Master’s Thesis:
Immersive Persuasion.
The effect of behavioral realism and social presence on social
compliance in interactive, immersive environments.
Graduate School of Communication Master’s programme Communication Science
Author. Jonas Schlicht Student-number: 11814144 Supervisor: Dr. Saar Mollen
Abstract
Social influence effects in virtual environments have often been found to be dependent on how realistic the virtual persuader behaves, if he is controlled by a human or a computer, and most importantly if he/she can elicit a feeling of “social presence”, the feeling of “being with another”. The majority of these studies assesses social influence in interactive, non-immersive environments. However with the rise of non-immersive technology like Virtual Reality and its high potential for social presence the question was posed, if social influence (i.e. social compliance) effects can be explained by the same mechanisms in a high-immersive, dyadic-conversation setting.
The study however found no significant relationship between more behavioral realism, social presence or social compliance. Conversely a negative effect of very high social
presence on social compliance was found. Results indicate that high social presence is not necessarily always positive for social influence. Further, studies in dyadic-immersive settings might be influenced by other external factors (e.g. reactance) and might need a different kind of operationalization to gain clearer results.
Introduction
Media-systems have been under constant change in the past years, generally following the trend to evermore immersive media experiences (Blascovich & Bailenson, 2011).
Beginning with mono-modal media like plain text or audio, modern audiovisual and interactive media offers a higher invitation to “lose oneself” in virtual worlds, escaping the real-life environment for some time. Surely the highest conceivable level of immersion today is through Virtual Reality headsets (VR), which finally appear to gain traction with producers and users (Rogers, 2019). The industry predicts 168 million sold VR-units by 2023, with the overall market approaching the 100-million-dollar mark by then. Not only has there been great optimism about VR as upcoming mass medium, but also as a potential communication tool (Biocca, 1995), particularly with equipment becoming more affordable and convenient to use. This hope is mostly based on two structural qualities of the medium: It’s capacity to create a feeling of ‘presence’, the sense of being and existing completely within the virtual world (Blascovich et al., 2002b). And secondly, it’s potential to create “social presence”, the feeling of sharing this virtual space with another human being.
With VR beginning to reach a mass audience, it does not surprise that the medium has been eyed as a future marketing and sales tool (Walker, 2018). Most of the research on the topic has examined how high levels of presence/social presence influences persuasive
outcomes (for a review: Fox, Christy, & Vang, 2009). Generally speaking findings have been promising: Higher levels of social presence have been found to make health-messages more convincing (e.g. Skalski & Tamborini, 2007), to benefit brand attitudes and purchase
intentions (Li, Daugherty, & Biocca, 2002; Yim, Cicchirillo, & Drumwright, 2012) as well as brand recall (Keng & Lin, 2006).
The connection between social presence and persuasion has been mostly studied in the context off social influence, i.e. change of behavior, which is due to the presence or action of another person (Allport, 1985; Blascovich et al., 2002b). Although virtual-environments can
be an escape from reality, social rules and boundaries are not easily discarded. Studies suggest that virtual humans can exert similar social influence-effects (e.g. conformity with others behavior, distance-keeping, feelings of social-anxiety (Blascovich 2002; Fox, 2009)), then their real-life counterparts.
Moreover, the mechanisms of how social influence of virtual-actors takes place seems relatively well understood. The social influence model by Blascovich (2002) gained much empirical support (for a review: Fox, Christy, & Vang, 2009) for their hypothesis, that social presence is much dependent on behavioral realism, i.e. how realistic a virtual actor behaves verbally and non-verbally. It was found that higher behavioral realism in a virtual
conversation-partner, raised levels of social presence, which was found to lead to more social influence (Blascovich, 2002).
What may sound like a panacea for communication-practitioners, however needs a more differentiated look. What makes behavioral realism and social presence studies hard to compare is their widely different approach to operationalize the subject. The persuasiveness of social presence have been studied in online-environments (e.g. Shin & Shin, 2011; Cyr, Hassanein, Head, & Ivanov, 2007; Skalski & Tamborini (2007), in so called CAVE-systems1 (Grigorovici, 2003; Garau, Slater, Pertaub, & Razzaque, 2005) and in typical VR-scenarios (e.g. Blascovich, 2002; Bailenson et al., 2005; Guadagno, Blascovich, Beall, & McCall, 2007). All these systems allow for very different levels of immersion and presence
(Grigorovici, 2003). Higher presence in low immersive-mediums (e.g. websites), might have different effects on persuasion then in a high-immersive like VR. Findings for VR are not yet conclusive, with some studies finding increased persuasion with higher social presence (e.g. Guadagno et al., 2007; Bailenson et al., 2005), while others do not (Beall et al., 2003;
Allmendinger, 2010). Therefore the connection between social presence and social influence in VR demands more academic attention.
1 An immersive environement , where the virtual world is projected onto a three-dimensional room, in which
The present research suggests that the beneficial effects of social presence on persuasion/social influence can be found on a wide basis of outcomes, from message-agreement, brand-perceptions, to behavioral indicators like mimicry or conformity (for an overview: Fox et al., 2009, Blascovich, 2002). However, social influence has not been studied in this context of social-compliance as defined by Cialdini (2001), i.e. the change of behavior following a direct request from another person. A study by Eastwick and Gardner (2009) found that social-compliance could be replicated via text in an online-game environment. However similar studies in an immersive environment are missing.
This appears problematic for multiple reasons. Social-compliance tactics are a number of strategies, which are very frequently used by salesmen and marketers alike (Cialdini, 2001). With VR becoming a communication medium (Biocca, 1995) and targeted by marketers (Walker, 2018) it appears more than likely that social-compliance tactics will follow, e.g. in the form of VR-marketing, political- or health campaigns (Biocca, 1995; Bailenson, 2018). Therefore it appears to be crucial to assess if social-compliance can be replicated in VR and if it adhere the same theoretical concepts (i.e. behavioral realism and social presence) (Blascovich, 2002) as other social influence-effects.
Addressing these two issues the present aims to study social-compliance within a high presence/social presence VR setting. We pose the following Research Question:
RQ: What is the effect of behavioral realism on social-compliance within a dyadic VR setting? And is this effect dependent on the level of social presence?
Virtual Environments, presence and persuasion
In the scientific literature there has been a struggle to find a clear terminology, for what commercially has been described as “virtual reality” (Lanier, 2017). Commonly used is the term “virtual environment” (VE), which describes the “organization of sensory
information that leads to perceptions of a synthetic environment as non-synthetic”
technology types, such as CAVE-systems, augmented reality, online-worlds like “Second Life”, virtual reality and videogames (for an overview see, Pan & Hamilton, 2018). Although all of these systems are capable of creating visual illusion, they widely differ in their potential for immersion (i.e., the medium’s technological capacity to create sensory-realistic
experiences; Slater & Wilbur, 1997), which makes studies using this term hard to compare. A somewhat more precise term is “immersive virtual environment” (IVE), which presupposes a medium’s capacity to create a strong feeling of “presence” (Blascovich et al., 2002b). While closely related to the technological-term immersion, the concept of presence describes the psychological experience of existing within a virtual world, while mentally leaving the own physical environment (Allport, 1985). According to a “Reality-Virtual continuum” (Grigorovici, 2003), VR via a head-mounted-display (HMD) ranks on top in its capacity for immersion and presence, only second to real-life experiences. Since the current study makes used of such a system, it will refer to it as IVE from now on, while laying the focus on research, which is in line with this definition.
