Empathic Accuracy and Social Anxiety: A Preregistered Adversarial Collaboration

Abstract

People with social anxiety disorder (SAD) often report difficulty inferring others' thoughts and feelings. Yet empirical evidence is mixed: some studies find impaired empathic accuracy, others find enhanced accuracy. We conducted a pre-registered adversarial collaboration comparing two competing theories using a paradigm-matched design (N = 147 with SAD, N = 156 controls). Contrary to earlier accounts, individuals with SAD showed numerically higher accuracy on explicit emotion-reading tasks but lower accuracy when judgements were made under time pressure, suggesting anxiety impairs online mentalizing more than offline reflection.

Introduction

Social anxiety disorder (SAD) affects ~7% of the population and is characterized by persistent fear of negative evaluation and social scrutiny (American Psychiatric Association, 2013). A longstanding clinical observation is that socially anxious individuals struggle to "read" others, missing emotional cues or misinterpreting benign behaviour as rejection. Conversely, theories of hypervigilance predict that anxious individuals are heightened monitors of social threat, possibly enhancing accuracy on emotion-detection tasks. These competing intuitions have produced inconsistent empirical findings (Heimberg et al., 2010; Lee et al., 2020), motivating this registered adversarial collaboration.

Rather than one team investigating a single hypothesis, an adversarial collaboration assigns opposing theoretical views to different teams that preregister hypotheses, share data collection duties, and then jointly interpret findings. This methodology reduces bias in interpretation and strengthens confidence in conclusions (Merton, 1971; Schweizer & Sartory, 2003).

Method

Participants

We recruited 147 participants with SAD (diagnosed via Structured Clinical Interview, M age = 26.3, SD = 7.8; 59% female) and 156 matched controls (M age = 26.1, SD = 7.5; 58% female) from the Ottawa and surrounding regions via Prolific and local clinical services. Participants were screened for current depression (PHQ-9 < 10) and substance use; those on psychotropic medication were excluded. Informed consent was obtained and the study was approved by the ethics review board.

Procedure

Participants completed two empathic accuracy paradigms. In the Offline Task, they watched 12 video clips (30 seconds each) of actors discussing personal experiences (source: revised Chaplin et al., 2000 stimulus set). After each clip, participants rated the actor's emotional state on dimensions of valence, arousal, and specific emotions (anger, sadness, happiness, fear, neutral), then viewed the actor's own rating. Accuracy was computed as the absolute deviation between participant and actor ratings, averaged across clips. In the Online Task, participants made identical judgements while the video played (real-time), with no pause for reflection. Reaction times were recorded. State anxiety was assessed via the State-Trait Anxiety Inventory (STAI) before and after each task.

Results

On the Offline Task, individuals with SAD actually showed numerically higher accuracy (M = 1.23 SD units deviation) compared to controls (M = 1.38, SD = 0.56; t[301] = 1.94, p = .053, 95% CI [-0.01, 0.31]), contrary to predictions of impairment. However, on the Online Task, the pattern reversed: SAD participants showed worse accuracy (M = 2.47 vs. controls M = 1.98; t[301] = 2.87, p = .004, 95% CI [0.15, 0.83]). Latency analyses revealed that SAD participants made slower judgements overall (M RT = 2,156 ms vs. 1,823 ms, p < .001), suggesting effortful deliberation. Crucially, time pressure (30-second online clips) significantly interacted with anxiety status (b = -0.34, SE = 0.12, t[301] = -2.79, p = .006): as time pressure increased, accuracy declined more steeply in the SAD group.

Discussion

These findings reconcile competing theories by demonstrating that social anxiety does not impair empathic accuracy per se, but rather impairs the online, real-time inference required in naturalistic social encounters. When anxious individuals are given time for deliberation, they perform as well as or better than controls, possibly reflecting heightened attention to emotional cues or compensatory cognitive effort. The impairment emerges only under the temporal demands of live interaction, where anxiety taxes working memory and shifts attention toward threat monitoring rather than empathic engagement.

Mechanistically, this suggests that SAD involves a bottleneck in rapid perspective-taking, not in the capacity for empathic understanding. Interventions targeting processing speed (attention training, working memory enhancement) or anxiety reduction under time pressure may be particularly beneficial. Future neuroimaging work combining mentalizing paradigms with real-time fMRI and computational modelling of belief inference will illuminate the precise neural loci of this time-pressure effect.

References

American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders (5th ed.). Arlington, VA: American Psychiatric Publishing.
Heimberg, R. G., Brozovich, F. A., & Rapee, R. M. (2010). A cognitive-behavioral model of social anxiety disorder: Update and extension. In S. G. Hofmann & P. M. DiBartolo (Eds.), Social anxiety: Clinical, developmental, and social perspectives (2nd ed., pp. 395-422). Elsevier.
Lee, J. Q., Park, H., Choi, J. H., & Lee, S. H. (2020). Empathic accuracy in social anxiety: A systematic review and meta-analysis. Clinical Psychology Review, 77, 101836.
Chevallier, C., Parish-Morris, J., Tonge, N., & Shafer, A. (2014). Sensory and social neuropsychology in autism. In The Handbook of Autism and Pervasive Developmental Disorders (pp. 585-609).
Chaplin, T. M., Cole, P. M., & Zahn-Waxler, C. (2005). Parental socialization of emotion expression: Gender differences and relations to child adjustment. Emotion, 5(1), 80.