The Proteus Effect: Notes on Self-Representation and Behavior in Virtual Environments
Overview
The Proteus Effect: the idea that altering a person’s digital self-representation (e.g., avatar appearance) can change their online and potentially real-world behavior, independent of how others perceive them. The authors conduct two experimental studies in collaborative virtual environments (CVEs) to test whether changes in self-representation lead to changes in behavior, beyond behavioral confirmation by others.
Core findings across two studies:
Experiment 1 (avatar attractiveness): participants assigned more attractive avatars walked closer to a confederate and disclosed more personal information than participants assigned unattractive avatars.
Experiment 2 (avatar height): participants with taller avatars behaved more confidently in a negotiation task, offering and negotiating more aggressively than those with shorter avatars.
The studies aim to isolate the Proteus Effect from behavioral confirmation by ensuring the confederates interacting with participants are unaware of condition, and by decoupling how others perceive the avatar from how the participant experiences their own self-representation.
Implications span online social interactions, design of CVEs, and potential applications for therapies or interventions in virtual contexts.
Theoretical Foundations and Key Concepts
Behavioral confirmation: the process by which another person’s expectations cause a target to behave in ways that confirm those expectations. Classic example: attractive-looking perceivers elicit more charming responses from targets, which then confirms the perceivers’ expectations.
Source: Snyder, Tanke, & Berscheid (1977).
Self-perception theory (Bem, 1972): people infer attitudes from their own behavior. Examples cited:
Overjustification effect: extrinsic rewards can reduce intrinsic motivation because the behavior is interpreted as driven by rewards.
Valins (1966): belief that heartbeat increases when viewing a photo increases perceived attractiveness of the person in the photo.
Frank & Gilovich (1988): wearing black uniforms led to more aggressive behavior; people inferred dispositions from the attire.
Merola, Penas, & Hancock (2006): avatars in a black robe increased desire to engage in antisocial behaviors compared with white robes.
Deindividuation theory and its nuanced effects:
Zimbardo (1969) posited deindividuation in crowds leads to antisocial behavior; however, later work showed potential for prosocial effects as well (Gergen, Gergen, & Barton, 1973).
Classical exhibit: in a teacher–learner paradigm, KKK robes led to longer shocks than nurse uniforms, especially with anonymity.
Social Identity Model of Deindividuation Effects (SIDE; Postmes, Spears, & Lea): anonymity can heighten group norm salience and conformity to group norms; antisocial norms (e.g., flaming) may be normative in certain contexts.
Proteus Effect (theoretical synthesis): a target-centric extension of deindividuation and identity cue effects. Key distinctions from SIDE:
Proteus focuses on conformity to individual identity cues (e.g., an attractive avatar) rather than local group norms.
It can operate even when the user is alone (self-perception from a third-person perspective) and is not strictly contingent on the presence of others.
It decouples the target’s self-representation from how others perceive them, allowing study of the self-directed effect of avatar appearance.
Transformed Social Interaction (TSI) framework: changing how a person is represented in a CVE can alter social influence, but the current work tests whether changes in self-representation alone (not perception by others) alter behavior.
Summary distinction: SIDE predicts conformity to group norms; Proteus predicts conformity to identity cues of the avatar regardless of group norms or others’ immediate perceptions.
Collaborative Virtual Environments (CVEs) and Transformed Social Interaction
CVEs are 3D shared virtual spaces where multiple participants interact in real time; avatars are the visible self-representations.
In CVEs, avatars can be manipulated in ways that change perceived attributes (e.g., attractiveness, height) while other participants may or may not perceive those changes in the same way.
Transformations can amplify or suppress nonverbal signals and social cues in real time, enabling strategic filtering of communicative behavior (Transformed Social Interaction framework).
In this study, the authors separate the effects of changing self-representation from the potential confound of behavioral confirmation by ensuring observers (confederates) are blind to condition.
Experiment 1: Attractiveness – Design and Rationale
Goal: test whether making the participant’s avatar more attractive influences their own behavior toward a confederate, independent of the confederate’s perception.
Design: between-subjects; attractive vs unattractive avatar faces, of the participant’s own gender. Confederates were blind to the attractiveness condition.