IVE’s capacity to create higher immersion than other media, is much based on it’S capacity to track and render various real-life behavior and cues (i.e. auditory, haptic and visual) almost in real-time (Blacovich & Bailenson, 2011). While IVEs therefore have the general capacity to transmit sensory-rich, high-presence content, it remains to be explored how this effect persuasive processes and outcomes.
Virtual Environments, social presence and social influence
Future marketing or persuasion-campaigns in IVEs might come in many shapes and forms: User might be invited to virtually try out products (as done by the Swedish car company Volvo; DMI Institute, n.d.) or might be positively engaged in an advergame with a brand(as done by Oreo; mbryonic, 2019). However, it appears to be likely that users will also be targeted by virtual-humans, e.g. by reporting or demonstrating the benefits of a certain product or behavior; this human-involved persuasion is based on the power of social
influence.
Social influence has been defined as changes in cognition, attitudes, physiological responses and behavior due to the presence or actions of another person (Allport, 1985; Cialdini & Goldstein, 2004). It is driven due to humans being extremely social and adjusting their behavior, e.g. through social norms, conformity, identification or compliance, to that of other people (Cialdini, 2001). Much of the research of social influence in IVEs is theoretically based on social presence, a subcomponent of presence, and the “threshold model of social influence” by Blascovich (2002). Therefore, it is important to discuss both in more precision. Social presence and Immersive Media. The definition of social presence is an ongoing struggle, with some researchers stressing the role of the medium (Becker & Mark, 2002) and others stressing the subjective experience of the individual of “being together with others” (Biocca & Harms, 1997). While belonging to the latter group, Blascovich et al. (2002b) defines social presence within IVEs as a “psychological state in which the individual perceives himself or herself as existing within an interpersonal environment” (p. 105). Since this paper is particularly concerned with interpersonal communication in IVEs, it will subsequently make use of the Blascovich definition.
The experience of social presence relies heavily on the amount of sensory-information (i.e. non-verbal cues) received from the conversational partner, independent of if this “other” is actually human- or computer-controlled. Since IVE-systems are able to transmit a wider variety of sensual cue, than traditional media they not only offer the capacity for high presence, but also for higher social presence (Grigorivici, 2003; Pan & Hamilton, 2018). Higher levels of social presence are generally believed to have a positive outcome on persuasion (for an overview; Fox et al., 2010). To a large degree these findings are however found within non-immersive settings, such as online-environments or websites (e.g. Skalski & Tamborini, 2007; Cyr et al., 2007). Given the high capacity for social presence in IVEs, it
seems particularly intriguing to study social influence effects in a medium which is able to create a feeling of socialness, which might come close to real-life encounters.
The threshold-model of social influence. According to the social influence model (Blascovich, 2002) the question if social influence takes place in an IVE, depends mainly on the level of social presence (Fox et al, 2009). According to the model, social influence
increases as a positive function of social presence.2 Blascovich identifies two external factors, which influence the level of social presence the most: agency and behavioral realism.
Behavioral realism describes the degree to which a virtual representation of a human or non-human “other” behaves in a realistic manner (Blascovich, 2002b). In particular this includes non-verbal behavior like eye-gaze, mimics, gestures, which need to be displayed in authentic, believable manner. High behavioral realism further demands a match between non-verbal cues and actions/spoken word, e.g. with angry hand movements only making sense, in combination with angry behavior or language. Overall the model predicts, that higher
behavioral realism creates more social presence.
The second factor agency is the degree to which the “other’s” digital representation is perceived as being controlled by a human, opposed to a computer. Visual representations entirely controlled by a human are called “avatar” (high-agency), while representation entirely controlled by an algorithm (low-agency) are defined as “agents”. In general avatars were found to create more social presence then agents (Fox et al., 2015).
According to the model social influence takes place, once the additive effect of agency and behavioral realism passes a threshold, however independent of one factor being stronger than the other (Blascovich, 2002b). Given that the “other” is controlled by a human, this avatar would need only little behavioral realism to achieve the social presence-threshold.
2 In another paper Blascovich et al., (2002) offers a slightly different model, arguing that behavioral realism and
social presence relate complementary and additive to each other until the threshold of social influence is reached: social presence is hereby somewhat equated with agency. The present study orientates itself more on the threshold model described in Blascovich (2002), which treats agency and social presence as separate concepts, with the latter2 being a positive function of behavioral realism and agency.
Conversely, a computer-controlled agent would need high-level behavioral realism in order meet it and exert social influence.
Where the threshold for social influence is located, however depends on the type of social influence. Blascovich differentiates between low-level and high-level-responses. According to theory and empirical evidence low levels responses (like reflexes or strongly socially conditioned before, like conformity) have a lower threshold for social presence, than high level responses like attitude-change or message-agreement (Blascovich, 2002).
Testing the social influence model. There has been much empirical support for the social influence model, concerning the relationship between agency, behavioral realism and social presence. The role of agency on social influence found convincing support from a recent meta-analysis by Fox and colleagues (2015). The study found overall stronger social influence for avatars compared to agents, supporting the hypothesis of higher agency being beneficial for social influence. Concerning the role of agency and social presence there is little controversy, indicating no further need to manipulate this concept.
Similarly well-established in the empirical literature is the connection between higher behavioral realism and higher social presence. Behavioral realism was found most effective when virtual non-verbal behavior matched typical human-behavior. Higher levels of social presence have been found as a result of natural nodding behavior (von der Putten et al., 2010), or a strongly blushing agent, after making an “embarrassing” mistake (Pan et al., 2008). The model’s prediction of higher social influence with higher levels of social presence also found much empirical support, however also leaving some questions unanswered for the current study. Firstly, a large part of studies claiming a connection between social presence and persuasion or social influence, studied those phenomena in non-immersive environment. Skalski and Tamborini (2007) showed that higher social presence increased behavioral-intentions in response to a health-related persuasive message, coming from an online conversational-agent. In an e-commerce set-up higher social presence was linked to more
favorable brand (Fortin & Dholakia, 2005) and product-attitudes (Lee & Nass, 2004; Lee & Nass, 2005). Further websites conveying more social presence were found to be more trustworthy (Keeling, McGoldrick, & Beatty, 2010). However, none of these studies took place in an IVE, and none of them placed users in an interactive scenario with another human. Therefore it is assumed that overall levels of social presence were likely to be low in these environments.
A similar critique hold for studies, which were conducted in IVEs. Bailenson & Yee (2005) found that agents, who mimicked a user’s head movements, were more effective in advocating a campus-security-policy than non-mimickers. Mimicking-behavior typically occurs in most face-to-face conversations (Chartrand & Bargh, 1999), therefore more
mimicking represented a higher level of behavioral realism. Similarly, Guadagno et al. (2007) found that persuasive messages delivered by an agent to be more persuasive with higher levels of behavioral realism. In both studies however the message was delivered one-way, without the possibility for a dyadic-interaction. Therefore, IVEs were studied as a distribution, rather than a communication-medium. Moreover the persuasive message was read by an agent, again suggesting a relatively low overall level of social presence, compared to the medium’s capabilities.
