Abstract:
|
Most of the existing research about the choice of missing data method for non-normal data has been carried out using binary data. This study however uses ordinal data to compare the different approaches of listwise deletion, mean imputation, and multiple imputation to determine how informative each method will be within an ordinal multinomial logistic model. Imputing categorical variables which are non-normal is challenging and it still is unclear which approach should be preferred (Lee et al., 2012). Considering the type of missing data (MCAR, MAR, or MNAR) is also important in determining how to handle missing values. In this study, after learning about the type of missingness by applying a logistic regression, an ordinal multinomial logistic regression is fitted to the ordinal data and within that model, different approaches of missing data are performed to evaluate the appropriateness of missing data handling procedures. This comparison is done by applying these methods to a dataset on the length of stay for people with severe mental illness at a live-in healing community in North Carolina, which includes longitudinal ordinal and multinomial data containing missing values.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.