Multiple Imputation Of Missing Data And Its Improvement

In the statistical investigation, the phenomenon of missing data frequently happened. Missing data may be caused by many reasons, and missing data under different background can bring different influence to the statistical analysis. If you want to enhance the statistical investigation data's quality, you may process the incomplete data set, so as to reduce the impact caused by flaw data. This article studies how to use the different interpolation and the adjustment method to reduce the statistical error caused by flaw value. The research indicated that, in each interpolation method, the multiple interpolations are the best. But when a rectangular continuous data set appears to have a longer tailed than normal distribution or it contains some values that are influential on statistical inference with normal distribution, the multivariate t distribution becomes useful for multiple imputations as an alternative to the multivariate normal distribution. First, when the data have a longer tailed than normal distribution. The multiply imputed data set using the t distribution allow more valid statistical inferences than using normal distribution with some "influential" observation deleted. Second, the t distribution is widely used in applied statistics for robust statistical inferences. Therefore, when an incomplete data contains some influential values or outliers, the t distribution allows for a robust multiple imputation method. Furthermore, the multiple imputation appears more useful that the asymptotic method of inference since the likelihood functions of the parameters of the t distribution given the observed data can have multiple modes. In all, these methods mentioned in the paper are very useful for datasets with missing data.