................... 53
Results for unattractive-source model with additional paths ...................... 55
Results for attractive source model with recognition memory measures ........ 57
Results for unattractive source model with recognition memory measures. 58

viii

Introduction

The astounding growth in the past decade of computerized media technology such
as the Internet and virtual reality has generated a great deal of interest in the potential for
such devices to affect persuasion (e.g., Fogg, 2003). These technologies promise to
surpass the persuasive success of traditional media like television through unique features
that allow users to exert greater control over communication processes. The most
prominent of these features is interactivity. Today’s computer-based technologies give
users the ability to interact with mediated people and objects in a manner conducive to
the experience of presence, or “the perceptual illusion of nonmediation” (Lombard and
Ditton, 1997). Understandings of presence are central to logic explaining the effects of
many new technologies, and have been recently applied to research on persuasion in
computer-generated virtual environments (e.g., Grigorovici, 2003; Li, Daugherty, &
Biocca, 2002). However, few studies have addressed the role of presence in new media
persuasion situations involving mediated sources. Given the important role of source
characteristics in prior research on media persuasion, its potential inﬂuence on persuasion
in new media is difﬁcult to overlook.

A focus on characteristics of source has played a substantial part both in
traditional persuasion research, and, coincidentally, in the design of new media hardware
and software. In the former, a sizeable body of literature in persuasion has accumulated
about the effects of such source characteristics as credibility, liking, perceived similarity,
and physical attractiveness (see O’Keefe, 2002 for a review). When these characteristics
are perceived, sources are generally believed to have stronger effects on persuasion. In

the latter case, consideration of source attributes is a central feature of research on new

technologies such as videoconferencing systems and virtual reality (VR) technology.
These machines are being designed to bring people in remote locations “together”
through communication media, or what some refer to as creating a sense of social
presence.

Social presence, simply understood as the feeling of being “with” another
mediated being, has become a major concern among developers of new media technology
because it is believed to enhance the effectiveness of mediated interpersonal and group
interactions. According to this logic, as social presence increases, so does the inﬂuence of
source cues on outcomes of source exposure. Since social presence has the ability to
improve social exchanges and strengthen source cue effects (Lombard & Ditton, 1997;
Skalski, 2004), technology developers are interested in identifying features that increase
the potential for social presence. One of the most striking examples of an application
thought capable of enhancing social presence is social agent technology.

Social agents are deﬁned here as computer software programs that both perform
functions in an “intelligent” way (e.g., helping users on their own initiative) and possess
the ability to interact with humans in a social way (Lieberman & Selker, 2002; Thorisson,
1996). One well known example of a social agent is Clippy, the animated paperclip
assistant in Microsoft’s Word program. The idea behind social agents such as Clippy is to
support computer users by allowing them to obtain information through (personal)
human-like channels instead of (impersonal) machine-like channels. Research on HCI
argues that humans, as a result of evolution, are already hard-wired to respond to
computers as people (Reeves & Nass, 1996), and social agents facilitate this natural

inclination.

Social agents can also be used as instruments for persuasion, though this avenue
has received little attention from scholars to date. Huang (1999) discusses the potential
for social agents to positively affect persuasion in health contexts based on their
interactivity and ability to convey a sense of social presence. He argues that animated
computer characters such as social agents foster a sense of involvement, leading to more
careful information processing and positive change in attitude and intention. While
Huang’s work is mostly non-theoretical and lacks empirical support, related ﬁndings
from studies employing the Heuristic-Systematic Model (HSM) (Eagly & Chaiken, 1993)
offer evidence consistent with this claim. Skalski (2004) suggests that source cues can
affect information and attitude change by prompting heuristic or systematic processing of
persuasive messages. Moreover, he argues that this effect should be enhanced by
characteristics of presence-inducing technology such as vividness and interactivity.

The present study attempts to extend recent work examining the effects of new
media on persuasion within the HSM framework by looking at how interactivity and
social presence affect information processing and persuasion in a health context.
Speciﬁcally, it will look at social presence with a social agent. Social agents, by virtue of
their ability to interact with computer users, are proffered to instill a greater sense of
social presence than non-social agents (e. g., media sources that do not respond to the
user). Subsequently, social presence is expected to affect the systematic and heuristic
processing of information in a persuasive message to inﬂuence resulting attitude and

behavioral intention regarding health issues.

Chapter 1
Literature Review

The Heuristic-Systematic Model (HSM)

Persuasion refers to any attempt to reconﬁgure belief, intention, attitude, and/or
behavior (Shavitt & Brock, 1994) The HSM has played a central role in recent research
on persuasion by explicating the underlying mental processes responsible for these
changes. Like the Elaboration Likelihood Model (ELM) (Petty & Cacioppo, 1981), the
HSM is a dual-process model of persuasion. It identiﬁes two concurrent modes of social
information processing--heuristic and systematic--and attempts to specify conditions that
trigger or govern each (Todorov, Chaiken, & Henderson, 2002). Systematic processing a
“comprehensive, analytic orientation to information processing in which perceivers
access and scrutinize a great deal of information for its relevance to their judgment task”
(Eagly & Chaiken, 1993, p. 326). Heuristic processing is thought of as a “more limited
mode of information processing that requires less cognitive effort and fewer cognitive
resources than systematic processing” (p. 327). This type of processing involves focusing
on a subset of available information that enables the use of simple decision rules (i.e.,

cognitive heuristics) to reach decisions.

One thing that separates the HSM from other dual-process models, such as the
ELM, is the notion that heuristic and systematic processing can co-occur and
simultaneously exert an impact on judgments (Chen & Chaiken, 1999). This is one
reason why many (including this author) prefer the HSM over the ELM. Instead of
processing modes occurring one at a time and having an “either-or” effect, the HSM

allows processing modes to work together for or against persuasive outcomes. For

example, heuristic processing may work against the use of systematic processing when
the implications of message arguments (systematic information) are ambiguous and bias
targets toward the use of heuristic cues to guide their judgments. For example, if a
message about blood pressure has unclear arguments about the importance of getting a
checkup, individuals may be swayed by a negative source cue (e. g., unattractiveness)
away from having a favorable attitude or intention toward a checkup. If the implications
of processing modes are congruent, however, the two can have an additive effect on
attitude and intention (Todorov, Chaiken, & Henderson, 2002). This might happen, for
example, when a positive source cue (e. g., an attractive person) is added to a persuasive
message, in which case the message content and heuristic source cue might exert a
greater impact on persuasion than either of the two elements separately. As these
examples demonstrate, several different types of processing-based outcomes are possible
within the HSM framework, giving the model more explanatory power than the ELM and

similar dual-process models (see Chen & Chaiken, 1999, for a discussion).

The choice of HSM processing mode(s) depends on both motivational and
cognitive factors (Trumbo, 2002). On the motivational side, Todorov, Chaiken, and
Henderson (2002) identify variables such as personal message relevance, need for
cognition, and task importance as factors affecting processing. For example, when a
message has high personal relevance to receivers, receivers should process more

systematically, since they are more motivated to possess the contained information.

The present study is primarily interested in the role of message involvement as a
motivational factor. Message involvement, a variable often examined in ELM research

(simply as involvement), is deﬁned as a state of focusing on and paying careful attention

to a communication message (Andrews, Durvasula, & Akhter, 1990). When this happens,
it should lead to an increased motivation to scrutinize messages using cognitive
mechanisms characteristic of systematic processing. Thus, message involvement is
expected to increase systematic processing and affect subsequent persuasive outcomes in
a predictable and meaningful manner. Though both systematic and heuristic processing
are capable of inﬂuencing attitude and intention, several studies demonstrate that
systematic processing produces greater attitude persistence and attitude-behavior
consistency than heuristic processing (Chaiken, Liberman, & Eagly, 1989). As a result,
systematic processing is generally considered more valuable than heuristic processing,

making it key to the success of a persuasive message.

On the cognitive side, the primary determinant of processing mode is a person’s
ability to process information. Since systematic processing demands more cognitive
resources, Chaiken and Eagly (1993) believe that it might be constrained or disrupted by
situational or individual factors that reduce processing ability. These can include
distraction, time pressure, and communication modality (Todorov, Chaiken, &
Henderson, 2002), the latter of which is of central importance here. When stimulus-rich
modalities such as TV are used to communicate a message, they can foster more heuristic
processing than less rich modalities such as print. This inﬂuence is expected to occur
because the additional stimuli in the more complex modalities can disrupt cognitive
ability by pulling attention away from persuasive message arguments and toward

heuristic information (e.g., the visible source of the message).

The concept of modality has long been used to account for the role of media in

HSM-based persuasion. However, the usefulness of this concept has recently been

challenged due to the emergence of highly vivid and interactive forms of new media
technology that obscure simple modality-based distinctions. The present study attempts
to reﬁne these obsolete conceptions by showing how modality-based effects can be better
explained as resulting from features of media technologies (e.g., interactivity) that

facilitate the experience of presence.

Reconsidering the Role of Modality

Modality is a term used to represent the different types of transmission channels
through which a message is communicated (e.g., print, audio, video). HSM scholars have
long considered modality as something that affects the way individuals process
persuasive information (e.g., Todorov, Chaiken, & Henderson, 2002; Eagly & Chaiken,
1993). The assumption has been that media providing visual or audio cues consume
greater cognitive resources than those providing only print, leading targets to process
information based more on heuristic cues than systematic consideration of message
content. The belief that audio-visual media exposure leads to more heuristic processing is
based primarily on a study by Chaiken and Eagly (1983) conducted before the HSM was
formally stated. In the study, subjects were presented with a persuasive message through
one of three modalities: print, audiotape, or videotape. In addition, the message was
delivered by either a likable or unlikable communicator. The results showed that the
likable communicator was more persuasive in the audio and video modalities (than
through print), whereas the unlikable communicator was more persuasive in the print
modality (than through audio or video). It was concluded that the video and audio
modalities enhanced the salience of communicator information, leading those

characteristics to exert a stronger impact on persuasion. However, more recent research

offers a different understanding of the processes leading to these outcomes. Part of this

reconsideration stems from ambiguity in understandings of modality as a concept.

The usefulness of “modality” as a way to distinguish media inﬂuence is becoming
increasingly problematic due to technological advancements that have occurred in recent
years (Skalski, 2004). The problem stems from the fact that traditional technologies vary
along several dimensions, and this variance occurs not only between modalities but also
within single modalities. For example, today’s televisions vary widely in terms of screen
sizes, from as small as a wristwatch face to as large as the wall of a room. Similarly,
many newer televisions, such as HDTV sets, display more realistic and higher-resolution
images than a traditional TV (Dupagne, 2002). In this manner, there is considerable
variance, within and across different modalities, along media attributes likely to account
for differences in outcomes of persuasive messages. Importantly, these differences are
accentuated in new media like VR and the Internet that offer the ability to interact with
media content. Such changes in technology call for new concepts to account for the
effects of media on information processing, persuasion, and other outcomes of exposure.
Of central importance to the present study is the concept of presence, particularly as it

applies to social-message sources.
The Role of Social Presence

Though this article began by deﬁning presence as “the perceptual illusion of
nonmediation,” (Lombard & Ditton , 1997), current work explicates presence in far
deeper layers of dimensionality and meaning. In a recent article on the concept, Lee
(2004) deﬁnes presence as “a psychological state in which the virtuality of experience is

unnoticed” (p. 32) and identiﬁes three subtypes, in line with earlier work (Biocca, 1997).

The ﬁrst, physical presence (aka telepresence; Steuer, 1995), involves feeling present in
a mediated environment or with virtual objects. The second, social presence, involves
feeling present with another (Biocca, Harms, & Burgoon, 2003). The ﬁnal type, self
presence, occurs when technology users perceive a virtual self as an actual self (Lee,
2004). The second subtype, social presence, is particularly relevant to research concerned

with the effects of source-delivered persuasive messages.

In the context of mediated communication, contemporary drinking about social
presence can be traced to the work of Short, Williams, and Christie (1976), who deﬁned
the concept as “the degree of salience of the other person in [an] interaction and the
consequent salience of the interpersonal relationships” (p. 65). Importantly, this
deﬁnition suggests that social presence is more than just a dichotomous “here or not”
judgment and instead exists along a continuum affected by individual perception and
communication technology. The work of Short et al. on social presence has been adopted
by scholars interested in comparing the appropriateness of different forms of media for
types of social interaction (e.g., Rice, 1993; Walther, 1996). Ultimately, it seems
concerned with the best ways to use media for social purposes.

Recent work on social presence has taken a more user-centered approach (vs.
technology-centered approach) consistent with contemporary notions of presence as a
psychological state (Lombard, 2000; Lee, 2004). In contrast to the work of Short et al.
which focuses on user perceptions of a medium’s ability to make others salient, this work
examines the actual perceived salience of others based on attributes of communication
media (Nowak, 2001). As a result, development of current social presence “theory”

focuses on two fundamental issues: (1) the technology question, i.e., how do changes in

properties of media interfaces affect social presence, and (2) the psychological question,
i.e., how do humans attribute social presence to mediated representations (Biocca, Harms,

& Burgoon, 2003).

In line with this recent thinking, this study adopts the deﬁnition of social presence
by Lee (2004), who refers to it as “a psychological state in which virtual (para-authentic
or artiﬁcial) social actors are experienced as actual social actors in either sensory or
nonsensory ways” (p. 45). According to Lee, social presence occurs when technology
users lose awareness of either the para-authenticity of mediated human sources or the
artiﬁciality of nonhuman social actors. In a videoconference, for example, social presence
would occur if a person appearing in video form (a mediated human source) was thought
of as a being in the same room. However, social presence can also occur in response to
“nonhuman social actors” such as animated virtual characters when a person perceives
the nonhuman social actor as real and present in the same location. As Lee’s deﬁnition
implies, social presence is mainly a function of individual perception. However, the role
of technology in creating a sense of presence must not be overlooked. Examination of
technology’s role in this process shows how the concept of modality can be reconsidered

in presence-related terms.
Media Technology and Presence: Prior Work and Unanswered Questions

Presence has been mainly thought of as a likely outcome of exposure to advanced
media technologies such as virtual reality (VR). However, it can be felt in response to
any medium, ranging from print to television to the Internet. Steuer (1995) identiﬁes two
dimensions along which communication technologies vary in their potential to induce

presence: vividness and interactivity. Vividness, according to Steuer, refers to “the ability

10

of a technology to produce a sensorially rich mediated environment” (p. 41). For
example, an increase in screen size and resolution on a television would make the
technology more vivid. Interactivity refers to “the degree to which users of a medium can
inﬂuence the form and content of the media environment” (p.41). For example, an
increase in the number of the features of a website that can be manipulated would
increase interactivity. The dimensions of vividness and interactivity shift the focus
research in this area from the use of the single categorical-level construct of modality to
the use of two continuous variables that distinguish characteristics of technology.
Notably, the dimensions of vividness and interactivity are consistent with a human-
centered presence approach, since each dimension is discussed in terms of how it might

affect the sensio-motor functions of users.

In line with contemporary new-media scholarship (e.g., Steuer, 1995; Li,
Daugherty, & Biocca, 2002), Skalski (2004) replicated parts of Chaiken and Eagly’s
(1983) experiment and extended this work by reconsidering modality in terms of variance
in media vividness and resulting feelings of social presence. To manipulate vividness,
Skalski exposed subjects to a videotaped, source-delivered persuasive message on a
small-screen, medium-screen, or large-screen television (with a print message included as
a control). He then measured subjects’ levels of experienced social presence, information
processing, and attitude toward the position of the message. Results were consistent with
explanations that vividness increased social presence, as expected, and that social
presence affected both systematic and heuristic processing leading to attitude. These
explanations are challenged, however, by an alternative hypothesis suggesting that the

observed change in social presence was driven mainly by the difference between the print

11

and television conditions, instead of by the screen-size induced vividness. Skalski
concluded that by itself, the screen size manipulation used in his study was not a strong
enough to provide a convincing test of the effects of social presence, especially through a

relatively non-interactive medium like television.

