Imputation methods for estimating public R&D funding: evidence from longitudinal data

Missing data are a common issue in datasets used for socio-economic research; thus, the implementation, application, and evaluation of imputation methods can lead to benefits in economic and social sciences. The purpose of this paper is to apply and compare the performance of different imputation procedures for a specific and original set of data on national public R&D funding, as well as to identify and evaluate the best method (among those proposed) for longitudinal data. The procedures shown here can be generalized to all social sciences contexts when data are missing or when there are problems of missing data in official socio-economic statistics. Our results indicate that the various imputation methods improve the estimates on the basis of data characteristics. Linear Interpolation fits our data better, while Two-fold Fully Conditional Specification (FCS) seems to be the best approach when the missing values are not in consecutive years, compared to Multiple Imputation by Chained Equations (MICE) and Full Information Maximum Likelihood (FIML) procedures.