The Effects of Ingroup Bias
on Public Speaking Anxiety in Virtual Reality
Lotte E. J. Biesmans, Pleun J. M. van Hees, Lisa E. Rombout, Maryam Alimardani and Eriko Fukuda
Department of Cognitive Science and Artificial Intelligence, Tilburg University, Warandelaan, Tilburg, The Netherlands
Keywords:
Virtual Reality, Virtual Agents, Social Agents, Ingroup Bias, Emotional Intelligence, Emotional Interaction,
Presence, Public Speaking Anxiety, Mental Health.
Abstract:
Virtual agents can be powerful elements in virtual reality (VR) applications, as their influence on user expe-
rience is governed by complex social mechanisms. Public speaking offers a relatively high-stakes situation
that involves interaction with virtual agents. We examined the effects of an ingroup versus outgroup virtual
audience on public speaking anxiety (PSA). Additionally, we looked at how emotional intelligence and VR
ecological validity modified these effects. Results indicated that the VR application succeeded in evoking
ingroup bias, and that in the ingroup condition, self-reported PSA was related to general PSA. Emotional
intelligence was also a significant moderator. Additionally, audience type influenced the level of presence
experienced by the user: ingroup audiences result in a higher level of presence. This study identifies potential
areas of interest for future research, approaches that could influence users in specific and measurable ways in
applications involving virtual social interaction, as well as the personalization of these virtual experiences.
1 INTRODUCTION
Virtual reality (VR) applications provide a rich en-
vironment for users with many different possibilities
for sensory experiences and interactive elements. The
design of the virtual agents that might inhabit these
spaces is one of the elements that can be used to in-
fluence the player in complex yet measurable ways.
Since virtual humanoid characters are experienced by
the user on a social level, their interaction is governed
by higher-level social mechanisms and the personal-
ity characteristics of the user (Hamilton et al., 2016;
Daher et al., 2017). There are therefore many dimen-
sions to consider when designing virtual characters.
One potentially interesting dimension is that of in-
group/outgroup distinctions.
Defining other people as ingroup or outgroup
(whether they belong to the same social group as the
person or not) is a common way to code and simplify
social interactions. It influences how others are per-
ceived and judged, and how they are behaved towards.
This often manifests in the form of a preference for,
or a more positive attitude towards, the ingroup mem-
bers, which is also known as ingroup bias (Paladino
and Castelli, 2008). Conversely, perception and be-
haviour tend to be more negative towards outgroup
members.
Judging whether someone is in the in- or outgroup
is an immediate and subconscious process based on
automatic stereotyping (Wittenbrink et al., 1997).
Grouping methods can intersect - for instance, some-
one can be in your ingroup with regards to race, but
in your outgroup with regards to gender. An indi-
vidual’s strength of group identification determines
how strongly they experience the effects. The way
the behaviour of the other is perceived is affected
by their group status. For instance, people are gen-
erally more sensitive to criticism when it is made
by outgroup members as opposed to ingroup mem-
bers (Elder et al., 2005). In virtual reality, the in-
group/outgroup status of a virtual crowd can influ-
ence presence (Kyriakou and Chrysanthou, 2018),
and other studies also show that ingroup bias extends
to virtual avatars (Guadagno et al., 2007). It should be
noted that ingroup/outgroup distinctions are a com-
mon narrative device, which is also known to affect
a VR experience (Troxler et al., 2018). Obviously,
an individual’s susceptibility to ingroup bias is depen-
dent on their actual identification with the group, and
might be affected by individual characteristics, such
as their level of emotional intelligence.
Emotional intelligence (EI) is not a single trait or
ability, but can be defined as the capacity to recog-
nize the meaning of emotions, to understand their in-
Biesmans, L., van Hees, P., Rombout, L., Alimardani, M. and Fukuda, E.
The Effects of Ingroup Bias on Public Speaking Anxiety in Virtual Reality.
DOI: 10.5220/0008951600170024
In Proceedings of the 15th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2020) - Volume 2: HUCAPP, pages
17-24
ISBN: 978-989-758-402-2; ISSN: 2184-4321
Copyright
c
2022 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
17
formation, and to manage them (Salovey and Sluyter,
1997). As expected, EI highly influences social inter-
action. People with low levels of EI generally expe-
rience more anxiety in social situations (Summerfeldt
et al., 2006; Jacobs et al., 2008) whereas high lev-
els correspond to lower communication anxiety (De-
waele et al., 2008). However, there is a gap in knowl-
edge on how an individual’s level of EI influences
their interaction with virtual agents, and whether this
is comparable to their interaction with real people.