To address these concerns, Skalski suggested that a replication manipulating
media interactivity would provide a better test of the extent to which the social presence-
inducing dimensions of technology can inﬂuence heuristic and systematic processing and
cause subsequent change in intention and behavior. This study would not only be the ﬁrst
attempt to examine the extent to which interactivity affects the heuristic and systematic
processing of persuasive-message content, but it would also add to growing body of
research investigating new media’s ability to stimulate social presence and the alter the
outcomes of mediated persuasion effort. Given the abundance of interactive media now
in existence, the role of this type of technology in persuasion remains an important
unanswered question in efforts to understand how media affects the processes suggested
by the HSM. Attempts to examine this question might begin by determining whether or
not interactivity with a source can increase social presence and source-based heuristic
processing to affect attitude change. However, any attempt to answer these questions

requires a detailed understanding of interactivity.
Interactivity and Social Presence

Interactivity is a key distinguishing feature of many newer media technologies,
including video games and the Internet. Interactivity generally refers to the degree to
which a machine can respond to a user and a user can respond to a machine. In other

words, interactivity might be thought of as the extent to which a user can be a receiver

12

and sender instead of just a receiver (Vorderer, 2000). According to Heeter (1989), the
most interactive media are those which respond to the needs and characteristics of users,
in human-like fashion. Media that respond with natural, human-like responses should
create a stronger sense of social presence because these types of responses come closer to
what users would get through face-to-face interaction, the mode of interaction believed to
be the standard for communication exchanges (Durlak, 1987). Thus, people who use
media that allow them to interact with a source are expected to feel a greater sense of

social presence than those who passively absorb source-communicated information.

One explanation for this comes from recent work emphasizing interactive
behavior, i.e., behavioral engagement, as a determinant of social presence (Biocca,
Harms, & Burgoon, 2003). Behavioral engagement can include face-to-face talking, text
chatting (perhaps considered a new media manifestation of talking), and a host of other
nonverbal communication behaviors (see Burgoon & Hoobler, 2002, for a discussion of
nonverbals). Looking at such a wide variety of interactive behaviors is a rather recent
development in empirical work exploring the link between interactivity and social
presence. Prior to the mid-1990s, most social presence research dealt with limited, low-
bandwidth interactive media, thereby limiting engagement to text-based verbal behavior
and a narrow range of nonverbal behavior (Biocca, Harms, & Burgoon, 2003). In recent
years, the possibilities for behavioral engagement have expanded considerably. High-
bandwidth web sites, immersive VR technology, computer games, and other new
technological developments have opened up a variety of new-media interaction forms
with the potential to create strong social—presence experiences. One such recent

development, and the focus of this research, is social-agent technology.

13

Social Agent Technology and Persuasion

Social agents are autonomous software “creatures” that possess some interaction
“knowledge” as a result of programming, and can therefore engage in social interaction
with people on some level (Thorisson, 1996). According to Kay (1984), the idea for the
agent originated from John McCarthy and Oliver G. Selfridge at MIT in the mid-19505.
They viewed agents as “soft robots” living in computers that could carry out actions for
humans and ask for help or advice when needed, in human terms. But their vision was not
realized until very recently, thanks to advances in computing speed and power and
corresponding increases in artiﬁcial intelligence. As mentioned in the introduction, the
most well know example of an agent is probably Clippy, the animated helper in
Microsoft’s Word program who debuted in 1997. Clippy is the embodiment of many
characteristics associated with agents, such as having a cartoon-like form and
communicating both verbal and nonverbal information to help humans. While Clippy was
not well received by many users (and has since been “ﬁred” by Microsoft, as part of a
publicity stunt), social agents are beginning to emerge as key players in the information
revolution. They now provide customer support for companies like Microsoft and
Symantec and are even being used to route and schedule warplanes on aircraft carriers
(Koprowski, 2003). In addition, researchers are beginning to explore their potential as a

tool for persuasion.

Social agents might be useful in persuasion and other contexts for several,
interconnected reasons. Numerous research ﬁndings demonstrate that people will respond
to computers as if they were real people. Reeves and Nass (1996) conducted a fascinating

series of 35 studies that recreated social and natural experiences with media taking the

14

place of people. They found that, almost without fail, people responded to media as if the
technologies were human. When a computer asked a person to evaluate it, for example,
the person gave a more polite response to the computer than when a third party asked
about the computer. When a computer ﬂattered a person, the person responded positively
to the ﬂattery. This all happened despite the known fact that the computers were just
“hardware and wires.” Reeves and Nass explain these odd reactions as being a result of
people not being evolved to twentieth century technology. The human brain evolved in a
world where only humans displayed social behaviors. When computers display these
behaviors, humans cannot help but respond as if the computers are people. Social agents
help to exploit this natural tendency by “putting a face” on physical objects. Social agents
are also helpful because they can perform tasks for humans at low cost and in user-
friendly fashion (Lieberman & Selker, 2003). Once an agent is programmed, it can
communicate fast and trusted information just as if the information was communicated by
an actual human source. This makes them particularly valuable to companies and
organizations looking to provide information. And since agents interact with users and
offer behavioral engagement, they should stimulate social presence, making them
particularly valuable from the standpoint of information processing and persuasion

(Huang, 1999).

The present study attempts to extend Skalski’s (2004) work on social presence
and persuasion by stimulating social presence through varying forms of social agent-
produced interactivity instead of screen-size produced vividness. This is done by using a
social agent in place of a human videotaped source and manipulating the degree of

interactivity found in the social agent. Use of an interactive agent is expected to instill a

15

higher level of social presence than that observed in Skalski’s one-way persuasive
communication exchange. If Reeves and Nass (1996) are correct in their belief that
people tend to respond to computers as if they were human, social agents should help to
exploit this natural tendency by “putting a face” on computer technologies. The study
will test two models (one with an attractive source and the second with an unattractive
source) positing that interactivity increases both social presence and message
involvement. Message involvement and social presence are expected to affect both
heuristic and systematic processing leading to changes in attitude and intention. A
description of predicted relationships in the two separate models is combined and shown
in Figure 1 below. The relationships are discussed in more detail in the rationale section

that follows.

Figure 1. Expected relationships for attractive and unattractive source

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

+
Social ' +/-Heuristic
Presence Processing (“UH
+ Attitude
Interactivity Intention
+
Message ' +Systematic +
Involvement Processing

 

 

 

 

 

Note. Path signs show model predictions. Signs in parentheses are for the attractive-source model only.
Signs in brackets are for unattractive-source model only. Signs with no parentheses apply to models for
both source types.

16

Chapter 2
Rationale and Model Predictions
This study is designed to demonstrate how interactive sources can affect social
presence, information processing, and persuasion within the HSM framework. The model

includes six paths connecting interactivity to attitude and intention.

To date, no research has attempted to connect interactivity to the HSM, despite
rapid growth in the use of interactive technology for persuasion (Fogg, 2003). While both
vividness and interactivity are identiﬁed as central determinants of presence (Steuer,
1995), only vividness has been connected to the HSM. Skalski’s (2004) research
indicates that vividness can increase social presence and shape subsequent information
processing that leads to and attitude change. However, the fact that the effect of vividness
on social presence was generally weak raises questions about the likelihood that
considerably stronger effects of social presence might result from technology that are
highly interactive. Because interactive media more closely match real-life social
interactions than passive media, a medium with an interactive social agent should create a
stronger sense of social presence than a medium with a passive social agent (i.e., one that
does not interact with the user). This should be evident in increased amounts of social

presence:

PATH 1: An interactive social agent will stimulate a higher level of social

presence than a non-interactive social agent.

In the only scholarly work to date exploring the social presence and persuasion
effects of social agents, Huang (1999) discussed the potential for interactivity to affect

persuasive outcomes in a health communication context. Several empirical studies have

17

demonstrated a positive link between interactivity and desired outcomes of health
communication messages (e.g., Street & Manning, 1997; Street & Rimal, 1997). Though
most of this work has looked at the type of interactivity created by hyperlinked Web
pages, Huang believes that social agent-based interactivity will have a similar effect. Like
the interactivity in Web pages, Huang expects that the interactivity of social agents will
increase message involvement. When users are able to actively shape the manner in
which information is presented, the information should engage them more than when
they see a non-interactive message. In this sense, the type of message involvement talked
about by Huang is the same as the concept of involvement discussed in the HSM and
ELM literatures (e.g., Petty & Cacioppo, 1981; Eagly & Chaiken, 1993), making it an

important mediating variable in this study:

PATH 2: An interactive social agent will stimulate more message involvement

than a non-interactive social agent.

The message involvement and social presence generated by an interactive agent
should have direct effects on information processing. Message involvement is expected to
increase motivation to process, while social presence decreases ability. Though Skalski
(2004) found paths from social presence dimensions to both heuristic and systematic
processing, the high levels of presence anticipated in the present study are expected to
lead speciﬁcally to more source-related thinking (i.e., heuristic processing). This follows
from expectations that a disruption of cognitive resources (or a constraint on ability)
should occur since more attention is called to the source. As such, social presence is
expected to affect heuristic processing. When subjects feel they are “with” a source, the

source information should create a stronger “mental model” or impression of the entity

l8

(Biocca, Harms, & Burgoon, 2003) and affect heuristic processing. The direction of
heuristic processing will be dependent on the nature of the source. If the source is
attractive, an increase in positive source thoughts will be observed. If the source is

unattractive, an increase in negative source thoughts is expected:

PATH 3: Social presence will relate positively to source-related heuristic
processing, with positive thoughts occurring in response to an attractive source

and negative thoughts occurring in response to an unattractive source.

Message involvement, to the contrary, should facilitate systematic processing. As
suggested by Huang (1999), when media users become more involved in a media
experience as a result of interactivity, they should become motivated to pay more
attention to message arguments, fostering greater systematic processing. While this effect
of interactivity might seem at odds with the idea that interactivity-induced social presence
will lead to more heuristic processing, the HSM does specify that heuristic and systematic
processing can co-occur, as mentioned earlier (Chaiken & Maheswaran, 1994). Skalski

(2004) suggests that this could be one of those instances:

PATH 4: Message involvement will relate positively message-related systematic

processing.

Notably, the model does not contain paths from social presence to systematic
processing or from message involvement to heuristic processing. One study does show
that social presence can correlate with systematic processing (Skalski, 2004). However,
no causal logic exists for the effect of source cues on systematic processing. Similarly,
there is no a priori reason to expect that message involvement will affect heuristic

processing since message involvement has to do with attention to the message and

19

nothing else. As such, no prediction is made for paths from social presence to systematic
processing or from message involvement to heuristic processing in this study. Instead, the
model continues with paths from heuristic and systematic processing to attitude and

behavioral intention.

The inﬂuence of heuristic processing on attitude/behavioral intention is the
distinguishing factor between the two hypothesized models. Heuristic processing is
expected to affect attitude depending on the nature of the source. When the message is
presented by an attractive source, heuristic processing of the positive source information
should have a positive effect on attitude and behavioral intention. In contrast, when the
same message is presented by an unattractive source, heuristic processing of negative
source information is expected to bias positive systematic processing (as suggested by
Chaiken etal., 1989) and lead to a less favorable attitude and behavioral intention toward
the same promoted health issue, even when positive systematic processing occurs.
Observations showing differences in the paths from heuristic processing to attitude and
behavioral intention provide important evidence of the underlying processes under
investigation. If only a positive source was used, a positive path from heuristic processing
to attitude and behavioral intention could be interpreted as simply showing that
systematic processing inﬂuences heuristic processing. However, observing paths from
heuristic processing to attitude and behavioral intention that differ as a function of source
attractiveness shows that heuristic and systematic processing can have independent and

conﬂicting effects on attitude and behavioral intention:

PATH 5+: Heuristic processing of attractive source cues will relate positively to

attitude and behavioral intention toward the health behavior promoted.

20

PATH 5-: Heuristic processing of unattractive source cues will relate negatively

to attitude and behavioral intention toward the health behavior promoted.

The ﬁnal path in the two models indicates that the impact of systematic processing is
expected to be a function of the persuasive message strength. Typical messages in a
persuasive health context contain strong statements supporting the adoption of the stated
health position. Given a message that contains arguments for awareness of a health issue,
more favorable thoughts about that issue should be evident. Systematic processing of
these messages should lead to a more favorable attitude and intention toward the behavior

promoted.

PATH 6: Systematic processing of a persuasive health message will relate
positively to attitude and behavioral intention toward the health behavior

promoted.

21

Chapter 3
Methods

Overview

A 2 X 2 between subjects factorial design was used in this study, with source
attractiveness (attractive or unattractive) as the ﬁrst factor and interactivity (interactive or
non-interactive) as the second. The outcome variables included measures of social
presence, message involvement, heuristic processing, systematic processing, attitude, and
behavioral intention.
Participants

A total of 125 undergraduate students (mean age = 21.56) enrolled in introductory
courses at a large Midwestern university (Michigan State University) were recruited for
this study and given course credit for their participation. Participants, 42 of whom were
male, were randomly assigned to one of the four experimental conditions. The non-
interactive, attractive-source condition had 32 subjects in it, and the other three
conditions each had 31 subjects in them.
Procedure

The experiment followed a scripted procedure (the entire script appears in
Appendix A). Upon arriving at a research laboratory, participants were greeted and asked
to ﬁll out a consent form and pre-survey. After completing both forms, a researcher told
them that they would be participating in a study on “people’s reactions to messages.”
They were told that they would view and give their reaction to a “randomly selected”
health message to be communicated through new media technology. After this

introduction, the researcher announced the topic of the speech.

22

The position advocated in the speech was, “Blood pressure matters.” This
relatively easy to understand message told participants about the importance of paying
attention to their blood pressure (see Appendix B for the message elements). Once the
topic was announced and the task introduced, participants were seated four feet in front
of a 6 by 6 foot screen and asked to put on a pair of headphones through which they
would hear the communicator. Then the researcher went behind the screen and pretended
to start the message, which was rear projected onto the screen in a 50-inch window.
Though the subject was lead to believe that the message was computer controlled, it was
actually controlled using the “Wizard of Oz technique” (Tang, 2004) involving a human
“puppeteer” behind a screen. This human controller provided the intelligence of the
source and determined what she would say by selecting response options from a message
generating software program. The on-screen source for the message was one of two
versions (attractive or unattractive) of an interactive social agent named “Cardia.” Both
versions are shown in Appendix C. The computer technology used in this study,
including the interface and Cardia, was adapted from technology created for the
HomeNetToo project (Jackson et al., 2003).

As discussed in Tang (2004), the computer interface for this study was created
using the Macromedia Director program. It spanned across two computer widows. The
ﬁrst window, projected onto the screen in front of which participants sat, showed a
talking agent on the upper-left side and a menu tree on the lower-left side highlighting the
current topic of discussion. Unique to this study, the right side of the window showed
only the ﬁve topics of discussion, as opposed to a variety of graphics and information in

Jackson et al (2003). The change was made to limit the amount of information subjects

23

were exposed to and thereby conﬁne their information processing mainly to source and
message thoughts so that the theoretical processes of interest in this study could be more
easily observed. On the second computer window, visible only to the human controller
behind the screen, colored buttons appeared, each corresponding to a message element in
Appendix B. In short, the message had ﬁve categories of information about blood
pressure. The ﬁrst, “effects,” deﬁned blood pressure and high blood pressure. The
second, “consequences,” discussed some harmful effects of high blood pressure. The
third, “risks,” stated several factors that increase the likelihood of developing high blood
pressure. The fourth, “diet,” discussed prevention of high blood pressure through dietary
change. The ﬁnal category, “other,” discussed some additional ways to control high
blood pressure. The controller could choose the information participants were exposed to
by clicking the buttons using a mouse.

Source attractiveness manipulation. Participants saw one of two different versions
of the animated source created speciﬁcally for this study using the Poser program by
Curious Labs. In the “attractive” condition, the source appeared as a pleasant-looking
woman. In the “unattractive” condition, the source appeared as an unpleasant-looking
woman. The same basic source was used in both cases, but in the unattractive condition,
computer programs were used to make the source unattractive. This unattractiveness
manipulation entailed the following changes to the source using the Poser program:
stretching the face so that it appeared slightly misshapen, making the eyebrows
asymmetrical, putting small bumps on the nose, and faintly discoloring the agent’s teeth.
In addition, the program Audacity was used to change pitch of speaker’s voice so that it

was lower and less “ladylike.” Importantly, Audacity changes the pitch of a sound ﬁle

24

without changing its tempo, so the message was delivered with the same speed and
intonations in both conditions. The pitch manipulation was designed merely to make the
unattractive source less attractive.