One study within IVE and within a dyadic-interactive research setting was conducted by Bailenson et al. (2005), who manipulated behavioral realism via more or less eye-gaze. Participants in this study were allowed to have an interactive discussion, while subsequently listening to a persuasive message. Avatars showing more eye-gaze were found to be more effective in attitude-change, than their low-behavioral realism counterparts. Since this effect was only found for women, this study suggests. that the social influence model works similarly in an interactive IVE-setting.
Some inconsistencies in the literature might arise through different results for low-level and high-low-level social influence responses.Low-level responses like interpersonal
distance, conformity and social-facilitation and inhibition have been consistently found to be influenced by higher levels of behavioral realism, agency and social presence (Blascovich, 2002; Bailenson, Blascovich, Beall & and Loomis, 2015). Again, it seems safe to assume that this connection has been theoretically settled, without further need of exploration. For high-level responses the picture appears somewhat less clear with results being mixed. Bailenson et al. (2005) found more attitude-change with higher levels of social presence (for women). On the other hand both experiments by Guadagno et al. (2007) failed to show a connection between higher social presence and attitude change, only finding an interaction effect with gender, with same-sex dyads being more persuasive than mixed-sex dyads.
Further high-level responses in IVEs until now have been solely studied in the form of attitude-change after receiving a message, however seldomly as concrete behavior. However social influence for salesmen and marketers is most relevant in the form of social-compliance as defined by Cialdini (2001), i.e. as a change of behavior resulting from a direct request. One of the few studies in non-immersive digital-environments (Eastwick and Gardner, 2009) have found that social-compliance techniques like the often used “foot-in-the-door” and “door-in-the-face” technique prove to be effective tools of persuasion in an avatar-mediated online-environment.3 However it remains scientifically- omitted if these responses can be found in IVE-settings and how they relate to social presence.
Overall it can be stated that as predicted from the social influence model, high-level responses appear to be harder generate (due to a higher threshold), which might explain the somewhat mixed results. The threshold for social-compliance as behavioral measure might lie even higher, since it is not passive, but requires real-life action. Therefore, it appears
particularly intriguing to study social-compliance in an interactive IVE-scenario, which is likely to elicit high-level social presence, which might be necessary for such effects.
3
While FITD effectiveness is believed to rely on the sense of self and consistency of the participant (Cialdini & Goldstein, 2004), DITF’s effectiveness is likely to rely on the principle of reciprocity, i.e. the urge to return a favor (Cialdini et al., 1975).
Behavioral realism and the importance of eye-gaze
Having established the crucial role of social presence for social influence, the question arises how this concept is best operationalized. Many studies particularly in IVEs manipulate agency or behavioral realism in order to increase perceptions of social presence. The latter is particularly often done, by altering the amount of eye-gaze which the virtual “other” engages in (e.g. Garau et al., 2005; Bailenson et al., 2005; Guadagno et al., 2007). The choice for eye-gaze among many possible non-verbal cues (e.g. mimics, hand gestures) is not random, but well founded in theory, which should be briefly addressed.
There is little controversy in the literature about the crucial role of eye-gaze for human communication (for a review: Kleinke, 1986; Segrin, 1993). Eye-gaze serves at least five communication functions (Garau, 2003), for instance being crucial for signaling attention (Breed, 1972; Kelly, 1978) or regulating conversational-flow, e.g. in the case of turn-taking (Argyle; 1988; Kendon, 1977). Moreover for virtual characters (like avatars) the eyes appear to be the single most expressive feature to convey realistic emotions (Thomas & Johnston, 1981).
In line with the social functions of eye-gaze, it has been found to also be a strong predictor of persuasion and social influence. A meta-analysis about the effect eye-gaze on persuasive outcomes in real-life scenarios, found a significant advantage of more eye-gaze , accounting for 20% of variance across studies (Segrin, 1993). Positive persuasive effects of eye-gaze were further found in an interactive IVE setting. Bailenson et al. (2005) manipulated behavioral realism by altering the amount of eye-gaze of an avatar, which was found to enhance persuasive outcomes for women. The authors therefore posed the idea that women might be more responsive to non-verbal cures like eye-gaze.
Most importantly for the current study however, this higher persuasion (Bailenson et al., 2005) was found due to increased feelings of social presence. A connection between more eye-gaze and higher social presence has previously been found in
non-immersive-environments and for conversational-agents (e.g. Rüggenberg, Bente & Krämer, 2005; Bente, Rüggenberg, Krämer, 2004). Conversely too prolonged eye gaze (16 second; Bente,
Eschenburg, & Aelker, 2007) or averted eye gaze has been linked to more negative
evaluations of the conversation-partner (Larsen & Shackelford, 1996) and less social presence (Bailenson et al, 2005).
Although the literature within IVEs is still sparse, the discussed studies offer
confidence that eye-gaze can be a powerful tool to operationalize behavioral realism and to influence feeling of social presence and influence both positively and negatively. Given the findings of Bente and colleagues (2007) the naturalness of gaze-behavior (i.e. avoiding staring behavior) appears important for a successful manipulation. Moreover the possibility of an interaction with gender (Guadagno et al., 2007; Bailenson et al., 2005) need to be further elaborated.
Experiment goals and hypotheses
Considering the inconsistencies within the literature three main reasons for uncertainty were identified. Firstly if results from non-immersive studies would translate into immersive IVE environments, which offer much higher potential for social presence. Secondly if non-interactive, one-way studies differ fundamentally in their working from settings, which feature a dyadic, interactive setting, which more strongly would resemble a face-to-face conversation. Such a conversation might more likely elicit feelings of “being together”, which should increase social influence. Lastly the effect of social presence on behavioral high-level responses like social-compliance are not well established yet.
Addressing these issues the current study aimed to test social-compliance within an interactive, IVE setting. Moreover it was aimed to assess external influences like gender, however in a subsequent exploratory manner. In line with the discussed literature two hypotheses are posed:
H1: High behavioral realism (eye-gaze) compared to low behavioral realism (no eye-gaze) will lead to more social-compliance behavior in a dyadic conversation in an IVE.
Secondly we tested the assumptions of the Blascovich model for social influence (2002), assuming they would be the same for social-compliance in the given setting: H2: The effect of behavioral realism (eye-gaze vs. no eye-gaze) is mediated by social presence.
Method
Research Design and Participants
The study employed a single-factor (high behavioral realism vs. low behavioral realism) between-subjects design. Behavioral realism was manipulated by altering if avatars engaged in eye-gaze during the conversation or not. Participants were recruited via the recruitment-website of the University of Amsterdam. Out of 117 participants, 26 were excluded from the analysis: Due to the interactive nature of the experiment some
conversations developed different than described in the script (n = 10), which threatened validity and reliability of the results. For the same reason, participants who voiced a suspicion about the “other” avatar not being a student (as pretended by the researcher), but a study confederate were excluded (n = 16) 4.
The final sample therefore consisted of 91 participants (19 male, 72 female), ages ranged from 17 to 32 (M= 20.68, SD= 2.61). The large majority of participants were enrolled as students at the University of Amsterdam (53.3% Undergraduates, 37.6% Graduate
students). The level of familiarity of participants with VR was relatively low: 53.3% indicated they had “little prior experience” in VR and for 22.2% it was the first-time using it.
