First Advisor
Tsai, Chai-Lin
Second Advisor
Yu, Han
First Committee Member
Merchant, William
Degree Name
Doctor of Philosophy
Document Type
Dissertation
Date Created
8-2024
Department
College of Education and Behavioral Sciences, Applied Statistics and Research Methods, ASRM Student Work
Abstract
Ecological Momentary Assessment (EMA) studies, also known as Intensive Longitudinal Data (ILD), involve participants that are intensively measured over time. Intermittent missing data tends to occur due to participants not responding when prompted. The high volume of assessments and intermittent nature of missingness have made some traditional longitudinal missing data methods unsuited to handle the missingness of EMA data. Two recent missing models intended for ecological momentary assessment missing data situations have emerged that jointly model the outcome and missingness, providing information about the latent trait of responding to prompts. Both models implement a shared parameter as a random effect but do so in different ways. X. Lin et al. (2018) model the missing process by using a random intercept logistic regression model for the binary missing prompt indicators. Cursio et al. (2019) model the missing process using item response theory to model responsiveness to the prompting device as a latent trait. The purpose of this study was to compare these two joint models used to handle missing data in ecological momentary assessment (EMA) studies and to evaluate their performance under different assessment and missing data scenarios. A simulation was designed to compare these two joint models under a few different assessments and percentage of missing prompts scenarios to evaluate their performance in terms of parameter estimate bias, empirical standard errors, and computation run time. Results in this missing data simulation displayed that the joint shared parameter missing data models consistently outperformed statistical software’s default missing data method list-wise deletion displaying the value of these models in ILD missing data situations. The Latent Trait Shared Parameter Mixed Model (LTSPMM) performed superior in this simulation and is recommended as a missing data model in ILD studies. The results of this study provides researchers with guidance on the performance of both shared parameter missing data models under missing data conditions that might be observed in real ILD data situations.
Abstract Format
html
Extent
118 pages
Rights Statement
Copyright is held by the author.
Recommended Citation
Harding, Justin, "Performance of Shared Parameter Missing Data Models for Intensive Longitudinal Data" (2024). Dissertations. 1091.
https://digscholarship.unco.edu/dissertations/1091