Interactivity manipulation. Participants interacted with the source in one of two
different ways. In the “interactive” condition, subjects were able to talk to source and
control the order in which message elements were presented by verbally selecting each of
ﬁve different message categories. These subjects were told by the researcher to chose the
order in which they wanted to listen to all ﬁve categories by selecting them one at a time
in any order they wished. In the “non—interactive” condition, the source simply discussed
the categories point-by-point, in one-way fashion, without giving subjects the ability to
interact by selecting the order. To make sure that subjects in both conditions were
exposed to the same information, all participants were given an “outline” of points to be
addressed in the interaction that appeared on the right side of the screen. Those in the
interactive condition used the categories as a guide to what they wanted to ask the source
to talk about. To keep the interactive and noninteractive messages consistent, the human
controller oversaw all responses of the agent. This person made sure that the experiences
in the two interactivity conditions were as close to one another as possible. Importantly,

all subjects listened to the exact same information about blood pressure contained in the

9, H 9’ 6‘

“effects, consequences, risks,” “diet,” and “other” sections. In the interactive
condition, there were a few small differences. First, subjects were asked to pick
categories and could control the order in which information was presented. Second, the
source asked subjects to repeat one of their category selections. This simple addition was

included as an attempt to create a stronger illusion of social presence. Other than this,

25

however, the experiences were nearly identical and ran between 4.5 and 5 minutes long,
with no subject going longer than 5 minutes.

Immediately following exposure to the persuasive message, participants
responded to the following questions (in order): measures of mode of processing
(systematic versus heuristic), social presence, message involvement, attitude toward the
position advocated in the speech, attitude toward the communicator (included as a
manipulation check of source attractiveness), perceived interactivity (included as a
manipulation check of interactivity), and behavioral intention.

Measures

Scale construction. Conﬁrmatory factor analysis (CFA) was used to test the
content validity of all multiple (three or more) item measures. Scale items were retained
if they passed an internal consistency test, involving (a) a check of face validity and (b)
an examination of factor loadings and errors. Items with poor face validity and factor
loadings of less than .50 and/or greater errors in association with other items than what
would be expected by sampling error were dropped. All of the items except one met the
criteria for factor loadings and passed the tests for error. The reliability of each scale was
assessed using Chronbach’s Alpha ((1).

Pretest. Upon arriving at the laboratory, subjects completed a two-page pretest.
Page one contained a two-item measure of attitude toward blood pressure amidst several
foil items measuring attitudes toward health issues. The two blood pressure items, “It is
important to have blood pressure checked regularly” and “Hypertension is a serious
problem,” were included to provide a baseline measure of attitude toward blood pressure,

since attitudes in general were expected to be high. Responses to these two items were

26

summed to create a measure of pre-attitude toward blood pressure. The reliability of this
index was a low but not unexpected (given only two items) a = .46. Page two of the
pretest asked about demographic information (age, sex, and race) and prior computer
experience. The computer experience measure asked subjects to check one of several
categories indicating their level of computer use in a typical day. The responses were
scored along a continuum tapping computer experience: “0” = 0 hours, “1” = O to .5
hours, “2” = .5 to 1 hour, “3” = 1 to 2 hours, “4” = 2 to 3 hours, “5” = 3 to 4 hours, “6” =
4 to 5 hours, “7” = 5 to 6 hours, “8” = 6 to 7 hours, and “9” = more than 7 hours. All
pretest items except for the foil questions appear in Appendix D.

Mode of processing. Two measures of mode of processing were used in this study
in an attempt to improve upon the measure used in Skalski (2004) and gain a deeper
understanding of new media technology-induced information processing. These two
measures were a thought listing task and a recognition memory test. The thought listing
task is a very direct measure of processing that asks participants what they were drinking
about. In Skalski (2004), thoughts were coded as systematic or heuristic and were found
to be related to several variables of interest in this study. Therefore, this procedure will be
repeated here as the primary assessment of information processing. However, since some
of Skalski’s ﬁndings suggested that systematic and heuristic ones may have been woven
together, the recognition memory test is included here to gain additional insight into how
information is processed, in terms of recall of source and message information. Since the
recognition memory test does not directly examine amount of processing, it is believed to
be a less direct measure of information processing and will therefore be treated in an

exploratory manner. Thus, the coded thought listing was included here as the principle

27

measure of processing mode for use in testing the hypothesized models. The recognition
memory test was included as a secondary measure for exploratory analyses.

First, as in Skalski (2004) and Chaiken and Eagly (1983), processing was
measured by giving subjects three minutes to “list their thoughts and ideas.” This
thought-listing technique is a commonly used method of assessing mental processes, and
is adopted from the work of Cacioppo and Petty (1981). The complete instructions are
shown in Appendix E. Responses to this thought listing task were then scored by two
independent coders as either source (S) oriented or message (M) oriented. In addition,
coders assessed whether statements were positively (+), negatively (-), or neutrally (0)
valenced. This was done to allow for the creation of six composite measures representing
the number of positive, negative and neutral thoughts related to both the source and the
message. Because the model was set up to examine positive and negative thoughts, and
since most thoughts fell into these categories (84%, or 582 out of 697), neutral thoughts
were dropped from further analyses. Intercoder reliability was assessed using Pearson’s
correlation, and results for each type are S+: r = .89; S-: r = .96; M+: r = .90; and M-: r =
.81. Examples of thoughts coded into the positive and negative categories are 8+,
“Spokesperson was effective!”; 8-, “The speaker was kind of scary looking”; M+, “The
message deﬁnitely made me think about my diet”; and M-, “The level of information
attempted to be conveyed was overwhelming.”

Following the procedures of Eagly and Chaiken (1983) and Skalski (2004), all
source thoughts are considered here as evidence of heuristic processing, and message
thoughts are considered to be evidence of systematic processing. While recognizing the

possibility that both source and message thoughts might be processed either heuristically

28

or systematically, the decision to use this coding protocol was made not only based on the
desire to follow earlier procedures but also based on the belief that the particular source
and message elements present in this manipulation minimize the likelihood that source
thoughts will be processed systematically and message thoughts processed heuristically.
The manipulation used in this study was chosen speciﬁcally, in part, because elements of
source attractiveness (based only on physical appearance) seem unrelated to message
strength (a constant in this study). It was expected that neither the attractive or
unattractive source would affect systematic processing about blood pressure beyond the
constant effect on systematic processing posited to result from source interactivity.
Likewise, the blood pressure message was not expected to affect processing of source
information.

For the second, more indirect measure of information processing, subjects were
given a recognition memory test (Shapiro, 1994) asking them to respond “true” or “false”
to statements about the message and the source. A total of 28 statements were included,
14 for systematic processing and 14 for heuristic processing. Seven of each type of
statement were true and the other seven were false. Of the seven false statements used to
assess systematic processing, four stated facts similar to information in the message, and
three stated common myths about blood pressure found on the website AOA-Net. The
complete set of items is shown in Appendix E, with the type of item each represents
shown in parentheses. Responses to these items were tallied and used as measures of
systematic processing (of message information) and heuristic processing (of source

information).

29

More speciﬁcally, systematic processing was assessed through three created
variables. The ﬁrst, message hits, represents the percent of correctly identiﬁed statements
that were in the blood pressure message. A higher percentage of message hits was used as
an indicator of more systematic processing. The second, message-fact false alarms,
represents the percent of statements subjects agreed with that were factually inconsistent
with information presented in the blood pressure message. The third measure, message-
myth false alarms, represents the percent of common myths about blood pressure not
contained in the message that subjects said they saw. These were dealt with separately to
see if the message primed existing knowledge about blood pressure. A higher percent of
false alarms in response to the second variable served as an indicator of less systematic
processing, since it suggests that attention was not paid to aspects of the message. More
false alarms in response to the “myths” variable served as an indicator of more systematic
processing, since it suggests that subjects were thinking about past information they had
heard about blood pressure.

Heuristic processing was assessed through two created variables. The ﬁrst, source
hits, represents the percent of correctly identiﬁed statements about the source. A higher
percentage of source hits was used as an indicator of more heuristic processing. The
second, source false alarms, represents the percent of statements subjects agreed with that
were factually inconsistent with information presented about the source. A higher
percentage of source false alarms served as an indicator of less heuristic processing.

Social presence. Social presence was measured using six items adapted from the
Champness (1973) social-presence scale discussed in Short, Williams and Christie (1976)

and used in Nowak and Biocca (2003). Subjects were asked to indicate the extent of their

30

agreement with each of a series of items on a 7-point scale ranging from “not at all” to
“very much.” Though the original Champness scale included ten items, Nowak and
Biocca found only six of the items to be reliable. Of these, two items apply more to the
medium than to the user. Therefore these two items were not used in this study, since
they were judged to be inconsistent with the Lee (2004) deﬁnition of social presence. The
four remaining items used in this study were: “To what extent was this like a face-to-face
encounter,” “To what extent did you feel in the same room as [your partner] [Cardia],”
“To what extent did you partner [Cardia] seem ‘real’,” and “To what extent were you
able to assess [your partner’s] [Cardia’s] reactions to what you said.” The two additional
items were: “To what extent did you feel like you were with an actual person,” and “How
much did you feel like you were ‘with’ Cardia?” Responses to these six items were
summed to create a measure of social presence ((1 = .91). The complete set of items and
factor loadings are shown in Appendix G.

Message involvement. Message involvement was measured through six, 7-point
Likert items. These items were adopted from work by Andrews, Durvasula, and Akhter
(1990) in research on conceptualizing and measuring the involvement construct. The
items were: “The message was engaging,” (reﬂected) “I did not concentrate on the
message,” “I carefully examined the message,” “I focused on the message,” ”The
message was involving,” and (reﬂected) “I did not focus on the message.” Responses to
these times were summed to create a measure of message involvement (or = .89).
Appendix H shows the complete set of items and factor loadings.

Attitude toward blood pressure. Attitude toward the position advocated by the

speech was measured through six, 7-point Likert items tapping the attitudes of subjects

31

towards the importance of blood pressure. One item, “Blood pressure is not very
important compared to other health issues,” failed the initial CFA test and was dropped.
The ﬁve items were summed to create a measure of attitude toward blood pressure (a =
.81). The complete set of items and factor loadings are shown in Appendix I. The scores
on this variable were subtracted from the pre-attitude scores to create a measure called
attitude change.

Attitude toward the communicator. Consistent with Chaiken and Eagly (1983),
attitude toward the communicator was assessed through 15-point Likert items asking
subjects to rate the speaker on the following dimensions: likable, knowledgeable,
intelligent, competent, warm, pleasing, and friendly. Three new items were added
(creating a total of ten) speciﬁcally tapping dimensions of attractiveness (appealing,
attractive, nice looking). Scores on the pleasing, appealing, attractive, and nice looking
items were used to check the attractiveness manipulation by summing the items to create
an overall measure of source attractiveness (a = .93). All communicator attitude items
and factor loadings (on the attractiveness items) are listed in Appendix J.

Perceived Interactivity. As another manipulation check, subjects also answered
ﬁve, 15-point Likert format questions about how interactive they felt their experience
was. The items were: “I felt like Cardia would respond to me during the message,” “I felt
like I could control how the message was delivered to me,” “I felt like I could respond to
Cardia,” “I saw myself as a message sender and receiver instead of just a receiver,” and
“I felt as though Cardia and I were interacting with each other.” These ﬁve items were
summed to create a measure of perceived interactivity (a = .92). The complete set of

items and factor loadings are shown in Appendix K.

32

Behavioral Intention. Intention was measured by providing participants with an
opportunity to sign up for an “actual” appointment to have their blood pressure tested. If
they chose to sign up, this was counted as behavioral intention. A total of 52 participants

(42%) chose to sign up for the test. The sign-up sheet is shown in Appendix L.

33

Chapter 4
Results
Descriptive Statistics
The descriptive statistics for all measured variables are shown in Table 1. Table 2
shows the correlations among these variables and the two manipulations.
Table 1

Descriptive Statistics

 

 

Mean SD Minimum Maximum

Age 21.56 1.75 18 30
Sex* .66 .47 0 1
Computer Use 3.71 1.82 1 9
Perceived attractiveness 7.80 3.63 l 15
Perceived interactivity 5.75 3.70 l 14
Social presence 3.1 l 1.40 l 6
Message involvement 4.95 l .89 1.50 7
Positive source thoughts .38 .72 3
Negative source thoughts .91 1.03 0 6
Positive message thoughts 2.65 1.54 7
Negative message thoughts .82 1.06 5
%Message hits .84 .14 .29 1
%Message false alarms .44 .23 0 1
%Message false alarms (myths) .24 .25 0 1
%Source hits .73 .18 .29 1
%Source false alarms .06 .09 0 .29
Pre-attitude toward blood pressure 5.21 .99 2.50 7
Attitude toward blood pressure 6.19 .68 3.83 7
Blood pressure attitude change .99 .98 -l .67 4.17
Behavioral intention" .51 .50 0 l

 

* Coded as a dichotomous variable with 0 = male and 1 = female.
"Coded as a dichotomous variable with O = did not sign up for test and 1 = signed up for test.

34

Table 2

Zero-Order Correlations between Studv Viables

 

 

 

 

 

l 2 3 4 5 6 7 8 9 10 11
1. Age 1.00
2. Sex ** -.12* 1.00
3. Computer use .03 .00 1.00
4. Interactivity manipulation .04 -.01 -.01 1.00
5. Attractiveness manipulation .09 -.03 .08 -.01 1.00
6. Perceived interactivity .03 -. 10 -.05 .52* .04 1.00
7. Perceived attractiveness .02 .05 -.06 .09 .70* .29* 1.00
8. Social presence .07 -.06 -.04 .30* .16 .65* .39* 1.00
9. Message involvement .03 -. 16 -. 16 .03 .12 .24* .22* .31* 1.00
10. Positive source thoughts .08 .10 -.10 .16 .20* .27* .33* .38* .11 1.00
11. Negative source thoughts -.03 -.Ol -.01 -.12 -.35* -.30* -.45* -.38* -.28* -.28* 1.00
12. Positive message thoughts -.15 -.07 -.06 .11 .07 .34* .16 .35* .35* .09 -.38*
13. Negative message thought .12 .10 .04 -.04 -.11 -.31* -.27 -.38 -.38* -.01 .25*
14. % Source hits .01 .ll -.10 -.06 .25* -.05 .14 .05 .17 .10 -.13
15. % Source false alarms -.05 -.02 .05 .03 .10 .07 .06 .02 -.02 .08 .00
16. % Message hits .07 .04 -.08 .17 -.20* .06 -.04 -.03 -.03 -.01 .06
17. % Message false alarms -.08 .11 .01 .14 -.10 .19* -.01 .17 -.02 .02 .05
18. % Message myth alarms .03 -.11 -.04 -.01 .12 .10 .12 .07 .28* -.01 -.06
19. Pre-attitude .07 -.04 -.01 -.09 .02 -.ll .12 -.07 .06 -.04 .01
20. Attitude .02 .17 -.12 .07 .02 .02 .17 .02 .26* .17 -.19*
21. Attitude change -.05 .16 -.07 .14 -.Ol .12 .00 .09 .12 .16 .14
22. Behavioral intent *** -.06 -.05 .07 .04 .19* .15 .16 .09 .21"' .14 .15
12 l3 14 15 16 17 l8 19 20 21 22
12. Positive message thoughts 1.00
13. Negative message thought -.48* 1.00
14. % Source hits .08 .011 1.00
15. % Source false alarms .05 .05 .02 1.00
16. % Message hits -.08 .02 -.09 .00 1.00
17. % Message false alarms -.09 -.09 .02 .21 * .12 1.00
18. % Message myth alarms .13 -.17 .12 .07 -.03 -.03 1.00
19. Pre-attitude . 14 -. 15 .04 -.08 .05 .01 .21 * 1.00
20. Attitude .08 . 18* -.02 .00 .17 .19* . 19* .36* 1.00
21. Attitude change -.09 .02 -.05 .08 .06 .12 -.08 -.76* .33* 1.00
22. Behavioral intent .25* -.13 .08 .14 -.06 -.01 .23* .06 .07 -.02 1.00

 

* Signiﬁcant at p < .05 for two-tailed t-test.
** Coded as a dichotomous variable with 0 = male and l = female.