Participants were either reimbursed with 2 Research Credits or with 5€. Additionally, each participant “won” 1,50€ during the game they took part in. The study was reviewed and
4 Although 16 people voiced suspicion about the “other” being a confederate, only 4 of them suspected the
approved on the 6th November, 2019 (2019-PC-11467) by the Ethical Review Board of the University of Amsterdam.
Procedure
Upon arrival, participants were led into a small room. As part of the cover story to have a conversation in VR, they were informed to play a quiz in “Virtual Reality” against a second student (actually controlled by the experimenter). Participants were informed that the “other” student was placed in a separate room and that they were not supposed to meet him/her in person, in order to avoid confounding effects. Further they were told about their possibility to win additional money within the quiz game, i.e. 5€ for a win against the “other”, 1,50€ for a draw and no money in case of a loss (i.e. 5€ for the opponent). This was followed by instructions about how to use the hardware (i.e. the HMD and the controllers) and about the 3-phase-procedure of the experiment (1. Introduction, 2. Quiz, 3. Small-Talk) (see Appendix 3 for the whole script, including the exact manipulation/duration of eye-gaze). Introduction. Prior to the game each participant was randomly assigned to one of the two behavioral realism conditions (eye-gaze vs. no eye-gaze). Once both “players” joined the IVE, they were supposed the introduced themselves shortly to the “other”. As part of the cover-story the researcher introduced himself as a Graduate student in Biological Sciences, who was working as an intern within a charity. The researcher altered his voice to ensure that he was not recognized as the researcher giving the instructions.
Quiz. Subsequently the “other” avatar started the quiz. Each player was given four multiple-choice-questions with four answer options out of the categories geography, food, film and the human body. To reinforce his expertise in Biology the “other” avatar answered his question about Biology with ease in the beginning. Later in the game (Question 3) the participant received a particular hard question about the Human body. The “other” avatar subsequently decided to help him/her by giving a clear hint for the correct answer, again
referring to his Biology background. This manipulation was done as part of a well-researched social-compliance technique, the reciprocity principle (e.g. Cialdini, 2001). The idea was that giving a favor creates the urge to return this favor in some way to the giver. The quiz finally always ended in a draw, due to experimenter always having the second question and therefore the opportunity to equalize scores. This ensured, that each participant won the same amount of money, but also strengthened the notion that the “other” sacrificed his own win/money in order to help the participant.
Small Talk. After the quiz participant and “other” were given 5 minutes to talk about a random topic. However, the “other” quickly directed the conversation towards the won money, voicing his intention to donate his 1,50€ to the charity (Water.org) he is working for. Subsequently he asked the participant if he would like to donate some of the money as well, since he was currently raising money for them. According to the reciprocity principle (Cialdini, 2001) we expected participants to be willing to return the previous favor from the quiz, by complying to the request.
Post-Game. Subsequently the VR session was ended, equipment was taken off and participants filled in a five-minute survey, asking questions about their perception about the conversation, their conversation-partner and their level of presence. They were remunerated with 1,50€ in 50 Cent for their draw. If they indicated in- game that they would like to donate or “maybe” would like to donate, the researcher subsequently asked for the donation in the name of the other student. If the participant denied the request in-game, no such thing was asked. The complete experiment took roughly 40 minutes to complete and was administered in a six-week time frame in November-December 2019 at the University of Amsterdam.
Independent Variables
The study assessed the effect of behavioral realism (eye-gaze) versus low behavioral realism (no gaze) on social-compliance in IVEs. This manipulation was chosen, since
eye-gaze was found to be a good operationalization of behavioral realism (e.g. Garau, 2003; Bailenson et al., 2005b), while being able to affect social presence (e.g. Bailenson et al., 2005a; Bente et al., 2007) and social influence (e.g. Segrin, 1993) both positively and negatively. Therefore, it was expected to find noticeable differences for avatars, who established or averted eye-gaze.
In the low behavioral realism condition the “other avatar” gazed straight ahead for the duration of the introduction and small-talk-phase of the game. Therefore, the gaze was located at approximately 90 degrees from the gaze-direction of the participant’s avatar (Figure 1), which did not allow for any mutual eye-contact.During the quiz the gaze-direction only changed when the experimenter read quiz-questions/answers from the screen, turning his head towards the screen (approximately 180 degree from the participant-gaze) before returning to his default direction. Therefore, the avatar was not static, however only avoided direct eye-gaze at the participant-avatar.
In the high behavioral realism condition direct eye-gaze by the “other” was established throughout the game, however slightly different in each of the three phases. During the
introduction and the conversation-phase, constant eye-contact was established by the “other”, both when he was listening or speaking himself. Since the avatar showed blinking-behavior we did not assume this to be perceived as staring, but rather as signaling attention to the conversational partner like in real-life dyadic settings (Garau, 2003).
During the quiz the experimenter-avatar gazed at the participant in periods of direct interactions (e.g. when giving the hint for the correct answer), however again turned the head towards the screen for reading the quiz-questions. This allowed for more interactivity, while proving that the gaze was not static or pre-programmed. This was particularly important since static gaze was found to create feelings of less social presence and influence (Garau, 2003; Bailenson et al., 2005). For the duration of the participant’s answer the experimenter-avatar turned his head again towards him/her, signaling listening behavior (Garau, 2003). In total the
“other” avatar therefore established more eye-gaze behavior while listening (vs. when talking himself), which mirrored typical real-life gaze-behavior (Garau, 2003). Overall this resulted in seven eye-gaze moments with varying length (see Appendix 3).
In both conditions the “other” avatar was easily identifiable as human-controlled, due to the interactive nature of the conversation. Therefore, the factor agency was consistently kept high across conditions. According to the social influence model (Blasocvich, 2002) this should allow for high social presence and a likely occurrence of some social influence in both conditions. Consequently, we aimed to single out the effect of behavioral realism (i.e. eye-gaze) on social presence and compliance.
Figure 1. Setting of the game and the other avatar within the eye-gaze condition (left) and the full “other” (right)
Materials and Equipment
The participants and experimenter wore an “Oculus Go” head-mounted-display. The display featured 5.5-inch LCD display, with a resolution of 2560 × 1440 pixels, refreshing
pictures at 72Hz. The social network platform in which the quiz and chat were held was
“vTime”, which features human-looking avatars and has been successfully used before to study dyadic-communication in IVEs (Baccon, Chiarovano, & MacDougall 2019). In order to
make the participant-avatar work for both sexes an avatar was created, with all visible parts of
“other”, a male avatar was created (Figure 1).
The game and conversation took place within the chat-room-template “Conference
Room”, featuring a large table and a big interactive screen, on which the quiz questions and answers were displayed (Figure 1).The participant-avatar was seated at the head of the table,
while the experimenter-avatar was seated on the right side of the table, roughly 90 degrees
from the participant, making both avoidance and establishment of eye-gaze easily feasible.
Both avatars engaged in automatic random eye-blinking and hand gestures, which however
did not necessarily match or correspond to their speaking.