We do know that people have varying reactions to vir-
tual characters and some people are quicker to accept
and trust avatars than others (Vinayagamoorthy et al.,
2006), which might be partly caused by characteris-
tics such as EI.
A quite common way of exploring the effects of
virtual character design is through the use of virtual
audiences in public speaking situations. This context
provides an interaction with a large group of people
in a relatively high-stakes situation. Virtual audiences
consisting of avatars that display clear human char-
acteristics are regarded as more realistic, competent,
credible, and attractive (Nowak and Rauh, 2005), and
strengthen the bond the VR user feels with the audi-
ence (Nowak et al., 2009). One study (Slater et al.,
1999) focused on how an outwardly positive or neg-
ative virtual audience affected public speaking anxi-
ety (PSA). They found that, similar to real life audi-
ences, positive virtual audience interest reduced PSA.
Additionally, they found that co-presence (the sense
of being in the same space as the virtual audience)
amplified this effect. A second study found that vir-
tual audiences with an outwardly neutral or positive
emotional state caused PSA to remain the same, while
those with a negative emotional state increased PSA
(Pertaub et al., 2002).
PSA is one of the most common social pho-
bias, estimated to affect around 20 to 34% of peo-
ple (Botella et al., 2010; Pull, 2012). People suffer-
ing from PSA experience physiological arousal such
as an increased heart rate or sweating, focus nega-
tively on how they come across, and show behav-
ioral concomitants like shaking or speaking gibberish
whenever they have to speak in front of an audience
(Bodie, 2010). PSA is usually treated with cognitive
behaviour therapy including exposure (Botella et al.,
2010). Exposure used to be either real-life or imag-
ined public speaking (Wallach et al., 2009), but is cur-
rently often addressed by VR mental health applica-
tions (Pull, 2012). Exposure therapy with VR can be
equally effective in reducing PSA as real-life expo-
sure (Wallach et al., 2009) and has long-term effects
(Anderson et al., 2005). VR hard- and software does
not need to be of the highest quality, and the exposure
does not need to happen in a therapy setting for it to
be effective (Lindner et al., 2019; Stupar-Rutenfrans
et al., 2017). Both self-reported PSA and heart rate
measurements reflect a reduction in PSA when VR is
used for exposure (Harris et al., 2002).
Because the public speaking experience provides
such a highly salient social interaction, with proven
effects of virtual audience manipulations on the level
of PSA, it seems an ideal setting to explore the in-
fluence of other social mechanisms, like ingroup bias
and EI. Therefore, this study aimed to investigate
whether an ingroup vs. an outgroup virtual audience
can influence the amount of public speaking anxiety
that users experience. Since there are indications in
the literature that the behaviour of ingroup members is
immediately and automatically perceived more posi-
tively than the behaviour of outgroup members (Elder
et al., 2005), and more positive audiences are less anx-
iety inducing (Slater et al., 1999), we hypothesized
that participants would experience less public speak-
ing anxiety in front of the ingroup virtual audience
compared to the outgroup audience.
Additionally, we were interested in the effects of
an individual’s personal traits, such as EI and gen-
eral predisposition towards public speaking anxiety,
on the level of specific experienced PSA in the above
ingroup/outgroup scenario. Since previous research
shows that low levels of EI correspond to high levels
of social anxiety (Dewaele et al., 2008; Summerfeldt
et al., 2006), we expected that participants with low
levels of EI would generally experience more public
speaking anxiety than those with high levels of EI,
and that the level of EI could help explain the poten-
tial PSA difference between conditions.
Finally, as the comparison between the two con-
ditions were done in a virtual environment, we spec-
ulated that the ecological validity of the VR simula-
tion could influence the level of PSA that users expe-
rienced in each condition. This includes the level of
presence that participants felt in the virtual room, and
their actual ingroup/outgroup identification. In this
context, we define ecological validity as a measure
of how effective the virtual environment is in terms
of making the participants feel like they are in a real
space, interacting with real people.
To summarize, three research questions (RQ) were
formulated:
RQ1: What is the effect of an ingroup vs. an out-
group virtual audience on the amount of PSA that
users experience during a virtual public speaking
simulation?
RQ2: How do individual traits such as emotional
intelligence and general predisposition towards pub-
lic speaking anxiety regulate the experienced PSA
HUCAPP 2020 - 4th International Conference on Human Computer Interaction Theory and Applications
18
level in each condition?
