A Comparison of the estimation methods for missing data in sample survey

Abstract

The objective of this research was to study three imputation methods including, Regression Imputation method (RI), Distance Regression Imputation method (DRI) and Multiple Imputation method (MI). The Monte Carlo simulation technique, conducted for 1,000 replications, was composed of one independent variable (X) and one dependent variable (Y) with normal distribution, the mean of 10 and the variance of 1. The correlation coefficients are 0.10, 0.30, 0.50, 0.70 and 0.90. The sample sizes are 30, 60, 100 and 300. The Percentages of missing at random in the dependent variable are 5,10 and 15. Confidence coefficient is 0.95. Coefficient of variation (CV) and coverage probability (CP) were used as the criteria of comparison. The result of this research showed that the three imputation methods led to similar coefficient of variations and coverage probabilities. When the sample sizes increased, the coefficient of variation would decrease, and the coverage probability would be close to the confidence coefficient of 0.95. Since the Regression method was uncomplicated and easier, so it was the most appropriate imputation method for this research.