Imputing Missing Values Using Support Variables with Application to Barley Grain Yield

Abstract

Missing values in a data set is a widely investigated problem. In this study, we propose the use of support variables that are closely associated with the variable of interest for the imputation of missing values. Level of association or relationship between the variable of interest and support variables is determined before they are included in the imputation process. In this study, the barley (Hordeum vulgare) grain yield in the semi-arid conditions of Cyprus was used as a case study. Monthly rain, monthly average temperature, and soil organic matter ratio were selected as support variables to be used. Multivariate regression employing support variables, bivariate, kernel regression and Markov Chain Monte Carlo techniques were employed for the imputation of missing values. Obtained results indicated a better performance using multivariate regression with support variables, compared with those obtained from other methods.