An Open-Source Standardized Pipeline for Equitable Observations of Interactive Behavioral Dynamics: Theory-driven Measurement, Analysis, and Masking

Behavioral observation has a longstanding tradition in psychology. However, its reliance on manual coding makes observational methods costly and prone to biases that relate to the inherent limits of coding schemes and to the limits of humans acting as measurement systems (i.e., raters). Behavioral observation research therefore has drawbacks in terms of applicability, accessibility, reproducibility, and equitability. Computer vision (CV) methods have promise in this regard. In the context of quantifying human movement, CV approaches to pose estimation (e.g., OpenPose) are becoming increasingly precise and have the benefit of being non-intrusive, reproducible, and low-cost. With such methods, human movement in videos can be processed as multivariate time-series data, offering a basis for a more comprehensive analysis of individual, interpersonal, and social dynamics, opening up the possibility for the widespread study of diverse populations in real-world contexts. Fulfilling such a promise goes beyond technical progress, it requires a theory-driven implementation of new observational methods, made accessible for use, and made equitable to learn. To showcase this promise, we present a CV pipeline with dynamical systems analyses of dyadic interactions, including a cross-cultural investigation of movement fluidity in sibling cooperation during task-focused interactions and longitudinal assessments of infant-caregiver coordination during free-play sessions. Our pipeline incorporates a pedagogical computational notebook, pose estimation (YOLOv8), theory-driven linear and non-linear time series analysis, and a CV method for more privacy-aware video data archiving. Together this pipeline showcases a next in behavioral observation, expanding both who can conduct research and who can benefit from it.