Beyond Face Value: Assessing the Factor Structure of an Eye-Tracking Based Attention Bias Task

Background Behavioral measurement of attention bias for emotional stimuli has traditionally ignored whether trial-level task data have a strong enough general factor to justify a unidimensional measurement model. This is surprising, as unidimensionality across trials is an important assumption for computing bias scores. Methods In the present study, we assess the psychometric properties of a free-viewing, eye-tracking task measuring attention for emotional stimuli. Undergraduate students ( N = 130) viewed two counterbalanced blocks of 4 × 4 matrices of sad/neutral and happy/neutral facial expressions for 10 seconds each across 60 trials. We applied a bifactor measurement model across ten attention bias metrics (e.g., total dwell time for neutral and emotional stimuli, ratio of emotional to total dwell time, difference in dwell time for emotional and neutral stimuli, a variable indicating whether dwell time on emotional stimuli exceeded dwell time on neutral stimuli) to assess whether trial-level data load on to a single, general factor. Unidimensionality was evaluated using omega hierarchical, explained common variance, and percentage of uncontaminated correlations. Results Total dwell time had excellent internal consistency for sad (ɑ = .95, ɷ = .96) and neutral stimuli (ɑ = .95, ɷ = .95), and met criteria for unidimensionality, suggesting the trial-level data within each task reflect a single underlying construct. However, the remaining bias metrics fell short of the unidimensionality thresholds, suggesting not all metrics are good candidates for creating bias scores. Conclusion Total dwell time by valence had the best psychometrics in terms of internal consistency and unidimensionality. This study demonstrates the importance of assessing whether trial-level data load onto a general factor, as not all metrics are equivalent, even when derived from the same task data.