*** Coded as a dichotomous variable with 0 = did not sign up for test and l = signed up for test.

The correlations between the variables in Table 1 and age, gender, and computer

use were examined to get a sense of how these variables were related before subjecting

the primary items of interest to path analyses. If any of the non-hypothesized variables

(i.e., age, gender and computer use) were related to the variables in the models, then they

35

would be included in the model analyses as controls. As Table 2 shows, this did not
appear to be the case - age, gender, and computer use were unrelated to other study
variables. Thus, these variables do not appear to have affected any of the processes or
outcomes of interest.

The results in Tables 1 and 2 were also examined to look for abnormalities and to
determine if the variables in the models were capable of showing the predicted
relationships. If problems were apparent with any variable, then their role in the model
would have to be reconsidered. Most variables appeared to have means, standard
deviations, and bivariate relationships that fall in ranges and directions expected.
However, one notable exception was discovered on measures of blood pressure attitude,
where the limited ranges, small standard deviations, and, in particular, the high means
suggested a ceiling effect. As the Table 1 shows, the mean for attitude on the 7-point
scale was 6.19 and the standard deviation was .68. Scores ranged from 3.80 to 7. On the
pre-test attitude index, the mean was 5.21 and standard deviation .99. This left little room
for change in attitude as a result of the manipulations in this study. Not surprisingly under
these conditions, a scan of the correlations between the attitude change and other
variables in this study showed no substantial relationships. This, coupled with poor
reliability on the pre-test measure of attitude, prompted the decision to not use attitude as
an outcome measure in the results section of this study and instead focus on behavioral
intention. While the inability to predict attitude change is a limitation that will be
addressed in the discussion section, behavioral intention is clearly a more desirable
outcome of persuasive health communication and was therefore used as the outcome

measure in this study.

36

Interactivity Induction Check

An independent samples t-test was used to determine if the interactivity
manipulation in this study increased subjects’ perceived interactivity as scored on the 15-
point scale. The manipulation succeeded. Though perceived interactivity was not high in
either condition, participants in the interactive condition reported signiﬁcantly more
interactivity (M = 7.67, SD = 3.37) than those in the non-interactive condition (M = 3.86,
SD = 2.98), t (123) = -6.68, p < .01, two-tailed.
Source Attractiveness Induction Check

Independent samples t-tests were also used to determine if the source
attractiveness manipulation was successful. First, a t-test on the 15-point scale showed
that the average source attractiveness rating in the appealing condition (M = 10.29, SD =
2.60) was signiﬁcantly higher than that in the unattractiveness condition (M = 5.26, SD =
2.63), t (123) = -10.74, p < .01, two-tailed. Second, a single sample t-test revealed a
signiﬁcant difference between the mean source attractiveness rating in the attractive
condition (M = 10.29) and the midpoint of the scale (8), t (62) = 6.99, p < .01, two-tailed.
This demonstrates that the attractive source was not only seen as more attractive than the
unattractive source, but also as signiﬁcantly above the midpoint on attractiveness. Third,
a ﬁnal single sample t-test compared the midpoint of the scale to the mean source
attractiveness rating in the unattractive condition, and this difference was signiﬁcant as
well. The average source attractiveness rating in the unattractive condition (M = 5.26)
was signiﬁcantly lower than the midpoint of the scale, t (61) = -8.18, p < .01, two-tailed,
indicating that the unattractive source was viewed as signiﬁcantly below the midpoint on

attractiveness.

37

A further examination of the effect of source attractiveness was done by running
independent samples t-tests to compare the number of positive and negative source
thoughts (as coded though the thought listing task) in the attractive and unattractive
source conditions. The results of these analyses ﬁrst indicate that study participants had
signiﬁcantly more positive source thoughts in the attractive source condition (M = .52,
SD = .82) than in the unattractive source condition (M = .24, SD = .56), t (123) = -2.24, p
< .05, two-tailed. The second ﬁnding of these analyses is that participants had
signiﬁcantly less negative source thoughts in the attractive condition (M = .56, SD = .84)
than in the unattractive condition (M = 1.27, SD = 1.09), t (123) = 4.14, p < .01, two-
tailed. All of the above results point to a successful manipulation of source attractiveness
in this study, indicating that the manipulated groups represent the two populations desired
for testing the predicted models. As such, data from the two groups were analyzed
separately to test the different path models hypothesized for the attractive and
unattractive source.

Descriptive Statistics for Model Variables

Before running the path analyses, descriptive statistics for all variables in both the
attractive- and unattractive-source models were examined separately. Descriptive
statistics for the attractive source are shown in Table 3, and those for the unattractive

source are shown in Table 4.

38

Table 3

Descriptive Sgtistics for Variables in Attractive-Source Model

 

 

Mean SD Minimum Maximum
Social presence 3.33 1.38 l 6
Message involvement 5.09 1.21 1.5 7
Positive source thoughts .52 .82 0 3
Positive message thoughts 2.65 1.54 0 6
Behavioral intention“ .51 .50 O l

 

*Coded as a dichotomous variable with 0 = did not sign up for test and l = signed up for test.

Table 4

Descriptive Statistics for Variables in Unattractive-Source Model

 

 

Mean SD Minimum Maximum
Social presence 2.89 1.41 l 5.5
Message involvement 4.80 1.15 1.5 7
Negative source thoughts 1.27 1.09 0 6
Positive message thoughts 2.44 1.66 0 7
Behavioral intention* .32 .47 0 1

 

*Coded as a dichotomous variable with 0 = did not sign up for test and 1 = signed up for test.

Evaluation of Models

Path analysis was performed on the hypothesized models using the least squares
method. This involves estimating the sizes of the model parameters and testing the
overall model ﬁt. Parameter size was estimated by regressing each endogenous variable
onto its causal antecedent, and model ﬁt was tested by comparing estimated parameter
sizes to the reproduced correlations (see Hunter & Gerbing, 1982, for a more complete
description of this analysis procedure). In short, a model that is consistent with the data is
one which (a) passes the test of overall model ﬁt, indicated by a non-signiﬁcant chi-

square goodness of ﬁt result, (b) has substantial path coefﬁcients, and (c) has differences

39

between parameter estimates and reproduced correlations (errors) that are no greater than
what would be expected through sampling error. For a model to be judged consistent with
the data, it had to pass all three of the above criteria. First in order to pass the test of
overall model ﬁt, a non-signiﬁcant chi-square test would have to be observed. Second, a
substantial path coefﬁcient is one that passes a test of statistical signiﬁcance at p < .05.
Third, for parameter error to be acceptable, all error terms must satisfy a conservative
criterion for z-differences set at p <.10. The PATH program was used to determine if
each model met these rigid criteria. It should be noted that the correlations reported in the
tables below were corrected for attenuation due to measurement error during the analysis
procedure. Both attractive- and unattractive-source models were tested and the results of
each test will be described in turn below.

Tests for the attractive-source model. The attractive-source model hypothesized
that interactivity would increase social presence and message involvement, both of which
were expected to affect information processing. Speciﬁcally, social presence was
expected to increase heuristic processing in the form of positive source thoughts (since
the thoughts were in response to an attractive source), and message involvement was
expected to increase systematic processing in the form of positive message thoughts
(since these thoughts would represent agreement with arguments in the persuasive
message). This processing was expected to have a positive effect on intention toward the
health issue. Since the impetus behind this research was to examine the role played by
interactivity and social presence in persuasion, this and all models were inspected for

evidence of substantial continuous paths from interactivity to intention. A model without

40

this type of continuous path was deemed incapable of showing support for the logic
underlying this study.

Results for the attractive-source model are shown in Figure 2 below, and
correlations used to test the model are shown in Table 5 below. Inspection of these results
shows two important things. First, the predicted model did not meet the three criteria
established to determine if the model was consistent with the data. Second, the observed
model does not show the type of substantial continuous paths from interactivity to

intention necessary to support for the logic underlying this study.

Notably, some observations were consistent with evidence of a good model ﬁt.
The chi-square test of goodness of ﬁt was non-signiﬁcant, x2 (9) = 7.08, p = .63.
Moreover, most of the predicted paths were signiﬁcant and in the expected directions.
Interactivity had a signiﬁcant positive effect on social presence, and social presence had a
signiﬁcant positive effect on heuristic processing. In addition, message involvement had
a signiﬁcant positive effect on systematic processing, which had a signiﬁcant positive
effect on behavioral intention. However, some observations were more problematic. First,
examination of the differences between predicted and obtained correlations for all
unconstrained bivariate relationships revealed a large residual between the predicted and
obtained correlation for the association between social presence and message
involvement (difference = .35, z = 1.77, p = .08). Second, two of the predicted path
coefﬁcients were insubstantial. Message involvement was not affected by interactivity,
and behavioral intention was not affected by heuristic processing. Third, these two
insubstantial coefﬁcients produced a break in the continuous path from interactivity to

intention needed to support the models underlying logic. Thus, although the chi-square

41

test of goodness of ﬁt was non-signiﬁcant, the large unpredicted relationship along with

paths that were weak and interrupted forced a decision to reject this model.

Figure 2. Results for attractive source

 

 

Socral ‘42,, +Heurrstrc .11
Presence Processrng

 

 

 

 

 

.32*

 

 

 

, , Intention
Interactrvrty

 

 

 

 

 

 

 

b Message .38* Systematic .24:
Involvement Processing

 

 

 

 

 

 

 

* Signiﬁcant at p < .05.

Table 5

Zero-order Correlations Used to Calculate Parameter Estimates in Model

 

 

1 2 3 4 5 6
1. Interactivity 1.00
2. Social presence .31* 1.00
3. Message involvement .04 .33* 1.00
4. Heuristic processing .15 .40* .08 1.00
5. Systematic processing .12 .29* .36* .10 1.00
6. Behavioral intention -.05 .05 .21 .13 .25* 1.00

 

Note. Interactivity was coded such that l = interactive and 0 = non-interactive. Behavioral intention was
coded such that 1 = signed up for test and O = did not sign up for test. * indicates p < .05, two-tailed.

Tests for the unattractive-source model. Similar to the ﬁrst model, the
unattractive-source model hypothesized that interactivity would increase social presence
and message involvement, both of which were expected to affect information processing.
In this case, social presence was expected to increase heuristic processing in the form of

negative source thoughts (since the thoughts were in response to an unattractive source),

42

and, once again, message involvement was expected to increase systematic processing in
the form of positive message thoughts. In contrast to the ﬁrst model, however, the
negative source thoughts were expected to have a negative effect on intention, while the
positive message thoughts were again expected to have a positive effect on intention. The
results for this model are shown in Figure 3 below, and the correlations used to calculate
the model are shown in Table 6.

Figure 3. Results for unattractive source

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Socral _. 41 * - Heuristic “02
Presence Processrng
.30* I .
. . ntentron
Interactrvrty
b Message '35:: Systematic .23
Involvement Processing

 

 

 

 

 

 

* Signiﬁcant at p < .05.

Table 6

Zero-order CorrelLions Used to Calculate Parameter Estimates in Model

 

 

1 2 3 4 5 6
l. Interactivity 1.00
2. Social presence .29* 1.00
3. Message involvement .03 .27* 1.00
4. Heuristic processing -.16 -.39* -.27* 1.00
5. Systematic processing .11 .40* .33* -.53* 1.00
6. Behavioral intention .14 .09 .16 -. 14 .24 1.00

 

Note. Interactivity was coded such that 1 = interactive and 0 = non-interactive. Behavioral intention was
coded such that l = signed up for test and 0 = did not sign up for test. * indicates p < .05, two-tailed.

43

Little evidence supported the unattractive—source model. First, the model failed
the chi-square test of overall ﬁt, x2 (9) = 18.89, p = .03. Second, the two paths found
insubstantial in the ﬁrst model (from interactivity to message involvement and from was
heuristic processing to behavioral intention) were near zero in this model, and a third
(systematic processing to behavioral intention) was only marginally substantial. This
once again produced a break in the continuous path from interactivity to behavioral
intention. More pointedly in this regard, although the other three predicted paths were
signiﬁcant, the direction of the sign for one of these paths (from social presence to
heuristic processing) was opposite to predictions. Third, individual link analyses again
showed substantial errors. Two residuals were substantial and signiﬁcantly different than
what would be expected through sampling error: the correlations between social presence
and systematic processing (difference = .42, z = 2.19, p = .05 and heuristic processing
and systematic processing difference = -.53, z = -2.92 p = .01. Thus, once again, the data
forced a decision to reject this model.

Although tests on data for both the attractive and unattractive source conditions
failed to produce the type of evidence needed to conclude that the data provide a good ﬁt
for the hypothesized model overall, the outcomes observed show patterns in line with the
underlying logic for the model. In this regard, I am hesitant to dismiss the hypothesized
model as completely uninforrnative. As already indicated, the data are consistent with
many of the paths predicted. Moreover, the observed error terms suggest that small
changes might produce a model that is both consistent with the underlying logic in the
original model and provides a strong ﬁt with the data. Keeping in mind the problems

endemic in the use of path analysis for testing non-hypothesized models, post-hoe

analyses were conducted to test a revised model using the data from both the attractive-
and unattractive-source conditions.
Post-Hoc Analyses

Revisions in the post-hoe attractive-source model began by simultaneously
considering two problems: the large residual error found for the predicted and obtained
correlations between social presence and message involvement, and the unsubstantial
path observed from interactivity to message involvement. The large residual error
suggested that social presence and message involvement were somehow related. In
addition, an undetected relationship between these two variables might account for the
weak path observed from interactivity to message involvement. More importantly, it was
apparent that inclusion of a path between these two variables would have implications for
factors affecting all other criteria used to test the model. All this suggested that a model
with a link from social presence to message involvement would better ﬁt the data, and led
to the decision to test an alternate model. Two changes were made in this model. First, a
path from social presence to message involvement was added. While the original model
posited that interactivity would affect both social presence and message involvement
directly, the revised model suggests that the affect of interactivity on message
involvement is indirect. Interactivity affects message involvement through its inﬂuence
on social presence. This minor change is easily explained given that social presence
would likely be activated when an encounter begins (upon perceiving the source) and
message involvement activated as the encounter progressed (as message elements were
communicated). Second, the non-signiﬁcant path from interactivity to message

involvement was dropped.

45

The decision to test an alternate model was made after careful consideration of the
problems evident in the use of path analysis for testing revised models. As Holbert and
Stephenson (2002) contend, models created through post-hoe respeciﬁcation are often
difﬁcult to replicate. For this reason, any decision to undertake this type of analysis
should be made judiciously and with the understanding that the results of this type of test
should be used only as a guide future replication. In the present case, the decision to test a
revised model was made because the changes in the revised model were deemed minor.
More importantly, the changes were consistent with the a priori logic in the original
model, which was chieﬂy concerned with understanding how interactivity affects
information processing and persuasion. Given these considerations, the replication
problems typically associated with testing respeciﬁed models should be minimized.

Figure 4 shows the results for the revised attractive-source model.

Figure 4. Results for revised attractive-source model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Social * + Heuristic

Presence '42 Processing '1 1

- . 32* Intention
Interactrvrty 37*
I

* III

Message '38 Systematic ' 4
Involvement Processing

 

 

 

 

 

 

* Signiﬁcant at p < .05.

