A Virtual Character's Agency Affects Social Responses in Immersive Virtual Reality: A Systematic Review and Meta-Analysis

Social responses to virtual humans in immersive Virtual Reality (VR) may depend on whether users perceive them to be controlled by a person (avatar) or by the computer (agent). As related evidence is growing, this meta-analysis set out to synthesize corresponding findings. Specifically, we analyzed whether the perceived agency of a virtual human affects (1) social presence, (2) the evaluation of the virtual entity, and (3) behaviors toward it differently. Additionally, we controlled for task type, the virtual entity's appearance, the type of interaction, and agency manipulation. A literature search yielded 20 studies examining 911 participants (agent condition: 460, avatar condition: 451). Differences between perceived avatars and agents in social responses were calculated using Hedges' g. Results showed differences for social presence (g = 0.65, p = .001) and evaluation (g = 0.30, p = .024) in favor of avatars over agents. No effects emerged for behavioral outcomes. Meta-regressions revealed larger agency-differences in social presence for neutral than for negative tasks, and differences in evaluation for positive compared to mixed and neutral tasks. Overall, our findings suggest that while deliberate social responses like social presence and evaluation depend on perceived agency, automatic behaviors do not. For future research, particularly consistent conceptualizations of key variables are necessitated.