Measures
Donation behavior. Social-compliance as the main outcome variable was measured via donation behavior. Hereby the success of the reciprocity manipulation was measured, assessing if the in-game favor of the “other” was reciprocated in form of a real-life donation by the participant (Cialdini, 2001). Within the game students were asked by the “other” avatar to donate their earnings from the quiz to charity. Only if they indicated their interest in-game, and only if they handed the money to the experimenter to the experimenter in real-life by request, the donation was counted. The total amount of donation was noted down. Although participants could have decided to give less than 1,50€, none of them did so. This resulted in a dichotomous variable with people either donating the full 1,50€ or not donating at all. In all other cases the variable was coded as “no donation”.
Behavioral realism. Behavioral realism as central concept to the Blascovich model (2002) was measured on a four-item seven-point Likert-scale (1 = strongly disagree to 7 = strongly agree) (“I felt like I was interacting with a real person.”; “The virtual person moved like a real person”). The scale was adapted from Guadagno et al. (2007), who used it within a very similar IVE-setting, while also manipulating eye-gaze. The scale showed good reliability (M = 4.31, SD = 1.41, Crohnbach’s alpha = .86)
Social Presence. Social presence as the second central concept was measured on a 5-item seven-point Likert-scale (1: strongly disagree to 7: strongly agree). The scale was based on Nowak and Biocca (2003), however slightly adjusted. Two items of the scale were not in line with our definition of social presence as psychological state/experience (Blascovich et al., 2002b), but rather conceptuaölized it as medium-quality (“How likely is it that you would choose to use this system of interaction for a meeting in which you wanted to persuade others of something?”; “To what extent did you feel you could get to know someone that you met only through this system?”) Therefore they were replaced by the items “To what extent did your VR partner seem “real”?” and “How much did you feel like you were ‘with’ your VR partner?”, as similarly done by Skalski and Tamborini (2007). This resulted in acceptable scale-reliability (M = 5.04, SD = 1.05, Crohnbachs alpha = .75)
Exploratory measures
The study assessed multiple exploratory variables (embarrassment, interpersonal liking, communication satisfaction and attitude), which have been previously shown to play a role in social influence processes in IVE or non-IVE settings. They were assessed as potential external influences or mediators. Since they are not part of the main analysis,
operationalization and results for these measures will be added to Appendix 1.
Covariates
Several demographic and background variables were assessed, that is participant age, gender, highest achieved education and previous experience with VR. Experience was
measured with a one-item seven-point Likert-scale (1: Strongly disagree to 7: Strongly agree) (“How would describe your level of practical experience with VR prior to this experiment?”) (M = 2.04, SD =.73). Correlation analysis revealed marginally significant correlations
between education and social presence (r = .18, p = .095), which will therefore be tested as a covariate in subsequent analyses. No other correlations approximated significance. However
due to its theoretical relevance for social influence outcomes in IVEs (e.g. Guadagno et al., 2007; Bailenson et al., 2005) it was decided to add gender as a second covariate to subsequent analyses.
Analysis plan
To test assumptions of a normal distribution the Kolmogorov-Smirnov test was run for the continuous behavioral realism and social presence. Behavioral realism showed signs for a significant non-normal distribution. However, levels of skewness and kurtosis did not exceed -1/+1, which indicates tolerable levels (Hair, Hult, Ringle & Sarstedt, 2017). Moreover, Levene’s test revealed a violation of the assumption of homoscedasticity for behavioral realism, F (1, 88) = 6.87, p = .010 and social presence, F(1, 88) = 4.10, p = .046. Since the size of the two conditions is roughly equal (49 vs. 42) it was decided to proceed with the analysis, using a bootstrapping procedure for robustness (Field, 2013). Education and gender were tested for each analysis as covariates. However, since they failed to reach significance in any analysis they were subsequently dropped.
In the beginning a one-way ANOVA was conducted to test if the manipulation of eye-gaze on behavioral realism was successful. Subsequently H1, that is the main effect
behavioral realism (eye-gaze) on donation behavior, was tested by the means of a chi-square test, taking into account that the dependent variable proved to be dichotomous. To test the mediation hypothesis of the effect via social presence (H2), a mediation analysis using Hayes SPSS macro PROCESS (Model 4) was run. In order to calculate the mediation-model
PROCESS generated 5000 bootstrap samples, resulting in point-estimates for the indirect effects together with bias-corrected 95% confidence intervals. The newest version of PROCESS (v. 3.3), automatically detects dichotomous variables, and analyses them via a logistic regression test.
Lastly several more exploratory analyses were conducted. Firstly another sequential mediation-model was tested (PROCESS, Model 6), assessing if the effect of eye-gaze was
sequentially mediated by behavioral realism and social presence. Finally, the role of gender was more closely examined, testing a possible main effect and/or interaction effects
(condition*gender) on donation behavior, via logistic regression.
Results
Manipulation check
To test if the manipulation of eye-gaze, i.e. its effect on behavioral realism was
successful a one-way ANOVA was conducted. Results indicated that behavioral realism in the eye-gaze condition (M = 4.61, SD = 1.11) was only slightly higher than in the no-gaze
condition (M = 4.05, SD = 1.59). Although in the intended direction, this difference proved to be small and only marginally significant, F(1, 89) = 7.18, p = .056, η² = .04. This result was not expected and reveals that the manipulation worked only very subtle. Acknowledging this the analysis continued with the main analysis.
Main analysis
Main effect. To test the hypothesis that high behavioral realism (vs. low behavioral realism) leads to higher levels of social influence (donation behavior), a chi-square analysis for the two eye-gaze conditions was conducted. Even though participants in the two
conditions showed differences in donation behavior in the expected direction (high: 57.1% donators and 42.9% donators, low: 42.9% donators and 57.1% donators), this difference was non-significant, χ2
(1) = 1.85, p = .174. Hypothesis 1 was therefore not supported.
Mediation analysis: behavioral realism. Hypothesis 2 stated that the effect of behavioral realism on donation behavior is mediated by social presence. Regression results indicate that behavioral realism did not significantly predict social presence, b = .17, SE = .22, p = .457. Therefor social presence did not significantly change when participants were looked at, compared to when they were not.
conducted a logistic regression analysis. Results showed that social presence was not able to significantly predict donation behavior, b = -.32, SE = .21, p = .129. Therefore, higher levels of social presence did not affect if participants donated. Consequently, the pathway for the indirect effect remained non-significant with the confidence intervals crossing zero, b = -.05, SE = .10, 95% CI [-.25, 1.46]. Hypothesis 2 was therefore not supported.
Exploratory analysis
Mediation analysis: sequential mediation. Previous analyses revealed two important insights. Firstly, the manipulation of eye-gaze as proxy for behavioral realism appeared to be weak, but (marginally) significant. Secondly the eye-gaze-manipulation did neither have an effect on social presence, nor on donation behavior. However, the question was posed if eye-gaze and perceived behavioral realism should be tested as separate concepts. Assessing them individually appeared to be justified in the present case, since eye-gaze (vs. no eye-gaze) somewhat failed to be a strong proxy-behavior for high (vs. low) behavioral realism. Testing H2 with separate concepts again, it was explored if the effect of eye-gaze on donation behavior was sequentially mediated by perceived behavioral realism and social presence. In line with previous findings, the regression showed no direct effect of eye-gaze (vs. no eye-gaze) on donation behavior, b = .51, SE = .45, p = .258, but a marginally significant positive effect of gaze on perceived behavioral realism, b = .57, t(87) = 1.94, p = .056. Participants who were gazed at (compared to no-gaze) therefore scored higher on perceived behavioral realism.
