Very often therapy change data is “nested”: sessional observations within each client, each client within therapist, each therapist within service etc. That violates the “independence of observations” assumption or model that makes most conventional statistical methods possible and requires alternative methods now generally termed “multi-level models/modelling” though still sometimes termed “hierarchical linear models (HLM)” or “random coefficients regression”.
Try also #
Multi-level modelling (MLM)
Independence of observations
Mainly chapter 5 but also 8 and 10.