RQ3: How does VR ecological validity, including
presence and actual ingroup identification of the par-
ticipants, regulate the experienced PSA level in each
condition?
2 METHODS
2.1 Participants and Study Design
Forty university students (23 female) were recruited
using convenience sampling. They were all native
Dutch speakers and aged between 18 and 28 (M =
22.55, SD = 2.43). The participants were quite com-
fortable speaking English (M = 4.93, SD = 1.51 on
a 7-point Likert scale). They had a medium experi-
ence with VR technologies (M = 3.75, SD = 1.81 on
a 7-point Likert scale). No pre-selection was made on
general PSA levels or presence of social anxiety. The
study was approved by the research ethics and data
management committee of Tilburg University.
We used a between-subjects design. Participants
were randomly assigned to one of the two condi-
tions: Ingroup (N = 20) or Outgroup (N = 20). In
the Ingroup condition, participants were told to give
a 2-minute presentation in English about their current
study for a group of freshman students, who wanted
to know what they could expect. In the Outgroup con-
dition, participants were also told to give a 2-minute
presentation in English about their current study, but
here it would be for a group of professors.
2.2 Materials
The virtual environment was created using Unity 3D.
It consisted of a simple room with a wooden floor
and white walls, with one big window through which
a static landscape view of a small city was visible.
Nine chairs were placed in a three by three arrange-
ment in front of the participant point of view. The
virtual audience was placed on these chairs. The vir-
tual avatars were created using the MakeHuman soft-
ware, and animated with basic breathing movements
from the Mixamo repository. Some general ambient
sounds, such as someone shifting in their chair, were
also added to the virtual environment.
The only difference between the conditions was
the physical appearance of the virtual audience. In the
Ingroup condition the avatars were physically smaller,
and they wore bright colours and informal clothes.
The avatars in the outgroup condition were bigger and
wore toned down colours and formal clothes (see fig-
ure 1).
(a) The Ingroup condition
(b) The Outgroup condition
Figure 1: The virtual characters.
The environment was presented to the partici-
pants through the Oculus Rift head-mounted display
(HMD), tracking only the head movements. Thus par-
ticipants were able to look around the room but unable
to further interact physically with objects in the envi-
ronment. Furthermore, a Grove ear-clip optic pulse
sensor by Seeedstudio was used to measure the heart
rate of the participants.
2.3 Measurements
2.3.1 Public Speaking Anxiety Scale
To measure subjective PSA, the Public Speaking
Anxiety Scale (PSAS) (Bartholomay and Houlihan,
2016) was used, consisting of 17 items on a 7 point
Likert scale. Before the VR experience, the partic-
ipants’ general predisposition towards PSA (general
PSA) was measured with the PSAS. After the pre-
sentation in VR, state PSA during the experience was
measured using the altered version of the PSAS ques-
tionnaire specifically targeted towards the recent pub-
lic speaking experience (state PSA) - so instead of
’I am confident when I give a speech’ the statement
reads ’I was confident when I gave the speech’.
2.3.2 Heart Rate
For an objective measure of PSA, mean heart rate
(HR) and heart rate variability (HRV) were collected,
both as a baseline and during the public speaking task.
These are both indicators of stress (Kim et al., 2018)
and complement each other (Kusserow et al., 2008).
HR is a general measure of arousal, whereas HRV
is linked to anxiety and emotional strain specifically.
The Effects of Ingroup Bias on Public Speaking Anxiety in Virtual Reality
19
Although both measures can of course respond to the
general VR experience also (especially in the case of
a novelty effect), we only consider the differences be-
tween conditions and not the raw values. Both HR
and HRV were derived from the normal-to-normal in-
terval of consecutive heart beats, with the variance
of the inter beat intervals (functionally, the SDNN)
used as the measure for HRV. To ensure that HR and
HRV were obtained from a stable signal, the first and
last ten heartbeats of each recording were excluded
from the analysis. Skin conductance was considered
as a similar measurement but proved less reliable and
more restrictive in the pilots.
2.3.3 Emotional Intelligence
For this study we focused on trait EI, referring
to emotion-related trait and self-perceived abilities
(Petrides and Furnham, 2006). To measure trait EI,
the Trait Emotional Intelligence Questionnaire - Short
Form (Petrides, 2009) was used. This questionnaire
consists of 30 items on a 7 point Likert scale, giving
a global trait EI score.
2.3.4 Ingroup Identification and Presence
As a measure of the effectiveness of changing the VR
avatars for the Ingroup/Outgroup manipulation, the
actual ingroup identification was measured with the
Group Identification Scale (GIS) (Sani et al., 2015),
which consists of 4 items on a 7 point Likert scale.
