Dissertation
Investigating alternative methods to recover level-2 covariates in multilevel models
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Summer 2022
DOI: 10.25820/etd.006562
Abstract
Hierarchical data is often observed in education data. Analyzing such data with Multilevel Modeling becomes crucial to understanding the relationship at the individual and group levels. However, one of the most significant problems with this kind of data is small sample sizes and very low Intraclass Correlations. The multivariate Latent Covariate Model is often accepted as the gold standard for analyzing hierarchically structured data. However, previous studies showed that this model did not work very well under the abovementioned conditions. This dissertation aimed to address two research questions around this situation.
The first research question intended to show how the Multilevel Latent Covariate Model worked under these conditions via a simulation study and a real data application. The second research question suggested six new candidate models as an alternative to the Multilevel Latent Covariate Model. The performances of all candidate models were assessed using the same simulation study and the real data application.
Raw bias, Root Mean Squared Error, Standard Error Ratio, Type 1 error rate, and Power were used to compare the feasibility of the candidate models while analyzing the multilevel data with low intraclass correlation and very small level-1 sample sizes. The results showed that the alternative candidate models outperformed the gold standard.
Details
- Title: Subtitle
- Investigating alternative methods to recover level-2 covariates in multilevel models
- Creators
- Ismail Dilek
- Contributors
- Lesa Hoffman (Advisor)Ariel Aloe (Committee Member)Brandon LeBeau (Committee Member)Ozlem Ece Demir Lira (Committee Member)Jonathan Templin (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Psychological and Quantitative Foundations
- Date degree season
- Summer 2022
- Publisher
- University of Iowa
- DOI
- 10.25820/etd.006562
- Number of pages
- xii, 177 pages
- Copyright
- Copyright 2022 Ismail Dilek
- Language
- English
- Description illustrations
- illustrations (some color)
- Description bibliographic
- Includes bibliographical references (pages 97-105).
- Public Abstract (ETD)
This dissertation examined the performance of the Multilevel Latent Covariate Model with multilevel data with very small level-1 sample sizes and very low intraclass correlations. The second ambition of the current study was to propose alternative methods to this model, and to investigate the performance of these candidate models under certain special conditions.
The results showed that these new six candidate models performed better than the Multilevel Latent Covariate Approach while using the simulation studies. An application of real data analyses was conducted by using all candidate models. The proposed models improved the accuracy of level-2 regression coefficient to recover level-2 covariates.
- Academic Unit
- Psychological and Quantitative Foundations
- Record Identifier
- 9984285248402771
Metrics
60 File views/ downloads
45 Record Views