UC Berkeley

We apply Random Forests to detailed survey data of social relations in order to derive an inductive typology of egocentric networks. Beginning with over 40 descriptors of 1050 northern California respondents' networks, we combine 21 of these into seven dimensions, the extent to which those networks display: (1) interaction with nonkin, (2) proximity to kin, (3) overall involvement with kin (including support), (4) support from nonkin, and the extent to which (5) church, (6) work and (7) extra-curricular activities shape connections with others. We use these dimensions to reliably place 985 of the 1050 observations into types: career-and-friends (24%), family-and-community (20%), family-only (16%), untethered (8%), energetic (7%), withdrawn (6%), and home-and-church (5%). In the second part of the analysis, we describe the social and demographic attributes of respondents that predict membership in each cluster to present a richer picture of the network typology, as well as to confirm that the types have face validity.