Moreover behavioral realism now significantly predicted social presence (b = .40, t(86) = 5.96, p < .001) when controlling for eye-gaze. With increasing behavioral realism therefore also levels social presence increased. Gaze-condition on the other hand showed no direct effect on social presence (b = -05, t(86) = -.31, p = .758), indicating that eye-gaze did not alter levels of social presence directly.
regression. When controlling for eye-gaze and behavioral realism, social presence significantly and negatively predicted donation behavior, b = -.54, SE = .27, p = .041. Looking at the odds ratio, a one-unit increase in social presence decreased the odds of
donation by roughly 30%, which was contrary to previous expectations. The direct path from perceived behavioral realism to donation behavior remained positive but non-significant, b = .28, SE = .19, p = .152. The complete indirect sequential-mediation-pathway was marginally significant, b = -.12, SE = .11, 95% CI [-.42, .01]. Given the confidence interval we cannot rule out the possible that the indirect effect is zero, however this seems unlikely.
The role of gender. Multiple studies have suggested that eye-gaze manipulations might vary in effectiveness for different gender, i.e. with women being more susceptible to them (e.g. Guadagno, 2007; Bailenson & Beall, 2005). Moreover same-gender dyads were found to be sometimes more persuasive compared to mixed sex dyads (Guadagno, 2007). Until now gender was treated as a covariate, however it seems valuable to explore if gender might have interacted with the eye-gaze manipulation, which might have driven the present results.
To test this idea a logistic regression analysis was conducted, assessing if gender and eye-gaze condition or the interaction term were able to predict donation behavior. Overall men donated less frequently (36.7%) than women did (53.5%), suggesting no ingroup-favoritism effect, but eventually a higher effect of eye-gaze for women. Results of the regression analysis however did not support this notion, with neither gender and eye-gaze alone, nor the interaction term predicting donation behavior significantly, b = -1.10, SE = 1.12, p = .333.
Further it was assessed if the interaction between gender and eye-gaze might have effect perceived behavioral realism and/or social presence. To test this idea, two separate regression analysis were run with gender, eye-gaze condition and the interaction term as predictors. Again, eye-gaze and gender on their own remained non-significant. Similarly, the
interaction term did neither predict perceived behavioral realism (b = -.46, t = -.62, p =.536) nor social presence (b =-.24, t = -.43, p = .666). Opposed to previous studies it can be
therefore stated, that gender of the participant and mixed- or same-sex-dyads did not affect the present analysis.
Discussion
The current study aimed to test the model of social influence within a dyadic,
interactive IVE setting. In particular it was assessed if higher behavioral realism (eye-gaze vs. no gaze) would increase levels of social presence, which in turn would increase
social-compliance. Thereby this study addressed three main gaps in the literature. Firstly it was assessed if positive results found in non-immersive, one-way communication settings, would translate into a high-immersive, dyadic-interaction settings. This idea was based on the notion that the latter scenarios offer far more potential for high-social presence (Grigorovici, 2003), which is likely to affect social influence. Lastly the study aimed to test social influence in a new way. While much studies measure it in terms of low-level responses (e.g. reflexes, conformity-behavior) or internal high-level responses (e.g. message agreement, attitude change), the present measured social-compliance, i.e. a real-life change in behavior followed by a request. It was pondered that real-life behavior changes might require high-level social presence (Blascovich, 2002), which was likely to be reached within this interactive-IVE- setting.
Hypothesis 1 predicted a main effect of higher behavioral realism (that was eye-gaze vs. no eye-gaze from the “other” avatar) leading to more donations from participants. In contrast to these expectations results did not show a significant difference between the two conditions. In absolute terms participants who were looked at during the conversation donated more often than not (57 % to 43%), this effect however could have been too small to be significant. Findings that more eye-gaze in real-life leads to strongly higher persuasive
outcomes (Segrin, 1993) were therefore not replicated in an interactive IVE setting. Further findings within IVEs, e.g. that eye-gaze was able to increase message agreement (Guadagno, 2007) or attitude change (Bailenson et al., 2005) did not translate to actual behavior (i.e. compliance behavior) in this study. Also, the above-mentioned studies found differing results for men and women. It was argued that eye-gaze manipulations might be more effective for women (Bailenson et al., 2005) and that same-sex dyads lead to more persuasion than mixed-sex dyads. To test this potential confounding effect an interaction between eye-gaze and gender was tested. However, the exploratory analysis did not find any connection, therefore gender and mixed/same-sex dyads cannot offer an explanation for the present results.
Although there was no direct effect, it was hypothesized (H2) that there is an indirect effect from higher behavioral realism to higher social presence, and from social presence to donation behavior. This connection is the key mechanism of the social influence model (Blasovich, 2002) and has gained much empirical support (e.g. Blascovich, 2002; Bente et al., 2007; Guadagno et al., 2007). Nonetheless the previous results could not be replicated by the present mediation-analysis. Participants, who were gazed at during the game did not report significantly higher levels of social presence than their no-gaze counterparts. Further higher social presence did not significantly affect if participants donated or not. The proposed relations therefore did not hold for the present setting. This appeared to be odd, particularly with Bailenson and Beall (2005) finding significant effect between eye-gaze, social presence and social influence in a similar interactive IVE.
It appears likely that the opposing effects were found due to differences in the operationalization. In the Bailenson et al. study (2005) social influence was tested in a three-person instead of a dyadic setting. Although the rest of the operationalization was very similar, other differences can be found in the amount of eye-gaze established. While in the augmented-gaze condition (Bailenson et al., 2005) participants were gazed at 100% of the
time, participants in in the present study (high behavioral realism) were only gazed at during some parts of the game, which was roughly 60% of the time. Particularly during the quiz however, the “other” avatar turned his back to the participant, therefore also avoiding eye-contact for some considerable time.
Results from the manipulation check indicate that this manipulation might have been too weak: Eye-gaze was found to have only a small and marginally significant effect on perceived behavioral realism, the concept it was supposed to operationalize. It is assumed that a stronger difference in eye-gaze behavior is necessary to generate clear differences. Similarly previous manipulation of gaze, e.g. a random-gaze condition (Garau et al., 2005) or a
completely motionless computer-controlled agent (Guadagno et al., 2007) might have been more noticeable, therefore leading to larger effects.
Given the problematic power of our manipulation, an exploratory-analysis was run, testing the eye-gaze manipulation and perceived behavior realism as separate concepts. Thereby we tested eye-gaze’s effect on perceived behavioral realism, behavioral realism effect on social presence, which then subsequently effects donation behavior. In this manner the complete indirect was found marginally significant. In detail this was eye-gaze marginally increasing perceived behavioral realism, which in turn strongly and positively predicted social presence. Higher social presence on the other hand significantly decreased the likelihood for donation, opposed to our expectation.
These findings indicate two main things: Firstly even in the absence of a strong manipulation behavioral realism and social presence were found to be strongly connected, therefore further supporting the theoretical relation between these concepts (Blascovich, 2002). However, a decrease of social influence with higher social presence is almost unprecedented in the relevant literature and needs be further discussed.
