Missing data in cross-sectional networks – An extensive comparison of missing data treatment methods

Abstract

This paper compares several missing data treatment methods for missing network data on a diverse set of simulated networks under several missing data mechanisms. We focus the comparison on three different outcomes: descriptive statistics, link reconstruction, and model parameters. The results indicate that the often used methods (analysis of available cases and null-tie imputation) lead to considerable bias on descriptive statistics with moderate or large proportions of missing data. Multiple imputation using sophisticated imputation models based on exponential random graph models (ERGMs) lead to acceptable biases in descriptive statistics and model parameters even under large amounts of missing data. For link reconstruction multiple imputation by simple ERGM performed well on small data sets, while missing data was more accurately imputed in larger data sets with multiple imputation by complex Bayesian ERGMs (BERGMs).