The authors have declared that no competing interests exist.
Conceived and designed the experiments: CKP MJ J. Bailey J. Bailenson CBT. Performed the experiments: CKP. Analyzed the data: CKP J. Bailey CBT. Contributed reagents/materials/analysis tools: CKP MJ J. Bailey J. Bailenson CBT. Wrote the paper: CKP MJ J. Bailey J. Bailenson CBT.
Body image disturbance (BID), considered a key feature in eating disorders, is a pervasive issue among young women. Accurate assessment of BID is critical, but the field is currently limited to self-report assessment methods. In the present study, we build upon existing research, and explore the utility of virtual reality (VR) to elicit and detect changes in BID across various immersive virtual environments. College-aged women with elevated weight and shape concerns (
Body image disturbance (BID) refers to negative self-evaluations of one’s physical appearance which lead to significant distress or impairment [
While BID has been studied widely, certain important features of this phenomenon remain poorly understood. This is partly the result of existing assessment techniques, which limit the scope of BID research by focusing on BID main as a stable attitude, or trait; few measures assess the degree to which one’s body image fluctuates at a given moment in time [
Recent research suggests that automatic cognitive and behavioral processes are highly involved in the state BD; however, traditional self-report measures are unable to assess automatic processes because they originate outside of conscious awareness [
Virtual reality (VR) offers tools to address limitations in BID assessment. VR has gained popularity in psychological research because it provides researchers excellent environmental control, high realism, and unparalleled control over assessment timing [
VR has been applied to BID assessment and intervention in past studies, and the present research aims to build on prior work in several important ways. First, VR studies have shown that individuals with ED exhibit strong affect responses to disorder-relevant virtual environments [
The present study presents a novel paradigm for measuring BID in VR. First, we propose to validate a series of virtual environments that are relevant to BID among women without eating pathology. The environments used in existing VR exposure protocols for ED patients were developed by administering traditional measures in the context of a virtual environment. We propose to use a similar method in which we determine the extent to which one’s experience in VR mirrors that expected in similar real–world situations. Observational, theoretical, and experimental studies suggest that situations in which one’s body is highly salient and/or in which one engages in body-based social comparison are highly relevant to BID [
Second, we aim to investigate the extent to which automatic nonverbal social behaviors are related to BID. VR allows us to explore visual attention in a realistic setting, whereas prior studies have been conducted in highly controlled laboratory settings by measuring attention toward static images, often presented on a computer screen. Visual gaze is easily measured in VR by deriving a user’s field of view as a function of her head movements (Bailenson et al., 2002). By measuring visual attention in virtual social contexts, we can determine whether attentional biases demonstrated in laboratory studies persist in situations in naturalistic settings, in which many stimuli (in addition to body-related cues) compete for one’s attention.
We further aim to extend previous work in automatic processes related to BID by exploring interpersonal approach behavior, which we believe may relate to BID. Specifically, previous VR studies in a variety of disciplines have shown interpersonal distance to be a sensitive indicator of implicit attitudes and evaluations [
Hypothesis I: Participants with greater weight-and-shape concerns will report higher levels of state BD across all environments compared to participants with low levels of weight-and-shape concerns.
Hypothesis II: Participants will report the most state BD in a VR environment with high body salience and social presence, compared to other environmental conditions.
Hypothesis III: Participants will vary their behavior toward female avatars based on their pre-existing levels of weight-and-shape concerns. Specifically, participants with elevated weight-and-shape concerns will approach more closely and demonstrate more visual attention toward thin avatars as compared to participants without elevated weight-and-shape concerns.
A convenience sample was recruited using the student research participant pool of the Stanford University Communication Department during the 2012–2013 Winter and Spring quarters and the 2013–2014 Fall quarter. Students received course credit for participation in the study. All students in available courses completed an online prescreen questionnaire at the beginning of the quarter to determine their eligibility. The Weight Concerns Scale (WCS) was included in the online questionnaire to assess level of body image disturbance. WCS scores greater than 47 are associated with greater BID overall, and have been used to indicate increased risk for developing an eating disorder [
A total of 112 students responded to an invitation to participate in the study. Of those 112 students, 103 students provided written informed consent and completed baseline assessments. Of the 103 students who completed baseline assessments, 24 were excluded from final analysis for the following reasons: (a) failure to meet inclusion criteria (
At baseline, participants were screened and sorted into two groups on the basis of WCS total scores: elevated overall BID (referred to as the “weight-concerned” group;
All procedures and materials were approved by the Ethical Committee of the Institutional Review Board at Stanford University. Written consent was obtained from all participants; the individual pictured in this manuscript, who was not a study participant, has given written informed consent (as outlined in PLOS consent form) for his image to be used.
