Degree Type

Dissertation

Date of Award

2015

Degree Name

Doctor of Philosophy

Department

Statistics

First Advisor

Jae-Kwang Kim

Abstract

Missing data, or incomplete data, inevitably occurs in many surveys. It is mainly due to nonresponse such that sample units do not fully or partly respond for the survey items. It can be also arisen from sample selection. For example, two-phase sampling can be viewed as a missing data problem in the sense that the study variable is not observed in the first-phase. In truncated data that are intentionally selected by researcher, it will be also missing data problem if we are interested in estimation of non-truncated data properties. Many statistical methods for handling missing data can be categorized into two types based on statistical treatment: one is weighting method and the other is imputation method. The weighting method such as propensity score adjustment that uses response probability as compensation for nonresponse is popular for reducing nonresponse bias. Also, the imputation approach is also prevailed to create complete data for statistical estimation or inference of those imputed data. In this thesis we investigate new statistical methods in both of weighting and imputation methods corresponding to three different missing data situations: (i) propensity score adjustment for nonignorable nonresponse data with several follow-ups, (ii) correlation estimation of singly tuncated bivariate samples and (iii) fractional hot deck imputation for multivariate missing data.

DOI

https://doi.org/10.31274/etd-180810-3968

Copyright Owner

Jongho Im

Language

en

File Format

application/pdf

File Size

86 pages

Share

COinS