The comparison study of imputation methods for missing data under different missingness mechanisms

• Main Authors: Yu-en Lu, 呂喻恩
• Other Authors: Huey-Ing Tzou
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/42735374308816134094

碩士 === 國立臺南大學 === 測驗統計研究所碩士班 === 96 === The purposes of this study are to compare the differences of four imputation method: Maximum Likelihood Estimators MLE、data augmentation、Metropolis algorithm。To compare the imputation efficiency how well these algorithms perform under different missingness mechanisms with different amount of missing data。 Got the empirical data from raw data, and to operate the three missingness mechanisms, MCAR、MAR、and MNAR。Then differing amounts were deleted at random causing MCAR data, which had 30%、40%、and 50%missing data.。The different types of missing data were examined the criterion: bias、RMSD、covariance matrix。The main findings are the following: 1. The results from the MCAR data show the impact of the mounts of missing rate on three multiple imputation methods . Results indicated that the bias was increasing as the mount of missing rate upward. The difference is small between three multiple imputation methods. 2. 30% and 40%missing data don’t be imputed the mean and covariance matrix nearly the empirical data。 For 50%missing data the data augmentation method more efficient than the others algorithm techniques, the covariance matrix is the nearest the the empirical data. The study recommend to think about imputing the MAR data, in this study the Maximum Likelihood Estimators is more efficient than the others algorithm techniques .For MNAR, three imputation methods is less efficient.