The lead researcher and one undergraduate research assistant were present for each trial. Upon arriving at the laboratory, all participants signed a consent form. Participants completed computer-based questionnaires assessing baseline weight-and-shape concerns, eating pathology, body satisfaction, and physical appearance comparison.
The researcher provided all participants with a verbal description of the purpose of the study explaining the components of the VR experience. Participants were told that they would be participating in a study examining whether certain virtual environments influenced women’s awareness of their own bodies and their behavior towards others. The researcher then assisted each participant in wearing the virtual reality headset. After completing the VR task, all participants received debriefing materials containing information about online resources for body image improvement and contact information for university counseling and psychological services.
The environments for the VR assessment task were designed to elicit varying levels of state BD. We used a 2x2 design (high vs. low) to vary the degree of (a) body salience and (b) social presence, resulting in four distinct environments. These factors were chosen based on previous research on state body dissatisfaction (state BD) in nonclinical samples using both traditional methods and VR exposure [
In order to confirm that equipment was functioning properly, all participants were first exposed to a virtual training environment designed to mirror the physical room in which the study took place. In the training room, the participants practiced walking while wearing the equipment to gain familiarity with the experience of being immersed in the virtual environment.
After the initial training room, the participants were randomly exposed to the four virtual experimental conditions to control for order effects. All participants were exposed to an empty beach, a beach populated with avatars, an empty party scene, and a party scene populated with avatars. Populated environments contained three groups of three avatars, each composed of two female and one male avatar.
The female avatars varied according to body size between the three groups: underweight, healthy weight, and overweight body sizes were represented (the male avatar remained the same). The groups were designed to represent a BMI of approximately 18 (underweight), 23 (healthy weight), and 27.5 (overweight) respectively. The groups stood in fixed positions within virtual space, equidistant from the participant’s starting position. The placement of each group relative to the participant’s starting position was counterbalanced within and between participants to control for environmental and individual factors that may influence approach behavior.
In the environments containing avatars, participants viewed all three groups from their starting position and then approached (“joined”) each group in the order of their preference. Participants spent 15–20 seconds observing each group. In environments without avatars, participants approached three areas of interest in the environment and observed each area for 15–20 seconds.
In the interval between exposure to each experimental environment, participants completed a measure of state BD, by reading the items on a virtual screen projected in the virtual reality headset and speaking aloud the number on the scale corresponding to their answer for each item. The research assistant was blinded to the scale items and audio-recorded participants’ responses. Responses were also hand-recorded by the researcher. Immediately following exposure to the final environment and completing the final measurement of state BD, the research assistant removed the virtual reality headset and the participant completed post-test measures.
Participants viewed the virtual environments using a head-mounted display (HMD), a virtual reality headset that provided three-dimensional stereoscopic views. They wore a nVisor SX111 HMD (NVIS, Reston, VA) with a resolution of 2560 x 1024 and a refresh rate of 120 frames per second (60 seconds per eye), enabling them to have naturalistic head movements while navigating the virtual worlds. An Intersense3 Cube accelerometer tracked their physical head movements (pitch, roll, and yaw) operating at 180 hertz with a 4-millisecond latency rate. In addition, participant’s gross body position on x-, y-, z-axis was tracked using an optical infrared camera system (Worldviz PPT-H) operating at 180 hertz with a 20-millisecond latency rate (with a precision of 0.25 millimeters). The optical tracking system allowed participants to navigate the virtual environment by walking naturally.
(1) An accelerometer allowed for naturalistic head movements during virtual exposure and (2) an optical infrared camera system recorded subjects’ gross body position in virtual space, allowing subjects to navigate the virtual environments via natural walking patterns.
Researchers assessed weight-and-shape concerns using the five-item Weight Concerns Scale (WCS) [
Self-reported height and weight was recorded at baseline to calculate approximate BMI [BMI = weight(kg)/height(m)^2].
The Eating Disorder Examination-Questionnaire (EDE-Q), a 39-item self-report version of the Eating Disorder Examination (EDE), is a well-established investigator-based interview used to diagnose and assess eating pathology [
State BD was measured using the Body Parts Satisfaction Scale–Revised (BPSS-R) [
Throughout virtual exposure, the VR tracking system collected data on participants x, y and z positions in the virtual environment at a rate of 60 frames per second, measuring participant’s position relative to the virtual objects in meters. A participant’s “minimum distance” was operationalized as the smallest absolute distance reached between a participant and each target avatar group. When calculating the distance between the participant and an avatar, analysis controlled for the relative body size of the avatar. First, the width of the avatar was calculated using the “Inspector” program of Vizard (WorldViz, Santa Barbara, 2012). In the analysis, the avatar was then considered to be a cylinder with a radius equal to the avatar’s calculated width, and the minimum absolute interpersonal distance was considered to be the smallest distance from the center of the participant to the outside edge of this cylinder, as shown in
The participant’s movement from her starting point at T1 to her idling position at T2 is recorded by the optical tracking system in the laboratory room. At T2, line C indicates the participant’s entire horizontal field of view of 1020 within the HMD. Avatar A is within the participant’s range of view, whereas Avatar B is not in view. A vector (line A) was extended from the center of the participant's head along the z-axis (i.e., extending out from the nose), and another vector (line D) was drawn between the participant and Avatar A, providing a measurement for angle E. The participant is considered to be looking at a given avatar if angle E is less than half of her entire field of view, or less than 510. Line B demonstrates the distance between the participant at T2 and Avatar B, measured as the distance from the participant to the outside edge of a cylinder with a radius equal to the width of Avatar B, as measured in Inspector.