Additionally, the short version of the Presence Ques-
tionnaire by Witmer and Singer (Witmer and Singer,
1998) was used to control for the level of presence ex-
perienced in the VR environment. This version con-
sists of 6 items on a 7 point Likert scale.
2.4 Procedure
The experiments were conducted in a neutral office
room. Participants were welcomed and read the in-
formation letter. After signing the informed consent,
they filled in the pretest questionnaire (demograph-
ics, English proficiency and VR experience questions,
general PSA, and EI). A baseline heart rate recording
was made for one minute while the participant sat still
and looked at a blank wall.
Then participants put on the HMD and got ac-
customed to the hardware and VR by experiencing
the virtual room without the audience in it for one
minute. They then took off the HMD and were in-
structed about the task. They were told that the exper-
imenter would leave the room and that the presenta-
tion would not be recorded. Following that, the par-
Figure 2: State PSA, GIS and Presence per condition.
Figure 3: HR and HRV increase per condition.
ticipants put on their HMD again and the heart rate
sensor started recording the speaking session.
After two minutes had passed, the experimenter
re-entered the room and announced that the partic-
ipant could stop talking and take off the HMD and
heart rate sensor. The participants then filled out
the post-test questionnaire (state PSA, GIS, and pres-
ence). Finally, the participants were thanked for their
participation and the experiment was ended.
3 RESULTS
3.1 Public Speaking Anxiety
3.1.1 State PSA
The post-test PSAS (to measure state PSA during
the VR experience) had good scale reliability (Cron-
bach’s α = 0.88 ). As shown in Figure 2, participants
in the Ingroup condition (M = 2.39, SD = 0.87) re-
ported slightly less PSA during the speech task than
participants in the Outgroup condition (M = 2.72, SD
= 0.74). The Levene’s test showed that the variances
in the two groups were homogeneous, F(1,19) = 0.71,
p = 0.46. The difference between the two conditions
was not significant, Mdif = 0.32, t(38) = 1.25, p =
0.22.
HUCAPP 2020 - 4th International Conference on Human Computer Interaction Theory and Applications
20
3.1.2 Heart Rate
Due to malfunctioning of the heart rate sensor (>15%
of measurements compromised in either baseline
or speaking measurement), eight participants (1 In-
group, 7 Outgroup) were removed from the analy-
sis of HR data. There was a significant general in-
crease in mean HR from the baseline to the public
speaking phase for all participants, regardless of con-
dition (Mdif = -17.41, t(31) = -6.38, p < 0.001). For
each participant, the difference between the baseline
and the speaking HR and the difference between the
baseline HRV and speaking HRV were calculated and
used for further statistical testing.
Participants in the Outgroup condition (M = 19.55,
SD = 18.10) had a higher mean HR increase than par-
ticipants in the Ingroup condition (M = 15.94, SD =
13.65), as can be seen in Figure 3. The variances in
the two groups were homogeneous, F(1,12) = 1.76,
p = 0.27. The difference between the two conditions
was not significant, Mdif = 3.61, t(30) = 0.64, p =
0.52.
Participants in the Ingroup condition (M = 358.11,
SD = 2445.14) had a lower increase of HRV during
the speaking task than participants in the Outgroup
condition (M = 637.71, SD = 1891.60), as can be seen
in Figure 3. The variances in the two groups were ho-
mogeneous, F(1,12) = 0.60, p = 0.37. The difference
between the two conditions was not significant, Mdif
= 279.60, t(30) = 0.35, p = 0.73.
3.2 Individual Traits
3.2.1 General PSA
The pre-test PSAS (to measure general PSA) had
good scale reliability (Cronbach’s α = 0.94). Partici-
pants in the Ingroup condition (M = 3.42, SD = 1.05)
reported slightly less general PSA than participants in
the Outgroup condition (M = 3.84, SD = 0.81), but
this difference was not significant. Because the level
of general PSA differed greatly (M = 3.63, SD = 1.33)
within all participants and cutoff scores are generally
used to categorize PSA (Bartholomay and Houlihan,
2016), we first split them into two groups, one with
low general PSA and one with high general PSA, with
a cutoff point of 4. The low general PSA group (10
Ingroup condition, 9 Outgroup condition) had an av-
erage of 3.05 (SD = 0.76), and the high general PSA
group had an average of 4.51 (SD = 0.31).