Allmendinger (2010) suggested for non-immersive environments, that social presence might not always pose an persuasive advantage, depending on how much social presence a
persuasion strategy requires. Also, within IVE scenarios there have been studies finding no direct effect of social presence and social influence (Beall et al., 2003; Guadagno et al., 2007). Similar to the present study Beall et al. (2003) attributed this to a not strong enough
manipulation i.e. due to the procedure of the experiment.5 However none of these studies indicated a negative relationship. It can only be speculated why this was the case in the current study.
The most likely explanation might be that the results were due to a ceiling effect. Overall levels of social presence differed only very little between eye-gaze conditions. Average scores in both conditions were very high, with an average score above 5 on a seven-point scale. These high-level social presence-ratings are not surprising, in the sense that they were intended by our research design and predicted by the social influence model
(Blascovich, 2002).
Independent of behavioral realism the model predicts high-level social presence for human-controlled avatars, which were used in both conditions of the present study. Further given the interactivity and immersion of the present setting it appears likely that the threshold for social influence was surpassed in both conditions. Therefore, participants in both
conditions likely felt the urge the reciprocate the favor similarly strong, independent of eye-gaze and the level of behavioral realism.
This would indicate that passed the social-influence-threshold, social presence’s positive effect on social influence might not rise indefinitely and linearly, but might level out (asymptotic relationship) or even regress (reverse U-curve). Although the present results offers support for the latter, negative results might be explained by a ceiling effect, with very high results (i.e. in social presence and donations), regressing naturally (Neumann, 2014). Alternatively high social presence might have affected a not-measured external
5
This weakness in manipulation was corrected in the described later study (Bailenson et al., 2005), which did find results subsequently.
variable. In particular it seems possible that higher social presence also produced higher levels of resistance/reactance in participants in response to the donation-request. Reactance is commonly found in real-life persuasion-scenarios, when persuasion-attempts are recognized as such and perceived as a threat to people’s personal freedom (Brehm & Brehm, 1981; Knowles & Linn, 2004). Given that participants scoring high in social presence felt even more “together” with the “other”, such a feeling might have been amplified. Negative feelings of reactance are commonly released by people trying to defend their freedom, i.e. by refusing behavior change or by acting in an opposing manner (Knowles & Linn, 2004). If participants high in social presence also felt more reactance, this would explain their decreased donations.
Limitations and implications for future research
Overall the current study offers a cautionary example, that social influence in one-way non-immersive scenarios is not necessarily equally found and operationalized in interactive IVE-settings. Within this study the eye-gaze manipulation was found to be a (too) weak manipulation for the concept of behavioral realism. Future studies might use stronger
manipulation or stronger contrasts between conditions, e.g. by testing more gradual eye-gaze levels (from constant eye-gaze, medium eye-gaze to no eye-gaze). In the current study in both conditions the “other” avatar showed interactive behavior due to head-movements. A stronger behavioral realism manipulation might therefore alter if the avatar behaves static or
interactive (for a possible operation see, Bailenson et al., 2005b).
Moreover this study raises the question, which effects of social presence on social influence can be expected once the threshold (as defined by Blascovich, 2002) is surpassed, contesting a linear relationship. Testing this idea, it seems valuable to gradually manipulate social presence in a similar social-compliance-scenario. This could most easily be done by also manipulating agency (e.g. using an agent) or comparing media with different capacities for social presence (e.g. IVE vs. computer-environment).
valuable to also assess possible feelings of reactance and how they relate to social presence. Reactance might differ depending on the perceived intrusiveness of the persuasive request (Knowles & Linn, 2004), therefore different request with different levels of intrusiveness (e.g. including or not including money) might be assessed in future research.
Naturally the present research comes with limitations, mostly due to restrictions of resources. Firstly, study-instruction were given by the same experimenter who later was later conversing with the participant as the ‘other’ avatar. This setting involved the risk that the experimenter was uncovered. However, it is believed that this potential problem was solved by excluding all people from analysis, who indicated only light suspicion within the survey or in person (n = 16). Among those participants only few recognized the researcher himself (n = 4), which speaks for overall success of the cover-story.
Further the behavior and conversation between participant and the “other” were highly interactive and therefore prone to some variation. While adhering to a script, the
conversations slightly differed from person-to-person, depending on how participants responded to the given questions or situations. Therefore, the duration of the conversation eye-gaze differed slightly for each participant. However, we do not expect these differences to have influenced the validity of findings negatively, given that variations were small and were equally found in both conditions. Moreover, this operationalization offered the benefit, that conversations were held in a very realistic, natural manner, being high in ecological validity. Moreover this study was the first one to study social-compliance as a form of social influence, therefore making a step from purely internal measures (e.g. attitudes, message agreement) to conscious behavior choices. Particularly given the ambiguous results of this study it might be however valuable, to additionally assess intermediate internal measures (like attitude and behavioral intention) (Fishbein & Cappella, 2008) to gain a clearer understanding how participants are internally and behavior-wise affected by the request.
therefore general claims about social-compliance as a whole cannot be made. Social-compliance in IVEs as a form of social influence appears to be widely understudied. It is strongly encouraged to address this gap further, e.g. by studying other compliance tactics (like the five principles proposed by Cialdini (2001)) and their behavioral outcome.
Since IVEs approach to become a communication- instead of a pure distribution-medium (Biocca, 1995), we encourage future research to study social influence/compliance in IVEs not purely in a one-way context, but in more realistic interactive two-way settings. Since marketing and persuasion are preparing to enter immersive world (Walker, 2018), it bears importance for marketers, regulation authorities and researcher alike, to understand underlying mechanism and outcomes more clearly, while also pondering the risk of “too much” social presence.
Appendix 1
Exploratory Analysis: Methods
Embarrassment. Embarrassment was measured on a three-item seven-point Likert scale (1: strongly disagaree to 7: strongly agree), based on Bailenson, Swinth & Hoyt (2005) (“I would be willing to pick my nose in front of my VR chat partner.”, “I would be willing to act out a scene from the movie “Titanic” in front of my VR chat partner.”) Embarrassment has been shown to be a reliable marker for social influence in virtual environments (Bailenson et al., 2003), with higher embarrassment ratings coinciding with stronger social influence effects. Therefore it was tested as a potential mediator of the social-compliance effect. The scale showed acceptable reliability (M = 2.98, SD = 1.45, Crohnbach’s alpha = .76)
Interpersonal Liking. Interpersonal liking was measured on a four-item seven-point Likert scale (1: strongly disagree to 7: strongly agree), based on the scale by Oh, Herrera and Bailenson (2019) (“I like my VR chat partner”, “I would like to get to know my VR chat partner better.”). The variable was assessed as possible mediator, taking into account the established interconnection between liking and social influence in real life (Cialdini, 2001). Moreover previous research suggests (Bente et al., 2007; Segrin, 1993) that mutual eye-gaze positively influences evaluations, while averted eye-gaze coincides with more negative
evaluations of a conversational partner. The scale showed acceptable reliability (M = 5.55, SD = 0.77, Crohnach’s alpha = .76).