The revised model was very strongly supported overall, and all but one of the
paths was ample and signiﬁcant. Interactivity increased social presence, with a path
coefﬁcient of .32, P (.20 < p < .44) = .95. Social presence, in turn, increased both

message involvement (path coefﬁcient = .37, P (.11 < p < .63) = .95) and heuristic

46

processing (path coefﬁcient = .42, P (.20 < p < .64) = .95). Message involvement
increased systematic processing, with a path coefﬁcient of .38, P (.14 < p < .62) = .95,
and systematic processing increased intention, path coefﬁcient = .24, P (.00 < p < .48) =
.95. Although heuristic processing did not signiﬁcantly affect intention (path coefﬁcient
= .11, P (-.15 < p < .37) = .95), this path was in the right direction and, as the conﬁdence
interval demonstrates, more likely than not had a positive effect, albeit a weak one.
Moreover, even without this path, a chain of signiﬁcant paths was found from the

exogenous interactivity variable to the dependent intention variable.

In addition to having substantial path coefﬁcients, the revised attractive-source
model fared well on the second and third tests for model evaluation. The differences
between predicted and obtained correlations for all unconstrained bivariate relationships
were examined, and none were signiﬁcantly different than what would be expected
through sampling error. Furthermore, this model passed the global test of goodness of ﬁt,
)8 (9) = 1.89, p = .99. Though the original model also passed this test, the revised model
result is a marked improvement over the original result of x2 (9) = 7.08, p = .63. Thus, the
analysis of this model shows several substantial path coefﬁcients, no signiﬁcant errors,
and an easily passed global test of goodness of ﬁt. Moreover, the model showed a

continuous path from interactivity to behavioral intention.

Upon ﬁrst look, the problems with the unattractive-source model appear
substantial, and raise great concern about the potential problems associated with
interpreting tests on re-speciﬁed models. However, closer inspection reveals a revised
model with the potential to be replicated. Notably, though initial analyses did not show a

continuous path from interactivity to intention, a conﬂuence of observations suggests the

47

existence of this connection. Speciﬁcally, errors in the unconstrained bivariate
relationships showing sizeable correlations from both social presence and heuristic
processing to systematic processing (r = .40 and r = -.53 respectively), along with the
considerable negative association between social presence and heuristic processing
observed in the original model, signals the presence of an unspeciﬁed negative path from
heuristic to systematic processing. In other words, the strong negative inﬂuence of social
presence on heuristic processing and the residual error showing a strong positive
inﬂuence of social presence on systematic processing suggests that the positive inﬂuence
of social presence can be explained by an indirect path from social presence through
heuristic processing and on to systematic processing. In this case, given the negative path
from social presence to heuristic processing, to positive inﬂuence of social presence on
systematic processing can only be explained by a negative path from heuristic processing
to systematic processing, as the residual term indicates.

The inclusion of this path is strongly called for by several factors. First, it should
help to produce the required continuous path from interactivity to intention. Second, it
should reduce error in the model considerably. Third, it would produce a model that
closely mirrors the model conﬁrmed for the attractive source while remaining consistent
with theoretical logic underlying the original model hypothesized for the unattractive
source. The apparent similarity in data patterns for the attractive-source and unattractive-
source groups suggest that this revised model, which closely replicates the model already
observed for the attractive-source group, would be a good ﬁt for the unattractive-source

data as well.

48

Based on these observations, the decision was made to test an alternative model. The
revised unattractive-source model, like the revised attractive-source model, added a path
from social presence to message involvement and deleted the path from interactivity to
message involvement. Further, the indicated path from heuristic processing to systematic
processing was added. Notably, given this additional revision, the model tested here is
more open to concerns about replication problems (Holbert & Stephenson, 2002). At the
same time, the fact that the revised model closely resembles the model observed for the
attractive source increases conﬁdence in this model’s potential to be replicated. The

results for this model are shown in Figure 5.

Figure 5. Results for revised unattractive-source model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Social - Heuristic
Presence “41* Processing
. . 30* Intention
Interactivrty 30* _ 47*
i' ii '

.22 S .

Message ystematrc

Involvement Processing

 

 

 

 

 

 

* Signiﬁcant at p < .05.

The results for the revised unattractive-source model were generally in line with
predictions. As expected, interactivity had a positive impact on social presence, path
coefﬁcient = .30, P (.08 < p < .42) = .95. This social presence, in turn, inﬂuenced
heuristic processing (path coefﬁcient = -.41, P (-.52 < p < -.30) = .95) and message
involvement (path coefﬁcient = .30, P (.04 < p < .56) = .95). Heuristic processing
inﬂuenced systematic processing (path coefﬁcient = -.47, P (-.27 < p < -.67) = .95, as did

message involvement, thought the path only approached signiﬁcance, path coefﬁcient =

49

.22, P (-.04 < p < .48) = .95. Of the ﬁnal two paths in the model, the one from systematic
processing to intention was sizable and in the expected direction but only close to
signiﬁcant, path coefﬁcient = .23, P (-.05 < p < .51) = .95 . The other path from heuristic
processing to intention was, as in the attractive-source model, virtually non-existent, path
coefﬁcient = -.02, P (-.32 < p < .28) = .95.

The revised unattractive-source model fared well and showed a marked upgrade
over the original model on most model evaluation tests. The differences between
predicted and obtained correlations for all unconstrained bivariate relationships were
examined, and none were signiﬁcantly different than what would be expected through
sampling error. In addition, the model passed the global test of goodness of ﬁt, x2 (8) =
2.51, p = .96. Not only does this show that the data ﬁt the model well, but it shows a
considerable improvement over the ﬁt for the original unattractive-source model which
had a failed chi-square test value of xz (9) = 18.89, p = .03. Though the model-evaluation
criteria equate substantial paths with conﬁdence intervals that do not cross zero, the small
shortcoming in the strength of the paths from message involvement to systematic and
from systematic processing to intention (represented by the minor extent to which the
conﬁdence intervals crossed over zero) can be explained by a lack of statistical power in
this study. This notion is supported by a test of the unattractive-source model using data
from all respondents — with the data from both groups included, the path from message
involvement to systematic processing is .28, P (.10 < p < .46) = .95, and the path from
systematic processing to intent .23, P (.05 < p < .41) = .95 . Neither of these paths cross
zero. Finally, with these paths in place, the revised unattractive-source model has an

unbroken chain of links from interactivity to behavioral intent.

50

It is important to recognize that the affect of social presence on heuristic
processing was an inverse one, such that social presence decreased heuristic processing in
the form of negative source thoughts. This is not surprising when we understand that in
the unattractive-source model this inverse relationship shows simply that social presence
reduced the negative source thoughts used to represent heuristic processing. The
implications of this ﬁnding will be talked about more in the discussion section.

Models Accounting for All Thoughts

One potential criticism of the models tested above is that they do not account for
all types of relevant heuristic and systematic processing. In other words, negative source
thoughts and negative message thoughts are not included in the attractive-source model,
and positive source thoughts and negative message thoughts are not included in the
unattractive-source model. The decision to do this was made for two reasons. First,
speciﬁc types of thinking were predicted in the study hypotheses. An attractive source
communicating a persuasive message about blood pressure was expected to generate
positive source and positive message thoughts. There was nothing in the study logic
suggesting that such a source would generate negative source thoughts or negative
message thoughts. Second, not including these types of thoughts made the models more
parsimonious. However, to circumvent criticisms of these models on the grounds of
hiding information, the decision was made to retest the revised models for attractive and
unattractive source with additional modiﬁcation to include all types of thinking.

Changes to the models were made wherever there were paths to or from
systematic and heuristic processing. Speciﬁcally, the single path from social presence to

positive source thoughts (heuristic processing) in the attractive-source model became two

51

 

paths, one to positive source thoughts and the second to negative source thoughts.
Similarly, the single path from social presence to negative source thoughts (heuristic
processing) in the unattractive-source model became two paths, one to negative source
thoughts and the other to positive source thoughts. The paths from message involvement
to positive message thoughts (systematic processing) in both models became two paths in
each, one from message involvement to positive message thoughts and the other from
message involvement to negative message thoughts. Finally, the paths from systematic
and heuristic processing to intention were expanded to include all types of processing,
such that positive source thoughts, negative source thoughts, positive message thoughts
and negative message thoughts all led to intention.

The results for the attractive-source model are shown in Figure 6 below. These
results are consistent with (and in some places more favorable than) the results for the
previous attractive-source model. Interactivity had the same signiﬁcant positive effect on
social presence, path coefﬁcient = .32, P (.20 < p < .44) = .95. Social presence had a
signiﬁcant effect on both types of heuristic processing — the path coefﬁcient to positive
heuristic processing was once again .42, P (.20 < p < .64) = .95), and the new path to
negative heuristic processing was -.31, P (-.55 < p < -.07) = .95). Social presence also,
once again, affected message involvement with a path coefﬁcient of .37, P (.11 < p < .63)
= .95. Message involvement had signiﬁcant relationships with both types of systematic
processing — the path to positive systematic processing was again .38, P (.14 < p < .62) =
.95, and the new path to negative systematic processing was -.38, P (-.62 < p < -.l4) =
.95 . Of the ﬁnal paths from the four processing types to intention, only one was

signiﬁcant, the path from positive message thoughts to intention, path coefﬁcient = .28, P

52

(.00 < p < .56). This path gave the model a string of signiﬁcant links from interactivity to
intention.

Figure 6. Results for attractive-source model with additional paths

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

+ Heuristic
Processing
.4
509131 _ 31,. - Heuristic
Presence —' "1 Processing
. . 32* Intention
Interactrvrty 37*
I
* .
Message '38 + Systematic

 

 

 

 

 

 

 

Involvement Processing

- Systematic
Processing

 

 

 

* Signiﬁcant at p < .05.

The paths to the additional positive and negative thoughts are almost reﬂections
of the original thoughts of each type. This makes sense considering that they are
opposites. For example, because positive source thoughts (positive heuristic processing)
increase as a result of social presence, it makes logical sense that negative source
thoughts (negative heuristic processing) would decrease in response to them. This pattern
can be seen above for both heuristic and systematic processing. The new attractive-source
model had one large error in unconstrained bivariate relationships, a difference of -.34 (z
= -1.96, p = .05) for positive and negative systematic processing, but this relationship can
be expected for the reasons just discussed. Positive and negative message thoughts are
simple opposites. Overall, this model passed the global goodness of ﬁt test, x2 (18) =
9.30, p = .95. Therefore, it can be viewed as consistent with the data and in line with the

revised attractive-source model preceding it.

53

 

The results for the unattractive-source model are shown in Figure 7 below. These
results are generally consistent with previous models. As in the revised unattractive-
source model, interactivity had a signiﬁcant positive association with social presence,
path coefﬁcient = .30, P (.08 < p < .42) = .95 . Social presence had signiﬁcant effects on
both types of heuristic processing — the effect on negative heuristic processing was again
-.41, P (-.52 < p < -.30) = .95, and the path coefﬁcient to positive heuristic processing
was .32, P (.08 < p < .56) = .95. The effect of social presence on message involvement, as
in the prior model, was .30, P (.04 < p < .56) = .95. Message involvement again had a
near signiﬁcant effect on positive systematic processing, path coefﬁcient = .22, P (-.04 <
p < .48) = .95, and it had a signiﬁcant effect on negative heuristic processing, path
coefﬁcient = -.41, P (-.65 < p < -.17) = .95. Negative heuristic processing also, as in the
previous model, had a signiﬁcant effect on positive systematic processing, path
coefﬁcient = -.47, P (-.67 < p < -.27) = .95. Of the ﬁnal four paths from types of
processing to intention, only one approached signiﬁcance, the path from positive message

thoughts to intention, path coefﬁcient = .21, P (-.11 < p < .53) = .95 .

54

 

Figure 7. Results for unattractive-source model with additional paths

 

+ Heuristic
Processing

V .06
Social

_ 41,. - Heuristic .01

Presence —‘—-1 Processin
.30* g

Interactivity 30*
I ' I

Message ’22 + Systematic

Involvement Processing -.06
.k

- Systematic
Processing

 

 

 

 

 

 

 

 

 

Intention

 

 

 

}-.47*

 

 

 

 

 

.21

 

 

 

 

 

 

 

 

 

 

* Signiﬁcant at p < .05.

On the other model evaluation tests, mixed results were found. The differences
between predicted and obtained correlations for all unconstrained bivariate relationships
were examined, and two were signiﬁcantly different than what would be expected
through sampling error. These included residuals of -.35 for the association between
positive and negative systematic processing (z = -1.94, p = .05), and -.35 for the
association between social presence and negative systematic processing (2 = -l.86, p =
.06). The ﬁrst error might be expected for reasons discussed earlier. The second was
unexpected. This might simply be due to chance, since the inclusion of additional
variables increases the probability of observing chance errors. Nevertheless, it cannot be
completely ignored. In spite of these errors, this model passed the global goodness of ﬁt
test, )8 (17) = 13.30, p = .72 and generally appears to be consistent with prior models

supporting the underlying logic of this study.

55

Models with Recognition Memory Measures

To gain further insight into how participants processed information in this study,
exploratory analyses were conducted using the recognition memory results in place of the
thought listing measures in the models accounting for all thoughts. These changes were
made simply by replacing positive source thoughts and negative source thoughts with
source hits and source false alarms, respectively, and then by replacing positive message
thoughts and negative message thoughts with message hits and both message false alarms
and message myth alarms, respectively. Despite the recognition memory results being far
less direct measures of information processing than the thought listings, the recognition
memory measures can still offer insight into the way in which information was processed
by participants in this study. Evidence of more “hits” in response to either social presence
or message involvement, for example, would indicate that subjects remembered more
about those aspects of the experience, and this could mean that that they were paying
closer attention to the source and/or message and perhaps even thinking about each more.
Evidence of more “false alarms” in response to social presence or message involvement
would indicate less attention or processing, and evidence of more “myth alarms” in
response to message involvement would likely indicate more systematic processing in the
form of priming existing schemas about blood pressure.

The results for the attractive source model are shown in Figure 8 below. As
expected, the paths from interactivity to social presence and social presence to message
involvement were identical. Of the remaining paths, only one was signiﬁcant, the link
between message involvement and message myth alarms, path coefﬁcient = .26, P (.00 <

p < .52) = .95. In fact, an examination of all path coefﬁcients in this model shows that

56

 

this model failed on the criteria of having substantial path coefﬁcients—of the 12 links in
the model, only three were signiﬁcant with a fourth, the path from message myth alarms
to intention, approaching signiﬁcance, path coefﬁcient = .22, P (-.04 < p < .48) = .95. As
a result, the model was rejected. The evidence in this model suggests that social presence
and message involvement had little or no impact on recognition memory, though it does
appear to have primed existing knowledge about blood pressure as evidenced by the
paths through myth alarms.

Figure 8. Results for attractive source model with recognition memory measures

 

 

 

 

 

    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Source
Hits
.03
309131 .12 Source Fls.
Presence L—Dl Alarms
. . “32* Intention
Interactrvrty 37,,
‘.
Message ‘01 Message
Involvement Hits
k.
.26* Message Fls.
Alarms
essage Myth
Alarms

 

 

 

* Signiﬁcant at p < .05.

The results for the unattractive source model are shown in Figure 9 below and
seem mostly consistent with the results of the attractive source model. Signiﬁcant paths
were found from interactivity to social presence and from social presence to message
involvement, and message involvement again had a positive impact on message myth

alarms, path coefﬁcient = .31, P (.07 < p < .55) = .95. Of the paths to intention, the one

57

from source false alarms was signiﬁcant, path coefﬁcient = .27, P (.01 < p < .53) = .95,
and the path from message myth alarms approached signiﬁcance, path coefﬁcient = .20,
P (-.04 < p < .44) = .95 . However, as with the attractive source model, the lack of
substantial paths (only 5 of 12) indicates a failed model. The results of this model again
suggest that social presence and message involvement only affected recognition memory

through myth alarms.

Figure 9. Results for unattractive source model with recognition memory measures

 

 

 

 

 

    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

 

 

 

Source
Hits
V
Socral _ 13 Source Fls.
Presence —’——il Alarms
. . 30* Intention
Interactrvrty 30,,
I
Message "02 Message
Involvement Hits
-.12
.31,. Message Fls.
Alarms
essage Myth
Alarms

 

 

* Signiﬁcant at p < .05.

