Dynamic Time Warping Detects Enhanced Dyadic Behavioral Matching as Facial Mimicry During Live Interaction

Dynamic time warping (DTW) is increasingly used to detect facial mimicry, yet the reliability of distance measures as indicators of mimicry remains uncertain. This study evaluated. DTW’s ability to infer mimicry by presenting participants with both live model performances and prerecorded videos of dynamic facial expressions. Simultaneous dyadic frontal facial videos of participants and models were acquired, and action unit (AU) time series were analyzed using Py-Feat, with DTW applied to estimate dyadic temporal pattern similarity. A principal component analysis of the models’ raw AU time series identified the highest-weighted AUs. For AU12 (lip corner puller) and AU06 (cheek raiser), temporal pattern distances were smaller in the positive-live condition than in the positive-video condition, while AU15 (lip corner depressor) exhibited a trend toward smaller distances in the negative-live than in the negative-video condition. These findings suggested that live interactions, within matching emotional conditions, enhanced dyadic behavioral alignment. However, all three AUs exhibited smaller distances when the corresponding emotion was inactive, raising the concern that reduced DTW distances could result from either mimicry or mutual inaction. Researchers should carefully consider design and theoretical frameworks when applying DTW in mimicry studies.