Upcoming manuscript
Funded by the Templeton World Charity Foundation (TWCF-2020-20505)
Our manuscript "Virtual Reality's Promise in Fostering Empathy and Compassion: Mode-Specific Interventions for Adolescents Facing Ethnic Discrimination" is an upcoming publication.
This work represents five years of research (2020–2025), funded by the Templeton World Charity Foundation.
Here are some of the key contributions.
The VR empathy field has used three main approaches: simulation of intergroup encounters, bystander perspectives on discriminative behaviours and first-person embodiment of outgroup members. However, these modes have rarely been compared directly in the same study. Our paper tests all three within a single RCT, holding content and hardware constant. Our finding that modes produce significantly different and even opposite effects is a major empirical contribution. It even overturns the implicit assumption in most prior work that "VR" is a single treatment.
The field has extensively assumed that first-person embodiment is the optimal mode for empathy and compassion, based on body ownership illusions and sensorimotor resonance theories. Some recent work has found first-person perspective did not directly enhance empathy despite producing stronger embodiment, but this remains poorly understood. Our study is one of the first adequately powered RCTs to demonstrate that third-person observation significantly outperforms first-person embodiment on multiple empathy and compassion outcomes (d=0.51–0.62), using emotionally charged ethnic discrimination content. The supplementary factorial analysis showing that embodiment produces a negative main effect on compassion outcomes strengthens this conclusion considerably.
Our paper integrates Theory of Mind, construal level theory and the empathy-as-hard-work framework to explain why third-person observation outperforms embodiment. This mechanistic account — that emotional overload from first-person immersion can cancel out empathy gains for high-intensity content — provides a theoretically grounded boundary condition: embodiment enhances empathy for mild stimuli; distressing content reverses this. I believe that this is a clear, falsifiable theoretical advance the field has needed. Prior reviews have noted that negative affect during embodied intergroup experiences can increase implicit bias and lead to withdrawal, creating a design dilemma for intervention researchers. Our paper gives that concern empirical precision.
No prior VR study had embedded a moral exemplar avatar into a discrimination scenario to test whether it enhances empathy and compassion. Our finding that the Role Model condition reduced motivation to act (d=−0.58) relative to control is genuinely novel and consequential. It generates three falsifiable explanatory mechanisms: bystander diffusion, psychological reactance, and cognitive distraction from perspective-taking, that others can now test. I believe this contribution to be original in both the VR literature and the moral exemplar/character education literature.
Early discussions about VR as an "empathy machine" often used the term empathy loosely without differentiating it from sympathy or compassion, drawing criticism from scholars. Our paper explicitly distinguishes empathy (affective and cognitive) from compassion (five subscales), measuring them with separate validated instruments and providing confirmatory factor analysis for each. The divergent patterns (Observe enhancing empathy, Role Model undermining motivation to act) validate this distinction empirically. This is a conceptual contribution to the field's conceptual hygiene, not merely a measurement choice.
A 2024 meta-analysis found that emotional empathy increased temporarily after VR but returned to baseline over time, while cognitive empathy remained enhanced. Our paper provides the most granular temporal profile yet in an adolescent discrimination context: significant immediate effects, attenuation by 2 months (T3), complete attenuation by 4 months. This has direct practical implications for intervention booster scheduling, a contribution to applied program design.
The equivalence of NPT to the passive control, despite identical discrimination content, suggests that text-based narrative requires imaginative effort that adolescents cannot sustain for emotionally charged content. This establishes a boundary condition for narrative perspective-taking and provides an argument for VR's specific advantage in this content domain.
The educational literature has increasingly used VR as an innovative tool for moral education, often assuming that greater immersion produces greater moral development. Our study's role model and embodiment findings challenge this assumption directly. For educators and programme designers, the practical takeaway is that design complexity and immersion do not reliably improve moral outcomes and can actively harm them.
The Role Model backfire finding has direct implications for character education designs that assume that embedding virtuous role models in experiential content is straightforwardly beneficial. The psychological reactance and bystander effect mechanisms we propose are testable and practically important.
Systematic reviews specifically call out the absence of long-term follow-ups and the scarcity of longitudinal research designs as fundamental limitations in the field. Even Herrera et al. (2018), considered a landmark longitudinal study, only followed up over 8 weeks. Ours extends follow-up to 6 months. The finding of effect attenuation by T3 is itself a contribution.
Herrera et al. (2018) compared VR against an active control (narrative perspective-taking) but lacked a passive no-intervention control, limiting causal inference. Our design includes both, which allows separating VR-specific effects from general perspective-taking effects and from natural change over time. This is a specific design contribution the field has explicitly called for.
One persistent problem in this field is that studies vary the medium (VR vs video vs text, or even comparing VR to television) and the perspective (first vs third person) simultaneously, making it impossible to attribute effects to either. Our study held the medium constant (all VR conditions used identical Meta Quest 2 hardware) while varying only the perspective. This directly addresses the Clark/Kozma media-versus-methods debate and allows cleaner inference than most prior work.
This is the first large-scale RCT of VR empathy/compassion in Sub-Saharan Africa with African adolescents. Besides contributing cultural variability to the field, the content of the intervention (Connected Spaces) was designed specifically for this interethnic context, making the findings ecologically valid in a way that cross-cultural transplants may not be.
This one is particularly close to my heart and was one of my hopes when we implemented the multi-mode design. The superiority of the controller-free Observe condition directly challenges the assumption that sophisticated hardware and active interactivity are necessary for VR empathy and compassion effects. This finding implies that effective VR empathy interventions may be achievable with smartphone-based VR viewers costing under $10, rather than premium HMDs costing $150 or more. In a field largely developed by well-resourced Western labs, this is a contribution with significant implications for scaling intervention access in resource-constrained settings globally.
This study is part of the VHCI Lab's ongoing programme of research into VR, empathy, and character development.
Explore Our Research