On average, participants with low general PSA
also experienced less state PSA during the experi-
ment (M = 2.17, SD = 0.63) than participants with
high general PSA (M = 3.13, SD = 0.72). Levene’s
test showed that the variances in the two groups were
homogeneous, F(1,36) = 1.04, p = 0.39. The differ-
ence in state PSA between the two groups was signif-
icant (two-way ANOVA, controlling for condition),
F(1,36) = 18.69, p < 0.001, with a large-sized effect
of d = 1.4. However, there was no significant interac-
tion effect with the Ingroup/Outgroup conditions.
3.2.2 Regression Model - EI & General PSA
The EI demonstrated good internal consistency relia-
bility (Cronbach’s α = .88). Participants in the Out-
group condition (M = 5.01, SD = 0.58) reported a
slightly lower mean than those in the Ingroup con-
dition (M = 5.26, SD = 0.66).
A moderated regression analysis was performed to
examine further whether the effects of EI and general
PSA (non-categorized values) on state PSA differed
by condition. The model summary for the regression
analysis was as follows: F(5, 34) = 9.84, p < .001,
R
2
= .59. Condition significantly moderated both the
effect of EI on state PSA and the effect of general PSA
on state PSA (see Table 1).
A post-hoc simple slopes analysis found a signif-
icant effect of EI on state PSA, after controlling for
general PSA, in the Outgroup condition (β = -.90, SE
= .30, t = -3.04, p = .005) but not in the Ingroup condi-
tion (β = .08, SE = .26, t = .30, p = .76). Conversely,
the simple slope of general PSA on state PSA, after
controlling for EI, was significant in the Ingroup con-
dition (β = .68, SE = .16, t = 4.21, p < .001) but not
in the Outgroup condition (β = .10, SE = .21, t = .48,
p = .64).
Table 1: Moderated regression model - the effect of individ-
ual traits EI and general PSA on state PSA.
β SE t p
gPSA .10 .21 .48 .637
Condition -.11 .19 -.58 .566
EI -.90 .30 -3.04 .005
EI*Condition .98 .40 2.47 .019
gPSA*Condition .58 .26 2.21 .034
3.3 VR Ecological Validity
3.3.1 GIS
The GIS had good scale reliability (Cronbach’s α =
0.91 ). As shown in Figure 2, participants in the
Ingroup condition (M = 3.33, SD = 1.54) identified
more with their virtual audience than participants in
the Outgroup condition (M = 2.45, SD = 0.94). Lev-
ene’s test showed that the variances in the two groups
were not homogeneous, F(1,19) = 0.37, p = 0.04. The
The Effects of Ingroup Bias on Public Speaking Anxiety in Virtual Reality
21
difference between the two conditions was significant
(independent samples t-test), Mdif = -0.88, t(31.4) =
-2.17 p = 0.04, with a moderate effect size of d = -
0.67.
3.3.2 Presence
On average, the level of presence experienced by the
participants was 5.2 (SD = 1.0). The presence ques-
tionnaire had sufficient scale reliability (Cronbach’s
α = 0.68). As shown in Figure 2, participants in the
Ingroup condition (M = 5.23, SD = 0.68) felt more
present than participants in the Outgroup condition
(M = 4.36, SD = 1.04). Levene’s test showed that
the variances in the two groups were homogeneous,
F(1,19) = 2.36, p = 0.07. The difference between the
two conditions was significant (independent samples
t-test), Mdif = -0.87, t(38) = -3.3, p = 0.002, with a
large-sized effect, d = -1.05.
3.3.3 Regression Model - GIS & Presence
A moderated regression analysis was performed to ex-
amine whether the effect of participants’ level of GIS
and presence on state PSA differed by condition. The
model summary for the regression analysis was as fol-
lows: F(5, 34) = 1.22, p = .32, R
2
= .85. Neither the
GIS nor the presence effects were significantly mod-
erated by condition (see Table 2). The model was re-
run without the interaction terms to allow estimation
of the partial effects of GIS and presence. Reducing
the model did not significantly decrease the propor-
tion of explained variation (R
2
= .02, F(2, 36) =
4.34, p = .65), but neither the partial effect of GIS nor
the partial effect of presence were statistically signifi-
cant (GIS: β = .20, SE = .12, t = 1.75, p = .089; Pres-
ence: β = -.19, SE = .17, t = -1.10, p = .28).
Table 2: Moderated regression model - the effect of VR
ecological validity GIS and presence on state PSA.
