State-dependent dynamics of therapist-client affect in psychotherapy

Objective: Psychotherapy unfolds as a continuous interaction between therapist and client, yet many process models implicitly assume that affective dynamics are stationary within dyads or can be summarized without modeling changes in interaction organization over time. We tested whether therapist--client affect is better characterized as a coupled process that shifts between distinct interaction modes.
Method: Using second-by-second categorical affect codes (negative, neutral, positive) from 133 sessions across 95 therapist--client dyads, we first compared independent within-person Markov models with coupled dyadic models under dyad-grouped cross-validation. We then fit a shared-latent hidden Markov model (HMM) to identify interaction regimes and evaluated those regimes using recurrence-based and sequential motif analyses.
Results: Coupled models outperformed independent models, supporting dyadic coupling in therapist--client affect. The inferred regimes differed in persistence, transition structure, and within-regime temporal organization. At the session level, regime use varied continuously along low-dimensional axes reflecting volatility and variability in regime duration.
Conclusion: Therapist--client affect is best understood as a nonstationary, state-dependent dynamical system organized by patterned movement among distinct interaction regimes rather than by a single fixed set of interaction rules.