Date of Award

Spring 5-14-2021

Degree Type


Degree Name

Doctor of Philosophy (PhD)


Educational Policy Studies

First Advisor

Audrey J. Leroux

Second Advisor

Katherine E. Masyn

Third Advisor

Hongli Li

Fourth Advisor

T. Chris Oshima

Fifth Advisor

Meltem Alemdar


In the social and behavioral sciences, it is common for event-history data to have a multilevel structure, such that individuals (e.g., students) in a lower-level are clustered into some higher-level context (e.g., schools). However, little work has explored the common situation that such data are not purely clustered, as in the situation where some students may have attended more than one school during the course of a study. In those cases, the use of a cross-classified discrete-time survival model may be needed to appropriately account for individual mobility across clusters. The purpose of this research was to understand the impact of ignoring a cross-classified data structure due to individual mobility across clusters in a discrete-time survival analysis and to examine how the baseline hazard function, variability of the cluster random effect, mobility rate, and within- and between-cluster sample size impact the performance of a cross-classified discrete-time survival model. A Monte Carlo simulation study was used to specifically examine the performance of a discrete-time survival model, a multilevel discrete-time survival model, and a cross-classified discrete-time survival model. Simulation factors included the value of the between-clusters variance, cluster size, within-cluster size, Weibull scale parameter, and mobility rate. The generating parameters for the simulation study were based on a review of the applied literature. The results indicated that substantial relative parameter bias and unacceptable coverage of the 95% confidence intervals is possible for all model parameters when a discrete-time survival model is used that does not account for either clustering or individual mobility across clusters, and to a lesser extent, when using a multilevel discrete-time survival model that does not account for mobility. Across nearly all simulation conditions and for all parameters, use of the cross-classified discrete-time survival model resulted in little to no relative parameter bias and acceptable coverage of the 95% confidence intervals. These findings will be useful for methodologists and practitioners in educational research, public health, and other social sciences where discrete-time survival analysis is a common methodological technique for the analysis of event-history data.


File Upload Confirmation


Available for download on Sunday, April 30, 2023