Education and Human Sciences, College of (CEHS)


First Advisor

James Bovaird

Date of this Version



Choi, D. (2022). Analyzing Clustered Longitudinal Data using Latent Curve Model with Structured Residuals (LCM-SR) (unpublished master's thesis). University of Nebraska-Lincoln, Lincoln, Nebraska.


A THESIS Presented to the Faculty of The Graduate College at the University of Nebraska In Partial Fulfillment of the Requirements For the Degree of Master of Arts, Major: Educational Psychology, Under the Supervision of Professor James Bovaird. Lincoln, Nebraska: February, 2022

Copyright © 2022 Dongho Choi


The LCM-SR can provide an inferential basis for understanding reciprocal relations while controlling for individual differences in the trajectories of young children’s psychological development. Yet, a hierarchical structure in the data has not been often adequately addressed even though that is common in social and educational research. The purpose of this study is to explore the impact of dependency among observations on the results when using the LCM-SR, and how to appropriately analyze the clustered longitudinal data for more accurate inference. To do this, the MLCM-SR (disaggregated approach; the “two-level” model) was introduced and compared with the single level LCM-SR considering nesting effects (aggregated approach; the “complex” model), and the single level LCM-SR ignoring nesting effects (conventional approach; the “default” model). This study used both simulated data and actual data to compare the performances of the models.

The simulation study results showed that all the models showed high rates of non-convergence or improper solutions in certain conditions, especially in low sample size conditions. The total number of proper solutions was higher for the complex/default model than for the two-level model in general. Also, bad model fit, severe bias, low coverage rate, and low power were found in conditions with a large percentage of variance as well as a large residual variance at the between-group level. The severity of bias increased as the sample size decreased. The two-level model showed little or no bias in general, thus showing a decent level of power and a nominal level of type 1 error rate. The actual data analysis results showed that even though there was a difference in the standard errors found between the models, using different modeling strategies did not lead to different conclusions.

Advisor: James Bovaird