Eye gaze is not inversion-proof: A robust gaze inversion effect in a large, sex-balanced sample

Turning a face upside down markedly impairs face recognition. Yet gaze judgements can be supported substantially by local eye-region information, so a strict account in which inversion preserves feature-based processing predicts little or no gaze inversion cost. Previous whole-face studies, generally based on small samples, have yielded mixed results. We estimated the magnitude of the inversion cost in a large, sex-balanced sample of 190 participants. They judged whether briefly presented upright or inverted faces, each followed by a mask, were looking directly at them. Inversion widened the cone of direct gaze: the mean 50% threshold rose from 7.29° for upright faces to 9.74° for inverted faces, an increase of 2.45° (95% CI [1.99°, 2.91°]; Cohen's d  = 0.76). Inversion also produced modestly but consistently shallower psychometric functions, indicating that direct and averted gaze were distinguished less sharply. Females showed a narrower cone of direct gaze than males in both orientations, with no consistent evidence for a sex difference in slope. The inversion-related threshold increase was 0.67° larger in male participants than in female participants (95% CI [−0.25°, 1.59°]; Cohen's d  = 0.21), but the between-sex difference was not statistically significant and the confidence interval was too wide to establish equivalence. Under brief, masked viewing, eye gaze is therefore not inversion-proof. These findings constrain strict feature-preservation accounts and provide a more precise estimate of the gaze inversion effect than was available from previous small-sample studies.