ESTIMATING MISSING VALUES IN UNIVARIATE QUANTITATIVE DATA: DISTRIBUTIONAL GOODNESS-OF-FIT APPROACH

Abstract

Missing values occur in almost all research. They are either removed or replaced with estimated value(s) before the data set is further analyzed. In this study, another method of replacing missing values is proposed for univariate quantitative data. The method requires fitting different distributions into the data set and replacing the missing value(s) with the mean of the distribution with the best Goodness – of – fit Statistic. This method is compared with some existing ones using both simulated data from symmetric, negatively and positively skewed distributions and some real life data sets. The Mean Square Error and Mean Absolute Error criteria were used to compare the methods. Results showed that the proposed method is either best or compare favourably with best method especially with skewed data sets.