Visual gaze was measured as the total percentage of time within each populated virtual environment that the participant spent looking at each avatar group. Participants’ head movement (yaw, rotation around the y-axis) was used to determine if they were looking at a group of avatars at a given time point (60 frames per second). Head direction, as opposed to eye movement, was used to indicate visual gaze. The technical challenges of measuring eye movement in immersive VR are significant, however head direction is very rarely decoupled from eye movement during VR exposure, and head direction is a more socially meaningful indicator of gaze than eye movement [
We determined which avatar group each subject chose to approach first in each populated environment by calculating the direction of the subject’s initial movement in each environment relative to the positions of each avatar group. This was also hand recorded by the researcher while observing the participant’s movement (i.e., the researcher noted which group the participant chose to join first).
All data analyses were performed in R [
Control ( |
At-risk ( |
|
---|---|---|
Age in years, |
20.07(1.47) | 20.67(2.52) |
Race/Ethnicity (%) | ||
White/European Descent | 50.00 | 48.64 |
African-American | 10.00 | 2.70 |
Asian-American/ Pacific Islander | 12.50 | 18.92 |
Latina/Hispanic | 12.50 | 13.51 |
Middle Eastern-American | 2.50 | 0.00 |
Other/ Prefer not to say | 12.50 | 16.22 |
BMI, |
22.19(2.45) | 22.70(1.96) |
Weight Concerns Scale Score, |
24.77 (12.09) | 61.70 (12.57) |
1 Due to missing data for one participant, age analysis is based on
Linear mixed-effects models were used to examine the association between BPSS-R scores, BID risk group, and virtual environment (
In examining the virtual scenes, there was a significant effect of environmental condition regardless of risk. Participants reported less body satisfaction in the populated beach scene compared to the empty beach scene (
Mean BPSS-R Body Subscale scores for both groups across all environments are presented in
Error bars represent 95% confidence intervals.
Linear mixed effects models were used to compare differences in participants’ minimum interpersonal distance from the avatars of different body sizes (average, overweight, and thin) in the two populated virtual environments (beach and party;
Distances are reported in meters.
Average Weight | Thin | Overweight | |
---|---|---|---|
Control | 0.54 (0.51–0.58) | 0.57 (0.53–0.60) | 0.60 (0.55–0.65) |
At-Risk | 0.59 (0.55–0.64) | 0.57 (0.52–0.61) | 0.59 (0.54–0.63) |
There was no significant main effects of BID risk group,
Contrasts applied to the model further examine the interaction effects between BID risk group and avatar body size. The first contrast compared the interpersonal distance of the average-sized avatars against both the overweight and thin sized avatars. The second contrast compared the overweight sized avatars with the thin sized avatar. There was no main effect of avatar size comparing the overweight and thin sized avatars to the average body size group,
There was an interaction effect between risk group and these contrasts. The Control group demonstrated differences in minimum differences according to avatar body size compared to the At-Risk group. The Control group stood significantly closer to the average sized avatars compared to both the overweight and thin sized avatars,
Linear mixed effects models were used to examine the effect of avatar body size and virtual scene on the percentage of time participants spent looking at the avatar groups (
Average Weight | Thin | Overweight | |
---|---|---|---|
Control | 27.52 (26.30–28.75) | 26.16 (25.17–27.15) | 26.45 (25.23–27.67) |
At-Risk | 26.04 (25.18–26.90) | 28.39 (27.26–29.51) | 25.28 (24.32–26.23) |
Using the average sized avatars as the comparison group, contrasts revealed additional interaction effects. The first contrast compared the average sized group against both the overweight and thin groups. The second contrasts compared the overweight avatar group with the thin sized avatar group. There was no significant difference in the amount of time that participants looked at the average sized avatars compared to both the overweight and thin avatars,
Two chi-square test of goodness-of-fit tests, were performed to determine whether the three avatar groups were equally preferred in terms of approach behavior in the two populated virtual environments. Preference for the groups was not equally distributed in the beach environment,
The present study investigated the use of VR in the study and assessment of BID. The primary aims of this study were (a) to identify environmental factors that relate to BID among women without eating pathology; and (b) to identify automatic behavioral indicators of BID. We found that all participants reported relatively stable state BD across a series of virtual environments, but greater state BD in a social environment with high body salience. With regard to behavior, we found that, compared to women without elevated weight and shape concerns, participants weight-and-shape concerns stood further from avatars of average weight and spent more time looking at thin female avatars than average or overweight female avatars. Finally, we found that all participants were reluctant to approach overweight female avatars.