Communication satisfaction.Communication satisfaction was assessed as a possible mediator on a four-item seven-point Likert scale (1 = strongly disagree to 7 = strongly agree), adapted from Oh, Herrera & Bailenson (2019) (“I would like to have another conversation like this one.”, “I was satisfied with the conversation.”). Communication satisfaction was believed to increase with more mutual eye-gaze and higher levels of social presence
(Blascovich et al., 2002b). The scale showed good reliability (M = 5.56, SD = .87, Crohnbach’s alpha = .87).
Attitude towards VR partner. Attitude towards the conversational partner can be considered both important for persuasion processes (Cialdini, 2001), but also influenced e.g. by feelings of social presence or gaze behavior in IVEs (e.g. Bente et al., 2007). Therefore, attitude towards the VR was measured on 12-item seven-point Likert scale (1: strongly disagree to 7: strongly agree). (I think my VR partner is: likable, honest, competent, warm, informed, credible, modest, approachable, interesting, trustworthy, sincere, friendly) adopted from Guadagno & Cialdini (2002). Since the scale measures a diverse set of attributes a factor analysis was conducted to find meaningful dimensions. A principal component analysis with Direct Oblimin rotation suggested a two-factor solution, explaining 47.95 (factor 1) and 9.1% (factor 2) of the total variance. The first factor was labelled “liking”, since it included: likable, honest, warm, modest, approachable, interesting, trustworthy, sincere, friendly. The second factor was labelled “competence” since it included: informed, credible, competent. Both factors were merged into a single variable, with liking showing good reliability (M = 5.89. SD = .68, Crohnbach’s alpha = .88) and competence showing almost acceptable reliability (M = 5.99, SD = .66, Crohnbach’s alpha = .69). Both factors were added separately to the analysis.
Results
Exploratory variables
To get a general overview of the relationships of the tested variables and to get further inside in this study’s results a correlation analysis was conducted (Table 1). As predicted by previous studies social presence was positively correlated with more positive evaluations of the conversation (conversational satisfaction) and the partner (interpersonal liking, partner-liking, partner-competence (marginally sig.) (see Table 1). Therefore, it is likely that the lower donation behavior for high social presence, which was found in the sequential
mediation analysis, was not due to negative evaluations of the partner. However none of these variables correlated significantly with either the independent variable or the outcome variable donation behavior, suggesting that there was no mediation through either of these variables. This notion was confirmed by five separate mediation analysis via PROCESS Model 4, which only found non-significant indirect mediation pathways. Therefore it is concluded that none of the exploratory variables mediated the effect of eye-gaze on donation behavior.
Moreover only partner-competence (r = .237, p = .025) appeared to be significantly correlated with donation behavior, confirming the role of competence, trust and credibility as source effects for persuasion (e.g. Belch & Belch, 2012). However it can be assumed that competence evaluations did not differ due to the gaze-manipulation: An independent sample t-test, testing differences in competence evaluations between gaze-conditions remained non-significant, t(88) = -.89, p = .401.
Table 1
Pearson's r for correlations between exploratory variables and independent, outcome and mediating variables Conversational satisfaction Interpersonal Liking Embarrassmen t Partner-Liking Partner- Competence Behavioral Realism .25* .19 -.08 .16 .07 Social Presence .35*** .30** -.01 .30** .20 Donation Behavior .82 .15 .10 .12 .24* Eye-Gaze condition 0.08 .06 -.14 -.04 .09 * = p < .05; ** = p < 0.01, *** = p < .001
References
Allmendinger, K. (2010). Social presence in synchronous virtual learning situations: The role of nonverbal signals displayed by avatars. Educational Psychology Review, 22, 41-56.
Allport, G. (1985). The historical background of social psychology. In G. Lindzey & E. Aronson (Eds.), Handbook of social psychology, 3rd ed.,Vol. I, pp. 1–46. New York: Random House.
Argo, J. J., Dahl, D. W., & Manchanda, R. V. (2005). The Influence of a Mere Social Presence in a Retail Context. Journal of Consumer Research, 32(2), 207–212.
https://doi.org/10.1086/432230
Argyle, M. (1988). Bodily communication (2nd ed.). London: Methuen.
Baccon, L. A., Chiarovano, E., & MacDougall, H. G. (2019). Virtual Reality for Teletherapy: Avatars May Combine the Benefits of Face-to-Face Communication with the
Anonymity of Online Text-Based Communication. Cyberpsychology, Behavior, and Social Networking, 22(2), 158–165. https://doi.org/10.1089/cyber.2018.0247
Bailenson, Jeremy N., Blascovich, J., Beall, A. C., & Loomis, J. M. (2003). Interpersonal Distance in Immersive Virtual Environments. Personality and Social Psychology Bulletin, 29(7), 819–833. https://doi.org/10.1177/0146167203029007002
Bailenson, J. N., & Yee, N. (2005). Digital Chameleons: Automatic Assimilation of Nonverbal Gestures in Immersive Virtual Environments. Psychological Science, 16(10), 814–819. https://doi.org/10.1111/j.1467-9280.2005.01619.x
Bailenson, J.N., Beall, A.C., Loomis, J., Blascovich, J., Turk, M. (2005a), Transformed social interaction, augmented gaze, and social influence in immersive virtual environments. Human Communication Research, 31(4), 511-537.
Bailenson, Jeremy N., Swinth, K., Hoyt, C., Persky, S., Dimov, A., & Blascovich, J. (2005b). The Independent and Interactive Effects of Embodied-Agent Appearance and
Behavior on Self-Report, Cognitive, and Behavioral Markers of Copresence in Immersive Virtual Environments. Presence: Teleoperators and Virtual
Environments, 14(4), 379–393. https://doi.org/10.1162/105474605774785235 Bailenson, J.N. (2018). Experience on Demand: What Virtual Reality Is, How It Works, and
What It Can Do. New York: W.W. Norton.
Barnes, S. (2016). Understanding Virtual Reality. Marketing: Nature, Implications and Potential. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.2909100
Beall, A. C., Bailenson, J. N., Loomis, J., Blascovich, J., & Rex, C. S. (2003.). Non-Zero-Sum Gaze in Immersive Virtual Environments.
Belch, G. E, & Belch, M. A. (2012). Advertising and promotion: An integrated marketing communications perspective. 9th ed. New York: McGraw-Hill/Irwin.
Bente, G., S. Rüggenberg, & Krämer, N. (2004). Social presence and interpersonal trust in avatar-based, collaborative net-communication. 7th Annual International Workshop Presence 2004.
Bente, G., Eschenburg, F., & Aelker, L. (2007). Effects of simulated gaze on social presence, person perception and personality attribution in avatar-mediated communication. Biocca, F. (1995). Communication in the age of virtual reality. Lawrence Erlbaum Associates,
Inc.
Biocca, F., & Harms, C. (1997). Defining and measuring social presence: Contribution to the Networked Minds. Theory and Measure, 36.
Blascovich, J. (2002a). Social Influence within Immersive Virtual Environments. In R. Schroeder (Hrsg.), The Social Life of Avatars (p. 127–145). Springer London. https://doi.org/10.1007/978-1-4471-0277-9_8
Blascovich, J., Loomis, J., Beall, A. C., Swinth, K. R., Hoyt, C. L., & Bailenson, J. N. (2002b). Immersive virtual environment technology as a methodological tool for
