HANDLING OF MISSING DATA FOR E-READINESS INDICATORS

摘要

Missing data is one of the main problems associated with composite indicators of electronic readiness (e-readiness), but the way in which these missing values are processed can have a serious impact on the results of e-readiness assessments. The complexity of this problem increases with the number of missing values. However, despite the known limitations on the performance of some missing data processing methods, such as imputation based on the following year?s values or the average of previous years? values, many composite indices of e-readiness continue to use these methods. The main objective of this article is to improve the estimation of missing data in a dataset used by the Networked Readiness Index (NRI) organisation. In order to improve existing estimates, we establish a predictive model based on multiple linear regressions for each indicator containing missing values. We also use variable selection techniques to choose the best input variables for each model.