A comparison study of tie non-response treatments in social networks analysis

Abstract

Analysis of social network data often faces the problem of tie non-response. Recent studies show that the results of social network analyses can be severely biased if tie non-response was ignored. To overcome the problems created by tie non-response, several treatments were proposed in the literature: complete-case approach, unconditional mean imputation, reconstruction, and multiple imputation. In this paper we assessed the impact of tie non-response on social network analysis and investigated the performance of four treatments to handle tie non-response. The simulation results showed that ignoring tie non-response data in network analysis could underestimate the degree and centralization of social networks depending on the types of network and the proportion of missing ties. We also found that unconditional mean imputation was the best tie non-response treatment. Multiple imputation could successfully correct for tie non-response in a few specific situations. Complete case approach and reconstruction, however, were not recommended. We advocate the importance of further research to better understand consequences of tie non-response in social networks analysis and to provide statistical guidance to researchers to tackle this problem in the field.