Predicting Dyadic Synchrony: A Theory-Driven Machine Learning Approach

Area of Science:

• Developmental Psychology
• Computational Social Science

Background:

• Dyadic synchrony, the coordination of emotional and behavioral signals between mother and child, is crucial for early socioemotional development.
• Predicting dyadic synchrony is challenging due to complex, context-sensitive interactions between relational and psychological factors, often not captured by linear models.

Purpose of the Study:

• To apply a theory-guided machine learning approach to identify predictors of observed dyadic synchrony in mother-child dyads.
• To compare the predictive performance of machine learning models (Random Forest) with traditional linear regression.

Main Methods:

• Utilized data from 204 mother-child dyads in the PEACE Study (collected Nov 2021-Aug 2022).
• Coded dyadic synchrony and affective behaviors from free-play interactions using the Coding Interactive Behavior system.
• Employed machine learning (Random Forest) and linear regression, with SHAP and ALE for model interpretation, examining maternal anxiety, resilience, parenting stress, and affective expressions.

Main Results:

• While linear regression showed slightly stronger predictive performance, Random Forest models identified complex non-linear and interaction-based patterns.
• Dyadic synchrony decreased with imbalances between parenting stress and child positive affect.
• Affective mismatch demonstrated a non-monotonic association with synchrony, highlighting context-dependent emergence.

Conclusions:

• Dyadic synchrony emerges from configurations of stress, affect, and emotional alignment, rather than simple additive effects.
• Interpretable machine learning offers a valuable complement to traditional statistical models for exploring complex relational processes in early development.
• Findings underscore the context-dependent nature of dyadic coordination in supporting socioemotional development.