Of course, these analyses were intended to be exploratory, so some leeway may
be given on judging them based on the normal model evaluation criteria. A close
inspection of the paths in both models suggests that message involvement may have
affected myth alarms leading to intention. The reason for this, as touched on earlier, may
be that close attention to the message (i.e., message involvement) primed subjects to

recall other schematic information about blood pressure, and the combination of the facts

58

in the message and prior knowledge (even though it was false) may have led to the
increased intention. If myth alarms are considered evidence of systematic processing,
then this result would be consistent with the earlier model results suggesting that message
involvement affects intention by increasing systematic processing leading to intention. In
that sense, both recognition memory models are consistent with the earlier model

ﬁndings.

59

Chapter 5
Discussion
This study set out to determine the impact of interactive sources on social

presence, message involvement, information processing, and persuasion within the HSM
framework. Speciﬁcally, it proposed a model linking the social presence and message
involvement generated by interactive social agent technology with the type of increased
heuristic and systematic processing expected to bring about change in attitude and
behavioral intention, depending on source attractiveness. Two models were tested, one
with an attractive agent and the other with an unattractive agent. Both models failed to
meet initial tests for overall model ﬁt. However, parts of each model were supported, and

these paths warrant attention.
Interactive Media and Social Presence

Regardless of the type of source subjects viewed in this study, interactivity
increased social presence. This happened even with the very limited interactivity
manipulation used in this study. The only differences between the two interactivity
conditions were that (1) participants in the interactive condition could pick the order in
which information was presented by telling the source which section they wanted to see,
and (2) participants were asked by the source to repeat one utterance. Otherwise, the
experiences of all study participants were nearly identical. Yet, even with this token
difference in the manner in which the information was presented, participants reported a

greater sense of being with another person in the interactive condition.

This ﬁnding foreshadows the exciting potential of media technology that allows

people to interact socially. Presently, videoconferencing systems and other advanced two-

60

way video interaction technologies provide people with the ability to connect in
compelling ways that can be difﬁcult to distinguish from face-to-face encounters. In other
words, these technologies can create a strong sense of social presence. However, this
study suggests that even exchanges with simple, computer-controlled entities such as
clearly identiﬁable social agents can induce social presence. The participants in this study
felt more social presence when they could respond to social agents in rudimentary ways,
despite being told that the agents were computer controlled. Such a ﬁnding is consistent
with the work of Reeves and Nass (1996), who argue that human-like responses to
machines are a natural function of human evolution. Agents put a face on this human
tendency. Given the constant improvements in the developing area of social agent
technology, it is only a matter of time before social agents are fully automated to respond
to human beings as if they were engaging in face-to-face, human-to-human encounters.
This will allow two-way messages to be communicated at low cost by sources with vast
knowledge bases. As agents grow in their intelligence and appearance, they should instill
a stronger sense of social presence and foster more information processing affecting

subsequent outcomes of exposure.
The Effect of Social Presence on Heuristic Processing

As expected, social presence increased positive source thoughts in response to an
attractive source, suggesting that more attention was called to the source and her positive
attributes when subjects felt present. This would support the HSM-based prediction that
social presence leads to more heuristic processing (Skalski, 2004). Contrary to
expectations, however, social presence decreased negative source thoughts about an

unattractive source. The revised models with all paths show that the paths from social

61

presence to heuristic processing were consistent for both the attractive and unattractive
source. In both cases, social presence increased positive source thoughts and decreased

negative source thoughts.

One possible explanation for the unexpected unattractive source ﬁnding is that
social presence increases the affect a person has toward a source, even if the source is a
computer agent. If a source seems real but happens to be unattractive, for example,
individuals who feel present with that source might be less critical of the source’s
appearance, leading them to think less negative thoughts. If this is the case, it suggests
that physical appearance might be less important in the new media environment (with its
high potential for inducing social presence) than it was with traditional media. This does
not mean, however, that appearance of virtual characters has no inﬂuence on social
presence. Recent work by Nowak and Biocca (2003) argues that the anthropomorphism
of a virtual human’s appearance has a considerable inﬂuence on level of social presence,
with less anthropomorphic characters (perhaps surprisingly) increasing social presence.
Given the imaginative range of possibilities for physical appearance in virtual worlds,
from stick ﬁgures to realistically unattractive humans to more fanciful characters such as
fantasy creatures, research on this variable would advance our understanding of the role
played by physical appearance in social interactions through interactive media such as
VR and online video games. This research will not only help in the design of these

technologies but can inform related theory.
Message Involvement and Systematic Processing

Both models also produced a signiﬁcant path from message involvement to

positive message thoughts. This intuitive ﬁnding highlights how careful attention (i.e.,

62

message involvement) to persuasive message arguments can increase the salience of
desirable message-related thoughts (i.e., systematic processing). Systematic processing,
as the attenuation hypothesis of the HSM states, can curtail or overwrite the impact of
heuristic cues when people are motivated to process. This type of processing, therefore,
should have the strongest impact on attitude and intention. Systematic processing has
been shown to produce greater attitude persistence and attitude-behavior consistency than
heuristic processing (Chaiken et al., 1989), which supports the idea that systematic
processing has a more powerful inﬂuence than heuristic processing, and one that we
might expect to extend to behavioral intention. Given the importance of systematic
processing to persuasive outcomes, this result points to the value of getting subjects to
focus on and carefully attend to persuasive message arguments, something the revised

models suggest can happen in response to social presence.
The Eﬂect of Systematic Processing on Behavioral Intention

Finally, and most important from a persuasion standpoint, the model results show
that systematic processing in the form of positive message thoughts had a positive impact
on behavioral intent. Given that message thoughts are directly relevant to the attitudinal
and/or behavioral outcomes of interest, the importance of such thoughts to persuasion
makes intuitive sense; moreover, empirical evidence corroborates the importance of this
type of processing (e.g., Skalski, 2004). It has been shown to lead to both attitude (in
Skalski 2004) and behavioral intent (in the present study). From a persuasion perspective,
these ﬁnding direct attention toward the need for variables that increase positive message
thoughts. One such variable, message involvement, has already been touched on and

shown in the revised models to emanate from social presence.

63

Implications of the Revised Models

The revised models in this study pull together the separate links discussed above
and, with some minor modiﬁcations of the original models, provide a complete chain of
relationships from interactivity to behavioral intent. The paths in the revised models are
strikingly consistent, with the only difference being the signiﬁcant inverse relationship
between negative source thoughts and positive message thoughts in the unattractive
source model. While this ﬁnding might seem a bit strange, it is consistent with the earlier
claim that social presence may remove the possibly detrimental effect of unattractiveness
in response to a source. As these negative thoughts decrease, positive message thoughts

increase, as the unique path in the unattractive source models shows.

Taken together, the revised models support the idea that interactivity in new
media can alter the way information is processed and shape persuasive effect. In both
cases, when subjects were exposed to an interactive message they experienced an
increased social presence which heightened attention to the message arguments (i.e.,
message involvement). This involvement, in turn, increased systematic processing, which
had a positive effect on behavioral intention. While other studies have found that
interactive media increase behavioral intention (e.g., Kinzie, Schorling, & Siegel, 1993;
Van Tassel, 1988), such studies have typically been different than the present one in at
least two important respects. The present study (1) offers insight into the cognitive
mechanisms underlying these effects, and (2) uses an interactive source instead of

another form of interactive media (e. g., hyperlinked web pages).

First, the models in this study are among the ﬁrst to offer insight into the

underlying cognitive mechanisms that explain why interactive media can increase

64

persuasion. Most extant research in this area has made little (if any) attempt at explaining
and providing empirical support for how interactive media affect users. Prior studies
merely reveal that interactivity increases desirable outcomes. This study not only shows
that interactive media can affect persuasion, but it also provides empirical evidence for a
process model linking interactivity to a set of mediating cognitive processes that
inﬂuence persuasion. In doing so, the research increases the potential for the prediction
and control of desirable persuasive outcomes, since key variables in the process of

persuasion are identiﬁed.

Much of the credit for this explanatory power lies in the rooting of these models
in the HSM. The HSM explicates how cognitive and motivational factors can affect
information processing styles. Moreover, HSM literature identiﬁes several of these
determinants (Todorov, Chaiken, & Henderson, 2002). The present research adds social
presence to the mix. Although social presence has recently been linked to the HSM
(Skalski, 2004), this study is the ﬁrst to look at interactivity-induced social presence
within the HSM framework. The ﬁndings show that while social presence does not seem
to be a cognitive impairment to processing, it does increase the motivation to process
information apparent in message involvement, and the heightened involvement that
results can increase systematic processing. Thus, social presence stemming from
interactivity may be an important variable in certain mediated persuasion situations due

to its ability to focus attention on message arguments.

This calls attention to a second important difference between this and other work
looking at the persuasive effects of interactive media—the use of an interactive source as

opposed to some other form of interactivity. Although interactivity may seem like a

65

simple concept, Vorderer (2000) illustrates how recent advances in computing
technology have resulted in many, sometimes ambiguous deﬁnitions of the term. The
advent of the Internet, for example, has focused much attention on interactivity as it
applies to web sites (e.g., Sundar, Kalyanaraman, & Brown, 2003), mostly involving the
surﬁng of hyperlinked web pages. This type of interactivity is different than the more
classical, communication-oriented interactivity looked at in this study — one based on the
idea of social interaction. As a result, these ﬁndings may be limited to source-centered
technologies and not easily generalized to other interactive media. The sizeable effect of
interactivity on social presence and lack of effect of interactivity on message involvement
lends support to this idea. It may be that interactive technology such as hyperlinked web
pages has a direct effect on message involvement while, as the present ﬁndings suggest,

source-centered interactivity has a direct effect on social presence.

Concerns about the applicability of these ﬁndings to all interactive media
illustrate the need for a broader conceptualization or perhaps even “theory” of
interactivity that attempts to unify the different manifestations of the concept. At the
same time, the ﬁndings in this study refocus our attention on traditional understandings of
interactivity in terms of social interaction and show it potential importance in new media.
In one sense, the technology used in this study provides a new arena in which to examine
the transactional nature of communication. In a broader sense, these ﬁndings point to the
road ahead. Since there are many emerging interactive technologies speciﬁcally designed
to create social presence, the ﬁndings of the present study have considerable value in

these new commercial areas and beyond.

66

 

Limitations

As with most studies, several issues with the procedures in this study raise reason
for concern. These include issues related to the message, measures, sample, and the

manipulation of interactivity.

Message. The blood pressure issue used in this study was one that subjects had a
very favorable attitude toward before taking part. The mean pre-test attitude toward blood
pressure was 5.21 (SD = .99), leaving little room for improvement and forcing a reliance
on behavioral intention as the outcome measure. Future research investigating the
processes under investigation here would beneﬁt from a message topic toward which
subjects are not as favorably inclined. While attitudes toward health topics tend to be
very favorable, messages could be created in another domain such as politics or social
issues. This would allow for a better test of the relationships under examination.
However, it should be noted that the strong pre-attitude toward blood pressure is
something that would work against obtaining any ﬁndings. Even with this limitation, the
model tested was still able to account for change in behavioral intention, and this
suggests the potential for even stronger effects in situations where there is greater room

for change in outcome variables.

An associated message-related limitation concerns the applicability of these
results to other persuasion messages. Will these results generalize to other health
messages, or even other types of inﬂuence attempts through interactive media, such as
commercial advertising? Since the components of the model in this study are not speciﬁc
to blood pressure, there is no reason to expect that this model would not work with other

persuasive messages, perhaps to an even greater extent. Many health issues, for example,

67

also have a strong pre-attitude (e.g., against smoking or for exercise) but might lack the
same high levels of behavioral intent that likely limited the inﬂuence of exposure in this
study. Similarly, it is not difﬁcult to think of commercial product use situations where
behavioral intent has no ceiling effect that would limit the inﬂuence of these processes in

advertising and other persuasive settings.

Measures. In addition to the fact that strong pre-attitudes toward blood pressure
constrain the potential to observe the inﬂuence of interactivity on related outcomes, the
dichotomous measure of behavioral intention further restricts this study’s ability to
observe covariation between the interactivity manipulation and behavioral intention.
Future research in this domain would beneﬁt from continuous measures of behavioral
intention that can provide more sensitive indicators of this association. For example,
participants could be asked to sign up for a series of behavioral intention steps, such as
signing up for a test on campus, signing up for a test and follow-up visit at the health
center, signing up for multiple tests, etc. With greater room for change in outcome

variables, stronger effects should be observed.

Sample. The sample used in this study may also be a limitation. A more
representative general population sample of subjects would have been preferred for this
research, to maximize external validity. However, there is no immediately apparent
reason to expect that the largely unconscious cognitive processes under investigation in
this study should differ for other populations. Moreover, a student sample was deemed
sufﬁcient for this study’s goal of testing the theoretical model under consideration given
its predominant concern with internal validity. Nevertheless, the low average age of the

participants in this study (M = 21.56, SD = 1.75) and lack of inexperienced computer

68

users (no subjects reported not using a computer) does raise questions speciﬁcally related
to the effect of age and experience on the technology and health issues examined in this
study.

A less experienced sample of older respondents at greater risk for high blood
pressure might have had difﬁculty with a message communicated through interactive
computer technology. They might, for example, have had a problem with accepting an
animated source, or difﬁculty processing the simultaneous communication of verbal and
visual stimuli while at the same time having to interact. If so, then the effectiveness of
this type of message would be limited to experienced users. The young sample in this
study beneﬁts from having experience with the type of technology used in the research
and would be less likely to have difﬁculty with it. Given that computer users are the most
likely recipients of these types of computerized messages, this limitation would not be a
common one. Moreover, we should expect inexperience with computers to become less
common over time. Still, the sample calls attention to the need for matching interactive
messages to appropriate audiences.

Age might also have affected the outcomes of this study by limiting the
receptiveness of targets to the blood pressure message. This type of health message
would more likely be attended to by individuals at risk, such as people over age 35, than
those not as risk, such as the younger people used in this study. Of course, this concern
would likely work against obtaining ﬁndings. The discovery that intention was still
positively affected by the message suggests that this limitation was minimal. Almost half

of the participants (42%) signed up for a blood pressure test.

69

Another sample limitation is the lack of males in the study. A more even split of
females and males would be more desirable, but instead 66% of participants were female
and only 34% male. This raises questions about whether these results were driven by
gender or some type of related interaction, such as gender with the femaleness of the
source. A scan of the correlations between gender and other variables in this study
suggested that this was not the case. Gender did not, for example, relate to perceived
attractiveness. Therefore, this limitation did not appear to be a problem.

A ﬁnal sample-related limitation is the sample size. Thought the cell sizes in this
study were fairly large for an experimental study, the split into separate attractive and
unattractive groups halved the base cell sizes and resulted in the testing of models with
N3 of 62 and 63. This reduction in the power behind the model analyses may account for
some of the weak paths, such as the paths in the unattractive source model from message
involvement to positive message thoughts and from positive message thoughts to
behavioral intent. With more power, greater conﬁdence in the substantial nature of these
relationships would be possible.

Manipulations. The last limitations have to do with the weak manipulations of
interactivity and source attractiveness in this study. In the case of interactivity, this
manipulation led to low perceived interactivity and social presence values. In the
interactive condition, the mean perceived interactivity was close to the midpoint of the
scale (M = 7.67, SD = 3.37), and the mean for social presence was below the midpoint of
the scale (M = 3.33, SD = 1.38). While somewhat disappointing, these means were still
signiﬁcantly greater than the means in the unattractive condition. Furthermore, they were

not entirely unexpected due to the subtlety of the interactivity manipulation discussed

70

earlier. To maximize control in this study and keep the conditions as close as possible,
there were only slight differences between the two interactivity groups, limiting their
potential for interaction.

A related limitation stems from the decision to use dichotomous manipulations of
interactivity and source attractiveness. Both variables had only two levels, and this may
have attenuated their relationships with other variables. Though the decision was made to
keep the manipulations in this study dichotomous based on technological and logistic
constraints, a greater range of manipulations would have been preferable, such as low,
medium, and high interactivity. Having three or more levels of interactivity could have
made the crucial relationship between interactivity and social presence stronger.

