Using the Multiple-Membership Random Effects Model to Estimate Treatment Effects With Condition- and Cluster-Switchers

• In Event: Issues in Clustered Data Structures

Abstract

In multi-site cluster-randomized trails (MS-CRTs), participants do not always stay in the same randomly assigned condition nor site. Methodologists have recommended ways to handle condition-switching when estimating a treatment’s effect using, for example, intent-to-treat, per-protocol or as-treated analyses. However, despite methodological research emphasizing the importance of not ignoring cluster-switching through use of the multiple-membership random effects model (MMREM), the inevitable cluster-switching that occurs when participants condition-switch is consistently ignored. The current study details how and why to enhance suggested methods for estimating treatments’ effects when participants change conditions and/or clusters in MS-CRTs by incorporating the MMREM rather than a conventional HLM. Analyses are conducted (pairing each method with each model) using two real datasets. Results are compared and discussed.

Authors

• Audrey J. Leroux, Georgia State University

• Susan Natasha Beretvas, The University of Texas - Austin