Further Comparisons of the Random Coefficients Model with Repeated Measures ANOVA in Longitudinal Group Randomized Trials

• Activity Number: 87 - Survival and Longitudinal/Clustered Data Analysis
• Type: Contributed
• Date/Time: Monday, August 9, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Biometrics Section
• Author(s): Jonathan Moyer* and David Murray
• Companies: National Institutes of Health and National Institutes of Health
• Keywords: group randomized trial; cluster randomized trial; repeated measures; random coefficients; longitudinal

Abstract

Longstanding guidance recommends the use of the random coefficients (RC) model over repeated measures ANOVA (RM-ANOVA) when modeling longitudinal group randomized trials, as the RC model has yielded nominal Type I error rates when applied to data generated under both approaches. In contrast, RM-ANOVA has exhibited inflated Type I error rate when applied to data generated under the random coefficients model. However, this guidance was developed using only cross-sectional data, assuming variance components covariance structure for the RM-ANOVA analysis model, and explicitly modeling group x time variation. Left unanswered are how well RM-ANOVA with unstructured covariance would perform on data generated according to the RC model, the impact of not modeling group x time variation if such variation is present in the data generating model, and if similar patterns are observed in cohort data. In this simulation study we investigate the effect these modifications have on Type I error rate over a range of intracluster correlations.