Advances in Network Clustering and Blockmodeling

7Treating Missing Network Data Before Partitioning

Anja Žnidaršič¹, Patrick Doreian^2,3 and Anuška Ferligoj^2,4

¹University of Maribor

²University of Ljubljana

³University of Pittsburgh

⁴NRU HSE Moscow

7.1 Introduction

The patterns of ties of a network are critical for revealing both macro and micro network structural features. This is especially true regarding the delineation of the macro structure of networks through network partitioning. Yet network data are prone to recording errors and/or missing data regardless of the substantive nature of the relationships measured. If specific real ties are not recorded and non-existent ties are recorded as if they are real, this creates major problems for analyzing network data. Also, correctly included ties can have incorrect values. Actors may refuse to respond regarding specific ties and can provide no information about ties to all other network members. The latter is known as actor non-response. It is crucial that the data used for clustering procedures are either error-free (a very rare event) or are treated appropriately when data were missing. Here, we continue an examination of the impact of actor non-response and treatments for it on the stability of partitions of actors obtained from different blockmodeling procedures. We use a set of real well-measured networks as the foundation for our analyses.

Missing network data take several forms (see Section 7.2 for more details). Previous studies tackled the problem of actor non-response ...