β SE t p
GIS .04 .25 .14 .888
Presence -.05 .23 -.21 .837
Condition -.33 .32 -1.05 .299
GIS*Condition .22 .28 .77 .447
Presence*Condition -.30 .38 -.80 .430
4 DISCUSSION
This study examined the effect of ingroup bias on
the level of public speaking anxiety that users expe-
rienced in VR, and how their personal characteristics
and the VR ecological validity regulated that experi-
ence. Our results showed that the Ingroup/Outgroup
conditions did not significantly affect the level of PSA
that participants experienced during the public speak-
ing task, answering RQ1 in the negative. The eco-
logical validity did not seem to influence PSA either
(RQ3). However, there were significant effects of per-
sonal traits such as emotional intelligence and general
public speaking anxiety (RQ2).
There was an interaction between EI and condi-
tion, where a higher EI predicted a lower state PSA
score, specifically in the Outgroup condition. This re-
sult suggests that EI had a modulating effect on state
PSA when presenting to an outgroup audience, but
not when presenting to an ingroup audience. It could
be interesting to explore this effect further in future
research, to see whether a high emotional intelligence
allows a person to be more comfortable with virtual
characters that they do not necessarily identify with.
If this is the case, virtual public speaking training
could be improved by adding emotional intelligence
training as well.
Not unexpectedly, state PSA was significantly af-
fected by general PSA. This means that participants
who experience more anxiety towards public speak-
ing in general were also more anxious during this
specific virtual public speaking experience, regardless
of the condition. The regression showed that there
was also an interaction effect with condition overall,
meaning that general PSA predicted state PSA in the
Ingroup condition. The more predisposed someone is
to public speaking anxiety, the more anxious they are
when they have to present in front of an audience that
they consider a part of their ingroup, whereas gen-
eral PSA was not significantly related to state anxiety
when presenting to outgroup members.
Besides the main research questions, a few other
noteworthy interactions were found. A significant in-
crease in heart rate was observed overall. This could
be caused by the novelty effect of the VR technology,
as many participants did not have a lot of previous
experience with VR. The 1-minute acclimatization to
the VR environment might not have been enough to
overcome this. If this is the case, the novelty effect
might have occluded any smaller effects of the con-
ditions, and it could be interesting to repeat the ex-
periment with more experienced participants or over
a longer period of time.
Participants in the Ingroup condition identified
significantly more with the virtual audience than par-
ticipants in the Outgroup condition. This is exactly as
intended, and indicates that it is likely that a type of
ingroup bias was triggered by the exposure to these
different audiences. Unexpectedly, the audience type
had a significant effect on presence, where partici-
pants in the Ingroup condition felt more present in
HUCAPP 2020 - 4th International Conference on Human Computer Interaction Theory and Applications
22
the virtual environment than those in the Outgroup
condition. This warrants further research, as it could
potentially be utilized in a wide variety of virtual ap-
plications to improve the user interaction. Although
it of course already makes sense from a narrative de-
sign point of view to improve user identification with
the virtual characters they encounter, this result shows
that it might have measurable and far-reaching effects
on how the medium itself is experienced.
The ingroup audience does not seem to lead to
enough of a positive association with the virtual char-
acters to significantly affect PSA. It is however not
unlikely that the increased presence and group identi-
fication have effects on other factors, which could be
explored in future research. As several studies have
shown, the valence of audience feedback affects the
user (Slater et al., 1999; Pertaub et al., 2002). Interac-
tion effects in this regard might be a worthwhile area
of study, since positive or negative audience feedback
could be perceived differently depending on whether
the virtual audience is a part of the participant’s in-
group or outgroup (Elder et al., 2005).
It should be noted that, as increased identification
with the ingroup audience was established not only
through the use of visuals, but also through means
of a narrative (first-year students who want to know
what to expect from their studies), it is possible that
this by itself explains the increased level of presence
(Troxler et al., 2018). As previous research shows that
co-presence in particular can amplify positive effects
of virtual audiences (Pertaub et al., 2002), it might be
interesting for future research to look at co-presence
specifically. Co-presence could also be induced more
strongly with several techniques, such as social prim-
ing (Daher et al., 2017).
One of the limitations of this study is that we
only looked at one type of in- and outgroup, namely
students and professors. As is often the case, these
groups hold a clear power differential in regard to how
they relate to the participant. This could have influ-
enced the results. Different types of groupings (for
instance on gender, race or social class) could poten-
tially lead to different effects. Additionally, although
many VR studies currently use similar or even smaller
sample sizes to the one used here, in-depth analyses
of the interaction effects require a more robust sample
size which could lead to deeper insights in the com-
plex mechanisms behind social VR experiences.