More potent results might be expected from technology allowing for greater
interaction, something that will become more common in the near future as agent
technology becomes more sophisticated. Even currently, agents have an impressive array
of capabilities. The agent in this study, for example, could have been programmed to
have more back and forth dialogue with subjects. Instead of just responding to spoken
commands about what section of the message to go to, the agent could have asked
subjects if they had any questions and responded with phrases like “I’m not sure what
you want me to do” if it did not know the answer. Future agents may be programmed
with enough information to answer just about any question a person would have,
particularly about a speciﬁc topic area such as blood pressure. In doing so, they can
approach the social interaction capability of an expert such as a doctor. As technologies

allowing for social interaction become more advanced and generate more social presence,

71

their effects on information processing modes and resultant persuasion should be even
greater.
Conclusion

The present study was confronted with a set of limitations capable of obscuring
the associations under investigation. The use of dichotomous and weak manipulations,
the dichotomous measure of the principal outcome variable, the use of a message with a
restricted range of attitude, the use of individuals with limited issue involvement, and a
small sample size are all factors that should limit the potential to observe any signiﬁcant
results. Yet despite these limitations, the study still produced ﬁndings consistent with
predictions. Obtaining these results speaks volumes about the potential for interactive
technology to affect persuasion.

The present theoretical extension and ﬁndings, rooted in the HSM, suggest that
increased social presence resulting from an interactive source can affect information
processing and resultant behavioral intentions related to health issues. In doing, the study
shows the exciting potential for these types of technologies from the standpoint of
persuasion. The possibility for such outcomes should increase dramatically as new media
continue to emphasize features that instill increasing amounts of social presence. Today’s
computer screen social agents may someday blossom into 3-D holographic
representations like the one depicted in the recent science ﬁction movie I, Robot. The
agent in I, Robot looked and talked like a person (actor James Cromwell) yet was a
projection controlled entirely by a computer that could simulate basic social-interaction
intelligence. As technologies continue to increase in vividness and interactive capability,

the role of social presence will become more important from both a theoretical and

72

practical, practitioner standpoint. Social presence has been shown to focus more attention
on a persuasive message — something that could become even more valuable in an
increasingly cluttered media landscape. To remain on the “cutting edge” and actively
shape technology development, a clear understanding of variables such as social presence

is vital. This study advances our understanding of both its determinants and inﬂuences.

73

APPENDICIES

74

Appendix A: Script

 

Message Effectiveness Study Script

 

Once subject is seated at table, RESEARCHER says:

Hi, are you ? (verify that correct subject has arrived from sign-up sheet)

 

Thank you for coming. My name is and I’ll be running the study today.
Before we begin, I need you to read and sign this consent form and then fill out this
brief survey.

Researcher hands subject consent form and pretest. Once subject has read and signed
consent form and ﬁlled out survey, both are collected. Researcher then says:

Thank you. Now, I would like to tell you a bit about today’s study, which is being
conducted by members of the Communication Department. The study deals with
“people’s reactions to health messages.” The Communication Department is
interested in developing messages to communicate information about health
practices through the use of new media technologies. Today we would like help in
the development of these by giving us your reaction to a message on one of a variety
of health topics. Your reactions will be kept confidential, so we would appreciate
your frank and honest opinions. If you have any questions at this time, please let me
know.

Researcher answers questions (if any) and then continues:

I randomly selected a message before you came in. You will hear a computer
generated health message. Your message is about blood pressure. The information
in this message will be communicated by computerized character named Cardia.

In interactive condition:

There are five sections of Cardia’s message, and we would like you to hear all of
them. However, the order in which the information is presented is up to you. On
the screen in front of you, five categories will appear on the right side, and Cardia
will appear on the left. Cardia has been programmed to communicate the
information in each category based on your spoken commands. When she begins
speaking, she will ask you what category you would like to learn about. You can
then pick one of the five categories for Cardia to talk about by talking to her. When
she finishes each section, she will ask you to select another category and continue
until you have heard about all five parts. When Cardia is done, I will return to see
what you thought about your experience.

75

Cardia will begin in a moment. Before she begins, we need you to put on these
headphones so you can hear what she says. Remember, you can speak to Cardia
once she begins and should tell her what information you want to find out about.
In non-interactive condition:

There are five sections of Cardia’s message, and we would like you to hear all of
them. On the screen in front of you, the five categories will appear on the right side,
and Cardia will appear on the left. Cardia will discuss the information on the right
side of the screen point by point. When she is done, I will return to see what you
thought about your experience.

Cardia will begin in a moment. Before she begins, we need you to put on these
headphones so you can hear what she says.

Researcher helps subject put on headphones, walks behind screen to pretend to start
program, and leaves room. Controller then starts program and oversees interaction.
Once experience is over, the researcher returns with THOUGHT LISTING
QUESTIONAIRE and says:

We are interested in what you were thinking about during the experience. It could
be thoughts about the message, or the speaker, or anything else. Please write down
your thoughts, and I will be back in three minutes to collect them.

Researcher leaves and comes back in three minutes to collect though listing task. The
researcher then hands the subject the TRUE-FALSE TEST and says:

Now, we would like to ask what you remember about the experience you just had.
Here are a set of statements. We would like you to tell us if they are true or false
based on what you heard. Please answer them to the best of your ability.

Researcher leaves and comes back once subject is done with true-false test. The
researcher then hands the subject the presence, attitude, and intention questionnaires
(three total) and says:

We would now like you to fill out three questionnaires dealing with your thoughts about
the experience, the message, and the speaker. When you are done with these, please wait
here and I will return with some ﬁnal questions.

Researcher leaves and comes back once subject is done. The researcher then hands the
subject the blood pressure checkups questionnaire and the blood pressure test sign-up
sheet and says:

You’re almost done. Before you go, we would like you to fill out this questionnaire about

blood pressure checkups and, if you are interested, we would like to offer you the
opportunity to sign up for a blood pressure checkup.

76

Researcher leaves and comes back once subject is done. The researcher then hands the
subject two sets of opinions of the speaker questions and says:

Before you go, we would like you do one more thing. I am going to play clips of two
different versions of Cardia for you. Then I would like you to evaluate each of the
two clips by answering a final few questions.

Researcher leaves and plays clip of both versions for subject. The researcher then returns
and says:

That’s all for today. Before you go, we would like to tell you about the purpose of this
study.

This study was designed to see how people respond to computer technologies featuring
different message sources. More specifically, we want to find out how the interactivity
provided by computer technology affects responses to persuasive messages delivered by
appealing and unappealing computerized sources. The speech evaluation task was
merely a cover story created for our study. The real purpose of the study was to see how
persuaded you would be by certain types of technologies and sources.

You were assigned to one of several different groups in the study. Two things varied from
one group to another. First, in different cases, the persuasive message was delivered by
either an interactive source or an unresponsive source. Second, in different cases the
speaker was presented as either an appealing or an unappealing character.

To properly test these inﬂuences, we could not completely disclose everything we were
doing at the start of this study. We hope this has not upset you in any way. Please accept
our apology if you are disturbed by anything that has happened.

Finally, you should know that all of your responses will be kept strictly confidential.
Your name will not be connected with our observations in any way. In fact, if you wish, I
will dispose of your responses now and they will not be included in any further part of
our study. You will receive full credit for your participation no matter what you decide.
However, since we feel that these types of persuasive situations are a normal part of
everyday life, we hope you will allow us to use these observations.

If subject agrees, research continues and says:

Thanks for helping us out. Since it is important for us to keep the purpose of this
study secret, we would appreciate it if you would not talk to anybody about this
until the end of the summer, at which time the research will be completed. If you
have any questions, Paul Skalski will be happy to talk with you about them.
Otherwise, thanks again. You are free to go.

77

Appendix B: Blood Pressure Message

[INTRODUCTION]
1. Hi! My name is Cardia! I’m here to tell you all about high blood pressure.
2. Tell me which section you would like to learn about ﬁrst.

[EFFECTS]
{SUMMARY} This section explains what blood pressure is, how it is created, and who it
affects.
1. Blood pressure is the amount of blood pumped by your heart. It is measured with
two numbers: the systolic and diastolic blood pressure.
2. Blood pressure is created when your coronary artery carries oxygen in your blood
away from the heart and lungs and to other parts of your body.
3. Your blood pressure is too high if your systolic blood pressure reading is higher
than 140 or your diastolic blood pressure reading is higher than 90.
4. 50 million people in America have high blood pressure, and 3 out of 10 of them
do not even know that they have it.

[CONSEQUENCES]
{SUMMARY} This sections tells you some of the more serious medical outcomes if high
blood pressure goes untreated.

1. High blood pressure can cause you to have a stroke. A stroke occurs when a blood
vessel that carries oxygen and nutrients to the brain is either blocked by a blood
clot or if it bursts.

2. High blood pressure leads to coronary heart disease, which is the most common
form of heart disease and the number one killer in America.

3. High blood pressure can also cause congestive heart failure, which is the buildup
of ﬂuid in your heart that affects the heart’s pumping action.

4. High blood pressure can also cause one or both of your kidneys to fail. Your
kidneys remove wastes from your body, and kidney failure would mean you have
to rely on a dialysis machine.

[RISKS]
{SUMMARY} This section tells you the factors that puts you at risk or increases your
risks of getting high blood pressure.

1. Obesity is a risk factor you can control. The extra weight not only puts you at risk
for high blood pressure, but could also lead to heart attacks, strokes, diabetes, and
even cancer.

2. Sodium chloride or table salt can increase your blood pressure drastically. You
should try to control your salt intake to no more than 3 teaspoonfuls a day.

3. Your alcohol intake is something you may want to control if you are at risk for
high blood pressure. Women should not have more than 1 drink a day and men,
not more than 2.

4. An inactive lifestyle can lead to poor health and obesity, which eventually leads to
high blood pressure. Getting more exercise reduces your risks.

5. If you constantly experience stress over long periods of time, you are at greater
risk of increasing their blood pressure levels.

78

6. A tendency to have high blood pressure runs in families. If your parents or other
close blood relatives have it, you’re more likely to develop it.

7. In general, the older you get, the greater your chance of developing high blood
pressure. It occurs most often in people over age 35. Men seem to develop it most
often between age 35 and 50. Women are more likely to develop it after
menopause.

[DIET]
{Summary} This section tells you how you can control your high blood pressure by
watching what you eat.

1. Deep fried foods and greasy foods such as fried chicken and French fries are high
in fat and cholesterol. Cheese and other dairy products may have a very high fat
content and should be avoided as much as possible.

2. If you use salt when you cook, you should not be eating more than three
teaspoonfuls of salt per person per meal portion.

3. If you drink alcohol frequently and regularly, just reducing your alcohol intake by
one glass can signiﬁcantly reduce your chance of getting high blood pressure.

4. Fruits and vegetables are high in potassium and low in sodium. In addition to
discouraging high blood pressure, it also helps prevent a potassium deﬁciency.

[OTHER]
{Summary} This section tells you how you can control your high blood pressure through
other ways other than diet.
1. Even if you are in good health, you should get a doctor to check your blood
pressure every two years. This is because blood pressure can rise unexpectedly.
2. Regular physical activity keeps your body healthy and helps reduce the risks or
effects of high blood pressure in the long run.
3. Your chances of getting high blood pressure and other cardiovascular diseases are
greatly reduced the very moment you quit smoking.
4. The easiest way to control your blood pressure levels is through prescribed
medication from your doctor. This method is usually used by patients with
extremely high blood pressure.

[TRANSITIONS]
I . Where would you like to go next?
2. You ’re now done with this section. Where to next?
3. You have already visited that section. Are you sure you want to view it again?
4. Hope you learnt something from that! Where to next?

[CONVERSATIONAL REPAIRS]
I. I didn ’t hear what you just said, could you please repeat that?

2. What was that again ?
3. I ’m not sure what you want me to do.

79

[INACTIVITY]
1. Please go on.
2. I’m waiting...
3. What next?

[POSITIVE/AFFECTIVE STATEMENTS]
1. Okay, we do it your way.
2. We ’re on our way!
3. Nice thinking!

[CLOSING]
1. Well, that’s all the time we have! Thanks for coming!
2. It was nice talking to you! Have a nice day!

Note: Interactive condition statements appear in italics

80

Appendix C: Cardia in Attractive and Unattractive Form

 

Attractive

 

Unattractive

81

Appendix D: Pretest Attitude, Demographic, and Computer Use Measures
Following are some statements about health issues. Please indicate the extent to which
you agree or disagree with each statement on a 1 to 7 scale, with “1” indicating
STRONGLY DISAGREE and “7” indicating STRONGLY AGREE. You may circle any
one number between 1 and 7 to indicate how much you agree with each statement.

1 It is important to have blood pressure checked regularly.
Notatalll 2 3 4 5 6 7 Verymuch
2. Hypertension is a serious problem.

Notatalll 2 3 4 5 6 7Verymuch

These last questions are about you. Again, all of your responses will be kept conﬁdential,
so please answer as accurately and honestly as possible.

How old are you (in years)?
Please indicate your gender. _Male Female

What is your race?

Asian Paciﬁc Islander
African American White
Hispanic Other

How much time do you spend using a computer (including surﬁng the Internet) in a typical
day?

_0 hours ___3 to 4 hours
_0 to 1/2 hour _4 to 5 hours
_1/2 to 1 hour __5 to 6 hours
________1 to 2 hours __6 to 7 hours
___2 to 3 hours _More than 7 hours

82

Appendix E: Thought Listing Instructions

We are interested in what you were thinking about during and immediately following
exposure to the message.

The NEXT PAGE contains the form we have prepared for you to use to record your
thoughts and ideas. Please state your thoughts and ideas as brieﬂy as possible... a phrase
is sufﬁcient. IGNORE SPELLING GRAMMAR AND PUNCTUATION.

 

You will have 3 minutes to write your thoughts. We have deliberately provided more
space than we think most people will need to insure that everyone would have plenty of
room to write their ideas. So don’t worry if you don’t ﬁll every space. Just write down
your thoughts. Please be completely honest and list all of the thoughts that you had.

83

Appendix F: Recognition Memory Items (with answers in parentheses)

Please indicate whether the following statements are “True” or “False” based on the
message you just heard by circling “T” for True or “F” for False.

l.

2.

10.

11.

12.

13.

14.

15.

16.

17.

High blood pressure is normal. (false; common myth)

Blood pressure is a measure of how fast blood is pumped by your heart. (false; it
is the amount of blood pumped)

Three out of 10 people with high blood pressure do not even know they have it.
(true)

High blood pressure can cause coronary heart disease, the number one killer in
America. (true)

In a blood pressure reading, the bottom number (systolic blood pressure) is more
important than the top (diastolic blood pressure). (false; common myth)

Systolic pressure of 100 + your age number is normal. (false; common myth)

High blood pressure develops most often in people over age 55. (false; it deve10ps
in people over age 35)

Sodium chloride consumption increases the risk of high blood pressure. (true)
No more than three teaspoons of salt should be eaten per meal portion. (true)
Dairy products discourage high blood pressure. (false; they contribute to it)
Fruits and vegetables discourage high blood pressure. (true)

A tendency to have high blood pressure runs in families. (true)

People in good health should have their blood pressure checked a least twice a
year. (false; every two years)

The easiest way to control blood pressure levels is through doctor-proscribed
medication. (true)

The speaker moved her head when she talked. (true)
The speaker had blue eyes. (false)

The speaker’s hair was straight. (true)

84

18. The speaker wore glasses. (false)

19. The speaker had braces. (false)

20. The speaker often gestured with her right hand. (false)

21. The speaker’s hair was brown. (true)

22. The speaker wore earrings. (false)

23. The speaker lifted her eyebrows when she talked. (true)

24. The speaker had strands of hair hanging over one eye. (true)
25. The speaker was female. (true)

26. The wall behind the speaker was a pale green color. (false)
27. The speaker was sitting behind a desk. (false)

28. The speaker blinked her eyes. (true)

85

Appendix G: Social Presence Items (with factor loadings)

Please consider the following statements about your feelings during the speech and
indicate the extent to which you AGREE or DISAGREE with each using a 7-point scale,
with “1” indicating NOT AT ALL or STRONGLY DISAGREE and “7” indicating
“VERY MUCH or STRONGLY AGREE. The point on the scale numbered 4 is the
midpoint. This would indicate that you don’t agree or disagree with the statement You
may circle any ONE number between 1 and 7 to indicate the extent of your agreement.

