Accounting for Correlated Random and Fixed Effects in Tests of Moderation in Group Therapy Studies

• Activity Number: 469
• Type: Topic Contributed
• Date/Time: Wednesday, August 7, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Mental Health Statistics Section
• Author(s): Susan Paddock*+ and Thomas Leininger and Sarah Hunter
• Companies: RAND Corporation and Duke University and RAND Corporation
• Keywords: hierarchical modeling ; conditional autoregression ; random effects ; fixed effects ; moderation ; substance abuse treatment

Abstract

Client outcomes for group therapy are clustered due to common client participation. Random effects modeling of such correlation can be problematic when random effects are associated with the same unit (e.g., therapy group session) as fixed effects of interest. We examine this issue when testing whether the content or leadership of group cognitive behavioral therapy (GCBT) sessions is related to changes in client symptom reporting of depressive symptoms. Tests of moderation were conducted by examining regression coefficients of interaction terms for session features and time since GCBT entry. To mitigate potential bias and variance inflation due to possible correlation of these effects, we restricted the random session effects' contributions to be orthogonal to those of the session-level moderators. Restricted models have better or comparable fit than unrestricted models for our data. We demonstrate the importance of accounting for correlation between random session effects and session moderators and mitigating variance inflation on statistical tests of moderation in open-enrollment group therapy studies.