As the technology and its applications advance,
VR involves more of the interacting variables that are
also present in real life, especially on a social level
(Hamilton et al., 2016). Uncovering the social pro-
cesses that govern the user interaction with the vir-
tual world allows us to design more effective VR ap-
plications. This study has shown that while ingroup
versus outgroup virtual audiences can affect group
identification, which can significantly influence pres-
ence, this does not have a significant influence on the
level of PSA that is experienced by users while prac-
tising public speaking in VR. It does however affect
the relationship between general and state PSA. Fur-
thermore, the effects on state PSA are moderated by
the individual characteristics (EI) of the player, poten-
tially allowing for personalization of the experience.
The overall results of this study indicate that ingroup
bias can be evoked in virtual environments and that
it could be an interesting method of influencing the
feelings and behaviour of users in specific ways. Al-
though this study focused on PSA, the effects of in-
group bias in VR could potentially be extended to-
wards any VR application that involves social interac-
tions, and might prove useful in a variety of contexts.
ACKNOWLEDGEMENTS
The authors would like to thank Lisa Neve for her
technical assistance with the creation of the virtual en-
vironment used in this study.
REFERENCES
Anderson, P. L., Zimand, E., Hodges, L. F., and Rothbaum,
B. O. (2005). Cognitive behavioral therapy for public-
speaking anxiety using virtual reality for exposure.
Depression and anxiety, 22(3):156–158.
Bartholomay, E. M. and Houlihan, D. D. (2016). Pub-
lic speaking anxiety scale: Preliminary psychometric
data and scale validation. Personality and Individual
Differences, 94:211–215.
Bodie, G. D. (2010). A racing heart, rattling knees, and ru-
minative thoughts: Defining, explaining, and treating
public speaking anxiety. Communication education,
59(1):70–105.
Botella, C., Gallego, M. J., Garcia-Palacios, A., Guillen,
V., Ba
˜
nos, R. M., Quero, S., and Alca
˜
niz, M. (2010).
An internet-based self-help treatment for fear of pub-
lic speaking: a controlled trial. Cyberpsychology, Be-
havior, and Social Networking, 13(4):407–421.
Daher, S., Kim, K., Lee, M., Schubert, R., Bruder, G.,
Bailenson, J., and Welch, G. (2017). Effects of so-
cial priming on social presence with intelligent virtual
agents. In International Conference on Intelligent Vir-
tual Agents, pages 87–100. Springer.
Dewaele, J.-M., Petrides, K. V., and Furnham, A. (2008).
Effects of trait emotional intelligence and sociobio-
graphical variables on communicative anxiety and for-
eign language anxiety among adult multilinguals: A
review and empirical investigation. Language Learn-
ing, 58(4):911–960.
The Effects of Ingroup Bias on Public Speaking Anxiety in Virtual Reality
23
Elder, T. J., Sutton, R. M., and Douglas, K. M. (2005).
Keeping it to ourselves: Effects of audience size and
composition on reactions to criticisms of the ingroup.
Group Processes & Intergroup Relations, 8(3):231–
244.
Guadagno, R. E., Blascovich, J., Bailenson, J. N., and Mc-
Call, C. (2007). Virtual humans and persuasion: The
effects of agency and behavioral realism. Media Psy-
chology, 10(1):1–22.
Hamilton, A., Pan, X. S., Forbes, P., and Hale, J. (2016).
Using virtual characters to study human social cogni-
tion. In International Conference on Intelligent Vir-
tual Agents, pages 494–499. Springer.
Harris, S. R., Kemmerling, R. L., and North, M. M. (2002).
Brief virtual reality therapy for public speaking anxi-
ety. Cyberpsychology & behavior, 5(6):543–550.
Jacobs, M., Snow, J., Geraci, M., Vythilingam, M., Blair,
R., Charney, D. S., Pine, D. S., and Blair, K. S. (2008).
Association between level of emotional intelligence
and severity of anxiety in generalized social phobia.
Journal of anxiety disorders, 22(8):1487–1495.
Kim, H.-G., Cheon, E.-J., Bai, D.-S., Lee, Y. H., and Koo,
B.-H. (2018). Stress and heart rate variability: A meta-
analysis and review of the literature. Psychiatry inves-
tigation, 15(3):235.
Kusserow, M., Amft, O., and Tr
¨
oster, G. (2008). Analy-
sis of heart stress response for a public talk assistant
system. In European Conference on Ambient Intelli-
gence, pages 326–342. Springer.
