To Disaggregate or Not to Disaggregate: A Focus on Covariates in Multilevel Models

• In Event: New Insights for Designing Multilevel Models: Predicting Random Effects, Using Predictor Aggregates, and Mediation Planning

Abstract

It is routinely suggested that appropriate actions are taken to disaggregate level-one variables in multilevel models if they are of substantive importance. However, disaggregation of level-one covariates is seldom recommended despite the well-established issues, including biased estimates which perturb the entire model. The current study uses a Monte Carlo simulation to explore both the benefits and consequences of disaggregation. We build off the work of Rights et al. (2020) by evaluating the impact of the size of intraclass correlation as well as the use of the latent aggregate model. The results of this study will inform researchers’ decision to disaggregate level-one covariates rather than relying on historically relied on rules of thumb.

Authors

• Remus Rose Mitchell, University of California - Los Angeles

• Craig K. Enders, University of California - Los Angeles