Document Type
Dissertation - University Access Only
Award Date
2010
Degree Name
Doctor of Philosophy (PhD)
Department / School
Mathematics and Statistics
Abstract
Longitudinal data analysis has become a popular statistical method. In order to produce accurate analysis, it is better to analyze longitudinal data with a complete dataset. Even in well-controlled situations, missing data often occurs in longitudinal studies. That is why the imputation methods are important and essential to learn. The effectiveness of these methods for different data structures has not been well studied. In this dissertation, four commonly used imputation methods are compared. Then, a simulation study and datasets in literature are conducted to evaluate the performance of imputation methods under a variety of circumstances. When the complete dataset is available, Missing Completely at Random (MCAR) is created and imputation methods are used to predict the missing values and analyze the mean of empirical means for each dataset. The experiments are concluded by outlining the conditions for each imputation method to produce reasonable and efficient statistical analysis. This dissertation emphasizes the need for improving the methodology for handling missing data when using imputations methods in longitudinal analysis.
Library of Congress Subject Headings
Longitudinal method
Missing observations (Statistics)
Multiple imputation (Statistics)
Format
application/pdf
Number of Pages
131
Publisher
South Dakota State University
Recommended Citation
Nakai, Michikazu, "Comparison of Imputation Methods for Longitudinal Data with Missing Values" (2010). Electronic Theses and Dissertations. 1677.
https://openprairie.sdstate.edu/etd2/1677