Uses and limitations of dichotomous aggregate relational data

Abstract

Aggregate relational data (ARD) on relationships between individuals and subgroups have been informative for studying egocentric network size and degree distributions, assessing segregation in contact with subpopulations, and estimating the size of unlisted groups. Scale-up models for ARD usually assume survey data giving counts of acquaintances in subpopulations, but a closed-ended response format that asks respondents to select a category covering a range of counts may be less burdensome. The simplest (dichotomous) such format distinguishes between having one or more acquaintances in a subpopulation and having none; many existing position generator data take this form. We assess the potential of dichotomous ARD by adapting existing methods for inference from count ARD to accommodate such data. We find that they permit estimation of degree distributions for basic scale-up models, though estimates are less precise than those from count ARD. Dichotomous ARD do not contain sufficient information to model segregation, however. These limitations may be addressed by estimating segregation using respondent heterogeneity on observed covariates, and/or by using a slightly expanded closed response format (0, 1, or two or more acquaintances). Our results suggest how the applications of position generator data – which typically collect dichotomous ARD – might be broadened to encompass questions studied using ARD.