Cross-classified Multilevel Models for Personal Networks: Detecting and Accounting for Overlapping Actors

Abstract

Multilevel models are increasingly used in sociology and other social sciences to analyze variation of tie outcomes in egocentrically sampled network data, particularly in studies of social support. Existing research assumes that the personal networks in the data do not overlap (i.e., they do not have actors in common), which makes standard hierarchical models suitable for analysis. This assumption is unrealistic in certain sampling designs, including the case of egos sampled from higher level groups or via link-tracing methods. We describe different types of ego-network overlap and propose a method to detect overlapping actors and analyze the resulting data with cross-classified multilevel models. The method is demonstrated with an application to research on personal networks and social support among Hispanic immigrants in rural U.S. destinations. Overlap detection and modeling result in better model fit, more correct partition of tie variation among different sources, and the ability to test new substantive hypotheses.