The Estimation Performance of Cross-Classified Multiple Membership Modeling and Consequence of Model Misspecification

• In Event: Issues in Clustered Data Structures

Abstract

Cross-classified multiple membership (CCMM) modeling is an extension of hierarchical linear modeling for multilevel nonnested data. This Monte Carlo simulation study evaluates the estimation performance of CCMM modeling under various conditions that emulate real research settings, and investigates the consequence of ignoring CCMM data structure. Preliminary results indicate that (1) Markov Chain Monte Carlo estimation can produce accurate parameter estimates for CCMM models, and (2) ignoring CCMM data structure does not affect parameter estimates of fixed effects, but it results in biased estimates of variance components.

Authors

• Shuyan Sun, University of Maryland - Baltimore County

• Wei Pan, Duke University