Investigating variability in multilevel models: Going beyond therapist effects

Although psychological therapies can benefit many people, over half of the patients who receive therapy do not recover. Also, across services and therapists there is a great deal of variability in patient outcomes. Studies from the USA, using multilevel modelling (MLM), have indicated that the variability between therapists has a significant effect on patient outcomes, with some therapists over twice as effective as others. However, some of these findings were derived from data samples that did not meet the recommended size for reliably estimating therapist effects using MLM.
This methodology-focused thesis discusses five studies, published between 2012 and 2017, that contain some of the largest samples of routinely collected service data to date. The initial aim was to replicate the USA studies with large UK samples. However in doing so, analytical methods were developed which utilised random slopes and residuals from multilevel models, to better understand therapist variability and ask research questions about ‘how’ and ‘why’ therapists vary in effectiveness.
The five studies in this thesis produced some of the most reliable estimates of the size of the therapist effect. They also include the first estimates of therapist effects for patient drop-out and deterioration. In addition, the methods developed were applied to: reliably identify the most effective therapists controlling for case-mix; show how the effects of important patient variables, like intake severity and number of sessions attended, are moderated by therapists; identify therapist factors associated with better outcomes and, for the first time, consider therapist variability on two outcomes simultaneously.
Collectively, the studies provide strong evidence of the importance of the therapist to patient outcomes and strong justification for focusing the research effort on therapists and therapist variability. This thesis provides some original methodologies which can contribute to such a research effort.