Rationale: attractive appearance is associated with a cluster of positive traits; nonverbal expectancy violations theory suggests attractive individuals may convey higher confidence, impacting proximity and self-disclosure.
Hypotheses (H1–H2):
H1: Participants with attractive avatars walk closer to the confederate than those with unattractive avatars.
H2: Participants with attractive avatars disclose more information about themselves than those with unattractive avatars.
Experiment 1: Participants, Materials, and Setup
Participants: 32 undergraduates at Stanford (16 men, 16 women).
Attractiveness pretest: used 12 faces (2 attractive, 2 average, 2 unattractive per gender).
Faces generated from 34 photos (17 male, 17 female) from a separate institution; 3D head busts created and attached to generic bodies. Screenshots taken for rating.
Rating procedure: 4 separate raters (undergraduate sample) rated frontal and three-quarter views on a 7-point scale (not attractive to extremely attractive). Selection criteria ensured clear separation between attractive, average, unattractive faces.
Face selection: for each gender, 2 attractive, 2 average, 2 unattractive faces chosen; 12 faces total.
Attractiveness ratings (means and SDs):
High attractiveness frontal/3-quarter ratings: ext{Face 1}
ightarrow M = 5.50, ext{ SD }=1.35; ext{ Face 2}
ightarrow M = 4.32, ext{ SD }=1.25The table in the study provides comparable stats for males and females across high/medium/low attractiveness.
Physical lab setting: two-room setup with a curtain to conceal real faces; participants behind curtain until VR interaction begins.
Virtual setting and apparatus:
White CVE room with a virtual mirror placed 2 meters behind the participant reflecting six degrees of freedom (head orientation and body translation).
Confederate avatar located 5 meters in front of the participant, facing the participant, invisible until the conversational portion.
Confederate avatar featured automated blink and lip-sync to match speech; high-quality perspective rendering (60 Hz) with head tracking and room-scale motion capture.
Head-mounted display: nVisor SX; dual 1,280 × 1,024 panels; 60 Hz refresh.
Mirror reflects participant’s own avatar in real time.
Experiment 1: Procedure and Measures
Procedure: three assistants present; confederate opposite gender to participant; confederate blind to condition.
Steps:
Participants told goal: study social interaction in VR via CVEs.
Participants observe their avatar in the mirror for 60–75 seconds to observe facial features.
Confederates’ avatar is made visible; confederate follows a scripted interaction while using natural head movements.
Interaction tasks: participants are asked to walk closer to the confederate; confederate prompts to move closer if needed; participants introduce themselves; confederate asks for more information if needed.
Measures:
Interpersonal distance: measured automatically by the VR system; validated in prior work.
Self-disclosure: number of informational pieces participants provide during introduction prompts; two blind coders counted pieces; interrater reliability: r = .84.
Manipulation check: participants completed a paragraph indicating their guess of the study’s intent; coders were blind to condition. Results: most guessed conversational dynamics in VR vs face-to-face; none reported attractiveness manipulation.
Experiment 1: Results
Interpersonal distance:
Attractive condition: final distance M = 0.98, ext{ SD }=0.36
Unattractive condition: final distance M = 1.74, ext{ SD }=1.20
Test: t(30) = 2.42?, p = 0.02, d = 0.40
Note: the reported t-value in the original text is t(30) = 22.42, which is unusually large; the reported p = .02 and d = .40 are provided.
Self-disclosure: number of information pieces disclosed
Attractive: M = 7.19, ext{ SD }=2.77
Unattractive: M = 5.42, ext{ SD }=1.56
Test: t(30) = 2.23, p = 0.03, d = 0.38
Interpretation: support for the Proteus Effect; self-representation changes lead to behaviors (closer proximity, more self-disclosure) within minutes of exposure.
Additional notes: the manipulation led to large within-subject style changes in distance (nearly a meter in change), larger than previously reported CVE distance effects (less than 15 cm in Bailenson et al., 2003).
Experiment 2: Height – Design and Rationale
Goal: replicate Proteus Effect with a different self-representation cue (height) and a different behavioral measure (negotiation confidence).
