Imputation for Missing Data in Statistical Matching Using Goal Programming

Abstract

Nearly all common statistical approaches assume complete information for all variables involved in the analysis, which making missing data problematic. Imputation is the process of substituting a missing value with a specific value, and it is most likely the most popular method for compensating for missing item values in a survey. This study suggests use of mathematical goal programming approach to impute missing data in statistical matching. The suggested approach adopts the regression method in imputation of the missing values. The regression coefficients are estimated using an estimated mathematical goal programming approach. The paper studies the cases when having variables with different skewed probability distributions (lognormal, Cauchy, chi square). The results of the simulation study indicate a good performance of the suggested approach in cases of skewed probability distribution .Using goal programming in regression is based on the minimizing the sum of absolute errors which is less affected by outliers compared to sum of squares of errors.

Keywords

• missing data
• imputation
• mathematical programming
• statistical matching