Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Dissertation_Final_Meng-Ting Lo.pdf (2.14 MB)

ETD Abstract Container

Abstract Header

Alternative Methods for Modeling Clustered Ordinal Data

Lo, Meng-Ting
http://rave.ohiolink.edu/etdc/view?acc_num=osu1586792558628762

Abstract Details

, Doctor of Philosophy, Ohio State University, Educational Studies.
Multilevel modeling is commonly used with clustered data, and much emphasis has been placed specifically on the multilevel linear model (MLM). When modeling clustered ordinal data, a multilevel ordinal model with cumulative logit link assuming proportional odds (i.e., multilevel cumulative logit model) is typically used. Depending on the research questions and inferences a researcher would like to draw from his/her findings, generalized estimating equations (GEE), a type of population average model, may be used as an alternative to multilevel cumulative logit model. Despite their usability, GEE for ordinal data and multilevel cumulative logit models are not often encountered in the applied literature. One of the reasons for underuse of these models may be that researchers are not familiar with the theory and specification of different approaches for analyzing clustered ordinal data. The goal of this dissertation was to investigate appropriateness of methods for modeling clustered ordinal data under different study conditions, and clarify interpretation differences between multilevel cumulative logit models and GEE approaches. In the current study, a simulation study was conducted to systematically examine the performance of multilevel cumulative logit models and GEE for modeling clustered ordinal data across different study conditions. Overall, two modeling approaches performed similarly regarding fixed effects biases and statistical power under different study conditions. However, when there was a smaller number of clusters (i.e., 10 or 30 clusters), the performance of GEE method was inferior to multilevel cumulative logit models in terms of 95% confidence interval coverage rates. Using multilevel cumulative logit models are useful to examine the contextual effects and the variation between clusters. In order to obtain reliable estimates of fixed and random effects, at least 50 clusters or more are needed, assuming all the assumptions of multilevel cumulative logit models are met. GEE can be used if the between-cluster variation and contextual effects are not of interest but the clustering effect should be taken into account. However, GEE should be used only when 50 clusters or more are available in the data when working with ordinal outcomes; if the number of clusters were less than 50 for GEE, the researchers should use the small sample size standard error correction.
Ann O’Connell (Advisor)
Jessica Logan (Committee Member)
Minjung Kim (Committee Member)
138 p.
Education; Statistics

Recommended Citations

Citations

  • Lo, M.-T. (2020). Alternative Methods for Modeling Clustered Ordinal Data [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1586792558628762

    APA Style (7th edition)

  • Lo, Meng-Ting. Alternative Methods for Modeling Clustered Ordinal Data. 2020. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1586792558628762.

    MLA Style (8th edition)

  • Lo, Meng-Ting. "Alternative Methods for Modeling Clustered Ordinal Data." Doctoral dissertation, Ohio State University, 2020. http://rave.ohiolink.edu/etdc/view?acc_num=osu1586792558628762

    Chicago Manual of Style (17th edition)

osu1586792558628762
158
© 2020, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.