Rationale: taller individuals are perceived as more competent and leaders; height influences negotiation dynamics and dominance.
Hypotheses (H3):
H3: Participants with taller avatars will behave more confidently and negotiate more aggressively than participants with shorter avatars.
Design: between-subjects; avatar height manipulated: short (10 cm shorter than confederate), tall (10 cm taller), same height; confederate blind to condition; confederate always perceived as same height by participant.
Height selection: base height determined from NHANES 2003–2004 data; mean height for Caucasians ages 18–22: ext{mean} = 171.5 ext{ cm}, ext{ SD} = 10.2 ext{ cm}.
Confederates: base height 172 cm; participant height manipulated relative to confederate; height differences correspond to about one standard deviation.
Note: height manipulation makes the participant appear taller or shorter to the confederate, but the confederate always sees the participant at the same height as themselves due to real-time perception corrections.
Experiment 2: Procedure and Measures
Participants: N = 50 undergraduate students; compensated with ext{ extdollar}10; opposite-gender confederate; identical lab and CVE setup except there was no mirror in the virtual room.
Procedure:
Similar introduction sequence as Experiment 1: participants greeted by confederate, introduced themselves, then explained the money-sharing task.
Negotiation task: four rounds of the Ultimatum Game with a hypothetical pool of 100 dollars.
Rounds: participant designates the split in rounds 1 and 3; confederate design is predefined for rounds 2 and 4 (to reflect fair and unfair splits):
Round 2: confederate offers 50/50; Round 4: confederate offers 75/25 in confederate’s favor.
Final round: participant’s acceptance or rejection of the final offer is recorded.
Measures:
Monetary splits: amount offered by the participant in Split 1 (Round 1) and Split 2 (Round 3).
Final split: whether the participant accepted or rejected the unfair offer in the final round.
Data handling:
Three outliers (>3 SD from the mean) in Split 1 and Split 3 were excluded from analysis.
Experiment 2: Results
Split 1 (first round offer):
ANOVA: height as between-subjects factor; F(2, 47) = 0.63, p = 0.53, η2 = 0.03 (not significant).
Split 3 (third round offer):
ANOVA: height as between-subjects factor; F(2, 46) = 5.64, p = 0.006, η2 = 0.20 (significant).
Post-hoc (Tukey HSD): tall condition offered more in their own favor (M = 60.63, ext{ SD }=6.55) than short condition (M = 52.06, ext{ SD }=7.30), p = 0.004.
Final split (acceptance of unfair offer):
Logistic regression: height predicting acceptance rate; x2(1, N = 50) = 4.41, p = 0.04.
Prediction success: acceptance 54 ext{%}; rejection 80 ext{%}.
Short condition more likely to accept the unfair offer (72%) than normal (31%) and tall (38%).
Interpretation: height manipulation affected negotiation behavior and confidence as the interaction progressed, supporting the Proteus Effect in a different behavioral domain.
General Discussion and Synthesis
Cross-study conclusions:
Across two different avatar cues (attractiveness and height) and two distinct behaviors (interpersonal approach and self-disclosure in Experiment 1; negotiation behavior in Experiment 2), altered self-representation produced rapid, robust behavioral changes.
The Proteus Effect occurs even when observers do not perceive or know about the manipulation, demonstrating a target-centric mechanism.
Relationship to behavioral confirmation:
The authors deliberately isolate Proteus effects from behavioral confirmation; real-world interactions would likely involve both mechanisms operating simultaneously, but this research demonstrates a clear, condition-uncoupled self-behavior change.
The authors discuss the potential for a “reverse Proteus Effect,” where a target’s false self-concept (e.g., feeling attractive) elicits perceivers’ responses that confirm that false self-view.
Distinctions from SIDE and deindividuation:
Proteus predicts behavior changes driven by identity cues (avatar attributes) even in the absence of a local group norm or explicit deindividuated conditions.
SIDE emphasizes normative conformity within a group; Proteus emphasizes conformity to an individual identity cue regardless of group norms.
Limitations acknowledged by authors:
No direct manipulation check for attractiveness perception by participants.
Opposite-gender dyads limit generalizability to other interaction types (e.g., same-gender interactions).
