A Comparison of Methods for Handling Cross-Classified Multiple Membership Data Structures

• In Event: Issues in Clustered Data Structures

Abstract

This study is designed to investigate the impact of mis-specification of cross-classified multiple membership data (ccmm) structures in multilevel models. Ccmm data structures are commonly encountered and yet ignored in longitudinal datasets. For example, in educational data, students are frequently cross-classified by more than one higher level unit (e.g., kindergarten and elementary schools) and some level one units (students) might be members of multiple higher level units (e.g. elementary schools). This simulation study found substantially biased estimates of the level two and the level one variance components and under-estimation of the level two predictor’s coefficient when the ccmm data structure was ignored.
Keywords: cross-classified multiple membership, multilevel modeling, student mobility

Authors

• Hyewon Chung, Chungnam National University

• Susan Natasha Beretvas, The University of Texas - Austin