Modeling Cross-Classified Data With Adequate Within-Subject Variance-Covariance Structures: A Lesson From the Project ELLA Data

• In Event: Poster Session 10
  In Poster Session: Multilevel Modeling: Complex Survey Data and Effect Size

Abstract

In this study we examine the impact of inadequately analyzing longitudinal cross-classified data using data from project ELLA (English language/Literacy Acquisition), a large scale longitudinal study with ELL students followed from kindergarten to 3rd grade. Similar to other educational data, data from project ELLA are not strictly hierarchical but are cross-classified by students and classrooms over time. Although the impact of inadequately analyzing the non-strictly hierarchical multilevel data has been thoroughly examined, the impact of misspecifying the within-subject variance-covariance structure in cross-classified random effect models (CCREMs) has not yet been investigated. We will fit several different models with data from project ELLA and compare the results. The implications of the findings and limitations of the study are discussed.

Authors

• Hok Chio Lai, Texas A&M University - College Station

• Fuhui Tong, Texas A&M University - College Station

• Myeongsun Yoon, Texas A&M University

• Rafael Lara-Alecio, Texas A&M University

• Beverly J. Irby, Texas A&M University - College Station

• Oiman Kwok, Texas A&M University - College Station