Kyriakou, M. and Chrysanthou, Y. (2018). How responsive-
ness, group membership and gender affect the feeling
of presence in immersive virtual environments popu-
lated with virtual crowds. In Proceedings of the 11th
Annual International Conference on Motion, Interac-
tion, and Games, page 12. ACM.
Lindner, P., Miloff, A., Fagern
¨
as, S., Andersen, J., Sige-
man, M., Andersson, G., Furmark, T., and Carlbring,
P. (2019). Therapist-led and self-led one-session vir-
tual reality exposure therapy for public speaking anx-
iety with consumer hardware and software: A ran-
domized controlled trial. Journal of anxiety disorders,
61:45–54.
Nowak, K. L., Hamilton, M. A., and Hammond, C. C.
(2009). The effect of image features on judgments of
homophily, credibility, and intention to use as avatars
in future interactions. Media Psychology, 12(1):50–
76.
Nowak, K. L. and Rauh, C. (2005). The influence of
the avatar on online perceptions of anthropomor-
phism, androgyny, credibility, homophily, and attrac-
tion. Journal of Computer-Mediated Communication,
11(1):153–178.
Paladino, M.-P. and Castelli, L. (2008). On the immediate
consequences of intergroup categorization: Activation
of approach and avoidance motor behavior toward in-
group and outgroup members. Personality and Social
Psychology Bulletin, 34(6):755–768.
Pertaub, D.-P., Slater, M., and Barker, C. (2002). An exper-
iment on public speaking anxiety in response to three
different types of virtual audience. Presence: Teleop-
erators & Virtual Environments, 11(1):68–78.
Petrides, K. V. (2009). Psychometric properties of the
trait emotional intelligence questionnaire (teique).
In Assessing emotional intelligence, pages 85–101.
Springer.
Petrides, K. V. and Furnham, A. (2006). The role of trait
emotional intelligence in a gender-specific model of
organizational variables 1. Journal of Applied Social
Psychology, 36(2):552–569.
Pull, C. B. (2012). Current status of knowledge on public-
speaking anxiety. Current opinion in psychiatry,
25(1):32–38.
Salovey, P. E. and Sluyter, D. J. (1997). Emotional develop-
ment and emotional intelligence: Educational impli-
cations. Basic Books.
Sani, F., Madhok, V., Norbury, M., Dugard, P., and Wake-
field, J. R. (2015). Greater number of group identifica-
tions is associated with healthier behaviour: Evidence
from a scottish community sample. British Journal of
Health Psychology, 20(3):466–481.
Slater, M., Pertaub, D.-P., and Steed, A. (1999). Pub-
lic speaking in virtual reality: Facing an audience of
avatars. IEEE Computer Graphics and Applications,
19(2):6–9.
Stupar-Rutenfrans, S., Ketelaars, L. E., and van Gisber-
gen, M. S. (2017). Beat the fear of public speaking:
Mobile 360 video virtual reality exposure training in
home environment reduces public speaking anxiety.
Cyberpsychology, Behavior, and Social Networking,
20(10):624–633.
Summerfeldt, L. J., Kloosterman, P. H., Antony, M. M., and
Parker, J. D. (2006). Social anxiety, emotional intelli-
gence, and interpersonal adjustment. Journal of Psy-
chopathology and Behavioral Assessment, 28(1):57–
68.
Troxler, M., Qurashi, S., Tjon, D., Gao, H., Rombout, L.,
and Nalepa, G. (2018). The virtual hero: The influ-
ence of narrative on affect and presence in a vr game.
In CEUR Workshop Proceedings.
Vinayagamoorthy, V., Gillies, M., Steed, A., Tanguy, E.,
Pan, X., Loscos, C., and Slater, M. (2006). Building
expression into virtual characters building expression
into virtual characters. The Eurographics Association,
pages 1–42.
Wallach, H. S., Safir, M. P., and Bar-Zvi, M. (2009). Virtual
reality cognitive behavior therapy for public speaking
anxiety: a randomized clinical trial. Behavior modifi-
cation, 33(3):314–338.
Witmer, B. G. and Singer, M. J. (1998). Measuring pres-
ence in virtual environments: A presence question-
naire. Presence, 7(3):225–240.
Wittenbrink, B., Judd, C. M., and Park, B. (1997). Evi-
dence for racial prejudice at the implicit level and its
relationship with questionnaire measures. Journal of
personality and social psychology, 72(2):262.
HUCAPP 2020 - 4th International Conference on Human Computer Interaction Theory and Applications
24