Across all virtual environments, women with higher baseline weight-and-shape concerns reported lower body satisfaction than women in the control group. This was expected, and builds upon prior VR studies, which have shown that ED patients experience distress during exposure to virtual foods and eating environments [
Contrary to our hypotheses, we did not find that interpersonal distance or approach preference toward thin avatars differed depending on BID. Given that BID is characterized by thin-idealization, it is surprising that we failed to find between-group differences with regard to behavior toward a thin avatar. Previous VR research using personal space as an outcome measure in human-avatar interaction has generally concluded that smaller interpersonal distances reflect greater affiliation or sense of “intimacy” between the subject and the target avatar [
We further found that compared to the control group women with BID spent a greater percentage of time looking at thin avatars as compared to the other avatar groups. This is consistent with existing research findings on automatic attentional biases among women with BID, which has shown that women with BID demonstrate automatic attention toward thin female bodies. It is argued that such biases in attention and processing increase women’s feelings of body dissatisfaction [
Unexpectedly, we found that participants in the present study, regardless of condition, were reluctant to approach overweight avatars. Given that overweight avatars were identical to thin and normal weight avatars on all features except weight, it is reasonable to interpret this preference as largely driven by body size. Overweight and obese individuals are frequently the targets of social stigma and, particularly among adolescents, overt social exclusion [
The results of the present study suggest that immersive virtual environments may offer unique advantages to measure weight stigma at the behavioral level. Such methods may inform the development of more robust interventions to reduce stigmatized beliefs about overweight and obese people, which prove difficult to modify through more traditional, dissonance-based approaches [
The present study may have important implications for the development of BID treatments. For example, the current set of environments could be expanded to provide an array of salient practice situations, as has been done in virtual exposure-based treatment for specific phobias and PTSD [
Virtual spaces are increasingly common components of modern social life. Millions of people interact in virtual social spaces each day, some of which involve embodiment with an avatar such as online role-playing games [
First, our relatively homogenous sample may be a limitation of this study. Because participants were recruited only from a small number of courses within the Communications Department, the sample may not be representative of college women in general. The final sample had a high percentage of white participants, and a relatively small range in age and BMI, which may reduce the generalizability of these findings. For example, our sample contained very few women with BMI > 25, which may limit our ability to draw inferences about the influence of an individual’s weight status on behavior, state BD, or their approach behaviors toward overweight avatars. In addition the present study did not include direct measurement of participants’ stigmatizing beliefs or attitudes, which limits our ability to interpret participants’ avoidance of overweight avatars. Future studies should recruit higher weight participants in order to better understand how personal characteristics may influence approach behavior, and more comprehensive studies are needed in order to draw strong conclusions. Finally, while men are certainly affected by BID, research on male experiences of body image is limited. Existing research indicates that BID is expressed differently in men than in women, making it difficult to extend existing models of BID to a male population. Moreover, there is no research on attention biases related to BID using male participants. In order to allow us to draw clearer conclusions based on prior work with women, we did not include male participants in the present study. Future research extending models of BID to apply to men as well as women is clearly indicated.
The current results suggest that VR provides effective tools for measuring BID in realistic settings. The present study provides preliminary evidence for environmental factors that are relevant for BID among non-ED women, although further research is needed to strengthen this paradigm. This is the first study to use VR to investigate implicit behavioral indicators of state BD, which may be helpful in measuring treatment outcomes or designing future behavioral exposure treatments. Emerging models increasingly indicate that automatic processes play an important role in the etiology and maintenance of BID, thus it is vital that researchers establish effective approaches by which to measure and understand such processes. The present study introduces a novel measurement of automatic behavioral processes associated with BID, lends support to growing evidence of visual attention biases in BID, and provides preliminary evidence that BID is also associated with specific patterns of automatic social behavior that may maintain the disorder.
(ZIP)
(DOCX)
(DOCX)
(DOCX)
We would like to acknowledge the Stanford University Virtual Reality Intensive Training Seminar (VRITS) students of 2012–2013 and 2013–2014, a group of exceptional undergraduate programmers who contributed to the development of this study. In particular, we wish to recognize Helen Hastings, Joanna Kim, and Melvin Low and for their individual contributions to this project.