Missing Data in Longitudinal Social Networks

Social data is a tremendously fertile soil for the missings to appear. In most cases, researchers avoid ignoring missings by excluding the cases that contain them and apply estimation or imputation techniques instead. For the past few decades, many algorithms were developed for treating missing data specifically in longitudinal social network data. In this paper, we evaluate the latest multiple imputation method for stochastic actor-oriented models (SAOMs) proposed by Krause et al. (2017). The evaluation takes place on the initially almost complete dataset consisting of the three waves. We generate missings of different nature (MCAR, MAR and MNAR) to estimate the bias of the imputed models depending on the mechanism behind data incompleteness. Keywords: network imputation, social networks, missing data, multiple imputation, statistical modelling