In psychotherapy, movement synchrony seems to be associated with higher patient satisfaction and treatment outcome. However, it remains unclear whether movement synchrony rated by humans and movement synchrony identified by automated methods reflect the same construct. To address this issue, video sequences showing movement synchrony of patients and therapists (N = 10) or not (N = 10), were analyzed using motion energy analysis. Three different synchrony conditions with varying levels of complexity (naturally embedded, naturally isolated, and artificial) were generated for time series analysis with windowed cross-lagged correlation/ -regression (WCLC, WCLR). The concordance of ratings (human rating vs. automatic assessment) was computed for 600 different parameter configurations of the WCLC/WCLR to identify the parameter settings that measure movement synchrony best. A parameter configuration was rated as having a good identification rate if it yields high concordance with human-rated intervals (Cohen’s kappa) and a low amount of over-identified data points. Results indicate that 76 configurations had a good identification rate (IR) in the least complex condition (artificial). Two had an acceptable IR with regard to the naturally isolated condition. Concordance was low with regard to the most complex (naturally embedded) condition. A valid identification of movement synchrony strongly depends on parameter configuration and goes beyond the identification of synchrony by human raters. Differences between human-rated synchrony and nonverbal synchrony measured by algorithms are discussed.
展开▼
