Comparing different imputation methods on different types of missing data

• Main Authors: Chien Hung Lin, 林建宏
• Other Authors: Esher Hsu
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/80464779124349884644

碩士 === 國立臺北大學 === 統計學系 === 95 === Social researchers commonly suffer missing data issue when they execute a research project. The lack of information easily influences the researcher making wrong decision on important matter. Therefore, many methods were developed to deal with missing data. Among these methods, imputation is one of popular ways in recent years. Generally, missing data covers several types including MCAR, MAR and non-ignorable missing value. The objective of this study is to provide references for the social researcher applying common imputations to impute missing data. This study selects five main methods of imputation including Em imputation, Regression imputation, Hot-Deck imputation, Multiple imputation and Ratio imputation. Monte Carlo Simulation is used to simulate the process of five imputations for two types of missing data, MCAR and MAR. The study evaluated the effect of each imputation method by “Bias” and “MSE”. EM imputation and ration imputation perform well in the MCAR or MAR type of missing data and suggested to put in use.