Beyond and Below Racial Homophily: ERG Models of a Friendship Network Documented on Facebook

Andreas Wimmer, Kevin Lewis
2010 American Journal of Sociology  
The most often noticed feature of social networks in U.S. society is their high degree of racial homogeneity, which has been attributed to the preference for associating with individuals of the same racial background (i.e. racial homophily). We unpack racial homogeneity using a theoretical framework that distinguishes between various tie formation mechanisms and their possible direct and indirect influences on the sociodemographic composition of a network; a new dataset based on the Facebook
more » ... es of a cohort of college students that contains information related to all of these mechanisms; and exponential random graph modeling that can disentangle their simultaneous effects. We first show that racial homogeneity is produced not only by racial homophily proper, but also by the aggregation effects of ethnic homophily hidden from sight when standard racial census categories are used, and by balancing mechanisms such as reciprocity and triadic closure that amplify the effects of racial homophily. In a second step, we put the importance of racial homophily further into perspective by introducing a comprehensive model of tie formation that estimates the simultaneous effects of a wide range of tiegenerating micro-mechanisms. It turns out that propinquity based on shared academic foci or co-residence and homophily on the basis of other categories than "race" are at least as important for the generation of the overall network structure as is racial homophily. These findings have important implications for our general understanding of how socio-demographic structures affect network composition via multiple causal pathways and the precise role and relevance of racial homophily therein. offered encouragement and advice. We are particularly indebted to Steve Goodreau and Dave Hunter, who gave extensive methodological feedback and generously took our concerns and needs into consideration when producing the latest version of statnet. Bob Hanneman and Mark Newman were equally supportive in helping us to find the most appropriate way to cluster student tastes, and Carter Butts advised us on alternative permutation-based methods to confirm our findings. We are indebted to outstanding AJS reviewers who helped to sharpen and re-focus the argument. Alas, all responsibility for errors of thought or fact remain with the authors. 1 Corresponding author, can be reached at
doi:10.1086/653658 pmid:21563364 fatcat:ww6sz337qzeqdgt77h34e6inbi