Factor

Lain. LOACLing

1. To what extent did this feel like you were with an actual person? .78
Notatalll 2 3 4 5 6 7 Verymuch

2. To what extent was this like a face-to-face encounter? .83
Notatalll 2 3 4 5 6 7 Verymuch

3. To what extent was this like you were in the same room as Cardia? .85
Notatalll 2 3 4 5 6 7 Verymuch

4. To what extent did Cardia seem “real”? .84
Notatalll 2 3 4 5 6 7 Very much

5. To what extent did you feel you were able to assess Cardia’s reactions
to what you said? .56
Notatalll 2 3 4 5 6 7 Very much

6. How much did your feel like you were “with” Cardia? .90

Notatalll 2 3 4 5 6 7Verymuch

86

Appendix H: Message Involvement Items (with factor loadings)

Please consider the following statements about the message and indicate the extent to
which you AGREE or DISAGREE with each using a 7-point scale, with “1” indicating
STRONGLY DISAGREE and “7” indicating STRONGLY AGREE. The point on the
scale numbered 4 is the midpoint. This would indicate that you don’t agree or disagree
with the statement. You may circle any ONE number between 1 and 7 to indicate the
extent of your agreement.

Item

1.

The message was engaging.

Strongly Disagree 1

2

3

4

I did not concentrate on the message.

Strongly Disagree 1

I carefully examined the message.

Strongly Disagree 1

2

2

I focused on the message.

Strongly Disagree 1

The message was involving.

Strongly Disagree 1

2

2

3

3

3

3

4

4

4

4

I did not pay attention to the message.

Strongly Disagree 1

2

3

4

87

7 Strongly Agree

7 Strongly Agree

7 Strongly Agree

7 Strongly Agree

7 Strongly Agree

7 Strongly Agree

Factor
Loading

.61

.84

.80

.90

.77

.73

Appendix I: Attitude toward Blood Pressure Items (with factor loadings)

Please consider the following statements about the message and indicate the extent to
which you AGREE or DISAGREE with each using a 7-point scale, with “1” indicating
STRONGLY DISAGREE and “7” indicating STRONGLY AGREE. The point on the
scale numbered 4 is the midpoint. This would indicate that you don’t agree or disagree
with the statement. You may circle any ONE number between 1 and 7 to indicate the
extent of your agreement.

Item

1.

It is important to have blood pressure checked regularly.

Strongly Disagree 1 2 3 4 5 6 7 Strongly Agree
Blood pressure is not very important compared to other health issues.
Strongly Disagree 1 2 3 4 5 6 7 Strongly Agree
People should pay more attention to their blood pressure.

Strongly Disagree 1 2 3 4 5 6 7 Strongly Agree
Blood pressure is an issue people should be concerned about.
Strongly Disagree 1 2 3 4 5 6 7 Strongly Agree

Blood pressure is one of the most important health issues facing people
today.

Strongly Disagree 1 2 3 4 5 6 7 Strongly Agree
It is not important to have blood pressure checked regularly.

Strongly Disagree] 2 3 4 5 6 7 Strongly Agree

* item dropped due to unacceptable loading

88

Factor
Low

.73

.43*

.81

.90

.52

.56

Appendix J: Attitude toward the Communicator Items (with factor loadings")

The following statements ask you to consider CARDIA, the SPEAKER of the message.
Please read them and indicate the extent to which you agree or disagree with each using a
15-point scale, with “1” indicating STRONGLY DISAGREE and “15” indicating
STRONGLY AGREE. The point on the scale numbered 8 is the midpoint. This would
indicate that you don’t agree or disagree with the statement. You may circle any one
number between 1 and 15 to indicate the extent of your agreement.

Factor
Item Loading
The communicator was.
1. Attractive .88

Strongly disagreel 2 3 4 5 6 7 8 9 10 ll 12 13 14 15 Strongly agree
2. Knowledgeable

Strongly disagreel 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Strongly agree
3. Intelligent

Strongly disagreel 2 3 4 5 6 7 8 9 10 ll 12 13 14 15 Strongly agree
4. Competent

Strongly disagreel 2 3 4 5 6 7 8 9 10 ll 12 13 14 15 Strongly agree
5. Nice Looking .93
Strongly disagreel 2 3 4 5 6 7 8 9 10 ll 12 13 14 15 Strongly agree
6. Pleasing .81
Strongly disagreel 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Strongly agree
7. Friendly

Strongly disagreel 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Strongly agree
8. Warm

Strongly disagreel 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Strongly agree

89

Factor

Item Loading
9. Appealing .89

Strongly disagreel 2 3 4 5 6 7 8 9 10 ll 12 13 14 15 Strongly agree
10.Likable

Strongly disagreel 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Strongly agree

* loadings shown only for items comprising source attractiveness scale

90

Appendix K: Perceived Interactivity Items (with factor loadings)

Finally, please consider the following additional items about your experience using
the same scale.

Factor
Item Loading
1. I felt like Cardia would respond to me during the message. .83

Strongly disagreel 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Strongly agree
2. I felt as if I could control how the message was delivered to me. .82
Strongly disagreel 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Strongly agree
3. I felt like I could respond to Cardia. .87
Strongly disagreel 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Strongly agree
4. I saw myself as a message sender and receiver instead of just a receiver. .87
Strongly disagreel 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Strongly agree
5. I felt as though Cardia and I were interacting with each other. .83

Strongly disagreel 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Strongly agree

91

Appendix L: Intention Measure

In an effort to improve blood pressure awareness, we have arranged for
participants in this study to receive a free blood pressure checkup. If you would like
to have your blood pressure checked, please sign below and write down your email
address and phone number. We will call or email you next week with information
about when and where to come for the test. If you do not wish to have your blood
pressure checked, do not sign the line below. There is no penalty for choosing not to
have your blood pressure checked.

 

Signature Date

 

Name (please print clearly) Phone Number Email Address

92

REFERENCES

Andrews, J. C., Durvasula, S., & Akhter, S. H. (1990). A framework for conceptualizing
and measuring the involvement construct in advertising research. Journal of
Advertising, 19(4), 27-40.

Biocca, F. (1997). The cyborg’s dilemma: Progressive embodiment in virtual
environments. Journal of Computer Mediated Communication, 3(2). Retrieved May
29, 2004, from http://wwwascusc.org/icmc/vol3/issue2/biocca2.html

Biocca, F., Harms, C., & Gregg, J. (2001). The networked minds measure of social
presence: Pilot test of the factor structure and concurrent validity. Unpublished
M.I.N.D. Lab Manuscript, Michigan State University.

Biocca, F., Harms, C., & Burgoon, J. K. (2003). Toward a more robust theory and
measure of social presence: Review and suggested criteria. Presence:
Teleoperators and Virtual Environments, 12(5), 456-480.

Brody, E. W. (1990). Communication tomorrow: New audiences, new technologies, new
media. New York: Praeger.

Burgoon, J. K. & Hoobler, G. D. (2002). Nonverbal signals. In M. L. Knapp and J. A.
Daly (Eds), Handbook of Interpersonal Communication (pp. 240-299). Thousand
Oaks, CA: Sage.

CaciOppo, J.T., & Petty, RE. (1981). Social psychological procedures for cognitive
response assessment: The thought listing technique. In T. Merluzzi, C. Glass, & M.
Genest (Eds) Cognitive Assessment. New York: Guilford Press.

Chaiken, S. & Eagly, A. H. (1983). Communication modality as a determinant of
persuasion: The role of communicator salience. Journal of Personality and Social
Psychology, 45(2), 241-256.

Chaiken, S. Liberman, A. & Eagly, A. H. (1989). Heuristic and systematic information
processing within and beyond the persuasion context. In J. S. Uleman & J. A.
Bargh (Eds), Unintended thought. New York: The Guilford Press.

Chaiken, S. & Maheswaran, D. (1994). Heuristic processing can bias systematic
processing: Effects of source credibility, argument ambiguity, and task
importance on attitude judgment. Journal of Personality and Social Psychology,
66(3), 460-473.

Champness, B. G. (1973). The assessment of the user reactions to confravision: Analysis
and conclusions. Unpublished Communications Studies Group paper no.
E/73250/CH.

93

Chen, S. & Chaiken, S. (1999). The Heuristic-systematic model in its broader context. In

S. Chaiken & Y. Trope (Eds), Dual process theories in social psychology (pp.
73-96). New York, NY: Guilford Press.

Durlak, J. T. (1987). A typology of interactive media. In M. L. McLaughlin (Ed.),
Communication yearbook 10 (pp. 743-757). Beverly Hills, CA: Sage.

Dupagne, M. (2002). High deﬁnition television. In Lin, C. A. & Atkin, D. J. (Eds),
Communication technology and society: Audience adoption and uses (pp. 279-
305). Cresskill, NJ: Hampton Press.

Eagly, A. H. & Chaiken, S. (1993). The psychology of attitudes. New York: Harcourt,
Brace, J ovanovich.

Fogg, B. J. (2003). Persuasive technology: Using computers to change what we think and
do. San Francisco: Morgan Kaufmann.

Grigorovici, D. (2003). Persuasive effects of presence in immersive virtual environments.
In G. Riva, F. Davide, & W. Usselsteijn (Eds), Being there: Concepts, eﬂects and
measurement of presence in synthetic environments. Amsterdam: 103 Press.

Greenwald, A. G. & Banaji, M. R. (1995). Implicit social cognition: Attitudes, self
esteem, and stereotypes. Journal of Personality and Social Psychology, 102, 4-27.

Heeter, C. (1989). Implications of new interactive technologies for conceptualizing
communication. In J. L. Salvaggio & J. Bryant (Eds), Media use in the
information age: Emerging patterns of adoption and consumer use (pp. 217-235).
Hillsdale, NJ: LEA.

Holbert, R. L., & Stephenson, M. T. (2002). Structural equation modeling in the
communication sciences, 1995-2000. Human Communication Research, 28(4),
531-55 1.

Huang, H. (1999). The persuasion, memory and social presence effects of believable
agents in human-agent communication. Paper presented at the Third International
Cognitive Technology Conference, San Francisco, CA.

Hunter, J. E. & Gerbing, D. W. (1982). Unidimensional measurement, second order
factor analysis, and causal models. In B. M. Shaw and L. L. Cummings (Eds),
Research in organizational behavior (Vol. 4, pp. 267-320). Greenwich, CT: J AI.

Jackson, L. A., Barbatsis, G., von Eye, A., Biocca, F., Zhao, Y., & Fitzgerald, H. (2003).
IT& Society, 1(5), 141-165.

Kay, A. (1984). Computer software. Scientiﬁc American, 251(2), 53-59.

94

Kinzie, W., Schorling, J. B., & Siegel, M. (1993). Providing prenatal alcohol education to
low income women with interactive multimedia. Patient Education and
Counseling, 21(1), 51-60.

Klein, L. (2001). Creating virtual experiences in computer-mediated environments. Paper
presented at the 2001 proceedings of the Experimental E-commerce Conference,
M.I.N.D. Lab, East Lansing, MI.

Koprowski, G. J. (2003, 23 July). Socially intelligent software: Agents go mainstream.
TechNewsWorld. Retrieved August 23, 2003, from
http://www.technewsworld.com/perl/story/41 172.html

Lee, K. M. (2004). Presence, explicated. Communication Theory, 14(1), 27-50.

Li, H., Daugherty, T., & Biocca, F. (2002). Impact of 3-D advertising on product
knowledge, brand attitude, and purchase intention: The mediation role of
presence. Journal of Advertising, 31(3).

Lieberman, H. & Selker, T. (2003). Agents for the user interface. In Jeffrey Bradshaw
(Ed.), The handbook of agent technology, MIT Press.

Lombard, M. (2000). Resources for the study of presence: Presence explication. Retrieved
May 1, 2004, from http://www.temple.edu/ispr/frame explicathtm

Lombard, M. & Ditton, T. (1997). At the heart of it all: The concept of presence. Journal
of Computer Mediated Communication, 3(2). Retrieved Oct. 15, 2002 from
http://www.ascusc.org/jcmc/vol3/issue2/lombard.html *

Nowak, K. L. & Biocca, F. (2003). The effect of the agency and anthropomorphism on
users' sense of telepresence, copresence, and social presence in virtual
environments. Presence: Teleoperators and Virtual Environments, 12(5), 481-
494.

O’Keefe, DJ. (2002). Persuasion: Theory and research (2nd ed.). Thousand Oaks, CA:
Sage.

Petty, R. E., & Cacioppo, J. T. (1981). Attitudes and persuasion: Classic and
contemporary approaches. Dubuque, IA: William C. Brown.

Reeves, B. & Nass, C. (1996). The media equation. Cambridge University Press.

Rice, R. (1993). Media appropriateness: Using social presence theory to compare traditional
and new organizational media. Human Communication Research, 19, 451-484.

95

Shapiro, MA. (1994). Think-aloud and thought-list procedures in investigating mental

proceses. In A. Lang, (ed.) Measuring psychological responses to media messages.
Hillsdale, NJ :Erlbaum (pp 1- 14).

Shavitt, S. & Brock, T. C. (1994). Persuasion: Psychological insights and perspectives.
Boston: Allyn and Bacon.

Short, J ., Williams, E., & Christie, B. (1976). The social psychology of
telecommunications. London: John Wiley & Sons, Ltd.

Skalski, P. (2004). Social presence and information processing: An application of the
heuristic-systematic model of persuasion. Unpublished manuscript.

Steuer, J. (1995). Deﬁning virtual reality: Dimensions determining telepresence. In F.
Biocca & M. R. Levy (Eds), Communication in the age of virtual reality (pp. 33-
56). Hillsdale, NJ: LEA.

Street, R. L. & Manning, T. (1997). Information environments for breast cancer
information. In R.L. Street, W. R. Gold, & T. Manning (Eds), Health promotion
and interactive technology: Theoretical applications and ﬁtture directions (pp. 1-
18). Mahwah, NJ: LEA.

Street, R. L. & Rimal, R. N. (1997). Health promotion and interactive technology: A
conceptual foundation. In R.L. Street, W. R. Gold, & T. Manning (Eds), Health

promotion and interactive technology: Theoretical applications and future
directions (pp. 121-139). Mahwah, NJ: LEA.

Sundar, S. S., Kalyanaraman, S. & Brown, J. (2003). Explicating web site interactivity:
Impression formation effects in political campaign sites. Communication
Research, 30(1), 30-59.

Tang, M. (2003). Spatial interface and interpersonal interface design: Design report for
HomeNetToo project. Unpublished master’s thesis, Michigan State University.

Thorisson, K. R. (1996). Communicative humanoids: A computational model of
psychosocial dialogue skills. Unpublished doctoral dissertation, Massachusetts
Institute of Technology.

Todorov, A., Chaiken, S. & Henderson, M. D. (2002). The heuristic-systematic model of

social information processing. In J. P. Dillard and M. Pfau (Eds), The persuasion
handbook Thousand Oaks, CA: Sage Publications.

Trumbo, C. W. (2002). Information processing and risk perception: An adaptation of the
heuristic-systematic model. Journal of Communication, 52(2), 367-382.

96

Van Tassel, J. M. (1988). Interactive television: The eﬁ'ects of varying levels of aﬂective
self disclosure and interactivity on viewer responses to mediated health messages.
Unpublished doctoral dissertation, University of Southern California.

Vorderer, P. (2000). Interactive entertainment and beyond. In D. Zillmann & P. Vorderer
(Eds), Media entertainment: The psychology of its appeal. Mahwah, NJ: LEA.

Walther, J. B. (1996). Computer-mediated communication: Impersonal, interpersonal, and
hyperpersonal interaction. Communication Research, 23, 3-43.

97

   

llljlll‘ljlljiljjljlljlIllljjll