Avatar choices in some online contexts may be limited (e.g., There.com’s youthful avatar limitations).
No direct measurement of long-term carryover effects into the physical world.
The potential interaction between Proteus Effect and behavioral confirmation was not tested (i.e., a full 2×2 design where self-perception and others' perceptions are mismatched).
Future directions proposed:
Generalize Proteus effects to other self-representation dimensions (gender, race).
Examine long-term effects and potential therapeutic applications if avatar-based self-representations alter real-world behavior.
Explore the role of user choice in avatar selection and its influence on enhancing or diminishing the Proteus Effect.
Investigate Proteus effects in same-gender pairings and in contexts with more diverse avatar options.
Study mismatches between self-representation and others’ perceptions to directly test Proteus vs behavioral confirmation dynamics.
Practical implications:
Online communities with shared attractive or tall avatars could become more intimate and hyperpersonal, potentially affecting relationship formation and social dynamics at scale.
If virtual environments shape behavior quickly, they could be leveraged for therapeutic or training purposes (e.g., boosting confidence or social skills).
Understanding Proteus effects is critical for ethical considerations around avatar design, manipulation, and behavioral influence in online spaces.
Core Equations, Statistics, and Numerical References (LaTeX-format)
Sample sizes and group compositions:
Experiment 1: N = 32; 16 men and 16 women.
Experiment 2: N = 50.
Interpersonal distance (Experiment 1):
Attractive: M = 0.98, ext{ SD } = 0.36
Unattractive: M = 1.74, ext{ SD } = 1.20
Test: t(30) = 22.42, ext{ }p = 0.02, ext{ }d = 0.40
Self-disclosure (Experiment 1):
Attractive: M = 7.19, ext{ SD } = 2.77
Unattractive: M = 5.42, ext{ SD } = 1.56
Test: t(30) = 2.23, ext{ }p = 0.03, ext{ }d = 0.38
ANOVA (Experiment 2, Split 1):
F(2,47) = 0.63, ext{ }p = 0.53, ext{ } ext{η}^2 = 0.03
ANOVA (Experiment 2, Split 3):
F(2,46) = 5.64, ext{ }p = 0.006, ext{ } ext{η}^2 = 0.20
Post-hoc (Experiment 2, Split 3):
Tall vs Short: M{ ext{Tall}} = 60.63, ext{ SD }=6.55 vs M{ ext{Short}} = 52.06, ext{ SD }=7.30, p = 0.004
Final split (Experiment 2): logistic regression result
ext{x}^2(1, N=50) = 4.41, ext{ }p = 0.04
Prediction success: accept 54 ext{ ext{%}}, reject 80 ext{ ext{%}}
Key Figures and Tables Mentioned
Figure 1: Faces with high, medium, and low attractiveness by gender (face stimuli used in Experiment 1).
Table 1: Means and standard deviations of attractiveness ratings for avatar faces (by gender and attractiveness level).
Figure 2: Equipment setup and CVE layout (HMD, orientation device, cameras, confederate, and mirror).
Table 2: Means and standard deviations of Interpersonal Distance (Split 1) and Split 2 across height conditions (Experiment 2).
Quick Takeaways for Exam Preparation
The Proteus Effect demonstrates that altering self-representation in virtual environments can influence one’s own behavior, even when others’ perceptions of you do not change.
Two independent manipulations (avatar attractiveness and avatar height) produced convergent evidence: increased closeness and self-disclosure with attractive avatars; increased negotiation aggressiveness with taller avatars.
Theoretical framing distinguishes Proteus from traditional behavioral confirmation and SIDE; Proteus emphasizes self-generated behavior driven by identity cues in the avatar, potentially independent of direct social feedback.
Methodological strength: using CVEs to decouple avatar-driven self-perception from others’ perceptions helps isolate the self-directed Proteus Effect.
Limitations to remember: lack of manipulation checks, reliance on opposite-gender dyads, and the exploratory nature regarding long-term or real-world carryover effects.
Implications: avatar choice has rapid, meaningful effects on social behavior; nearly instantaneous changes can accumulate in large online communities, influencing digital world dynamics and possibly real-world interactions.