Files
Abstract
The Health and Retirement Study (HRS) is designed to have a thorough investigation of retirement policy, health and well-being of the US elderly. The main focus of this thesis is to predict the time to dementia by demographic characteristics and self-respond questions measured in HRS. The Cox proportional hazard (PH) model was adopted in the analysis. However, there is a certain percentage of missingness in the HRS data. In order to ensure the accuracy of the study results, multiple imputation was used to deal with the missing data. Since not all variables are related to the response, we could reduce the number of covariates contained in the final model to improve the precision of model prediction. Stepwise and LASSO selection methods were then conducted on multiply-imputed datasets to select significant variables related to dementia. We evaluate the performance of the methods on selecting important variables by simulation studies.