Making friends: A nonexperimental analysis of social pair formation

Abstract

Economists and psychologists often assume that people sort into groups based on similarities. In this article, the validity of this assumption is examined using pairwise data and a new application of a proven analytical technique. I use a unique administrative data set that follows undergraduate students as they move from quasi-randomly assigned housing into self-selected residence groups; social pair formation is measured using behavior, not survey responses. Sorting is observed from the baseline ‘clean slate’ of random groupings, creating a quasi-experimental environment and circumventing problems inherent in examining selection processes when starting groups are not randomly formed. A wide array of covariates, including original campus location, race, and academic ability is exploited to determine the most important factors driving students’ choices about with whom to live. Results support a role for ability in friendship formation, both conditionally and unconditionally, but also a significant and more potent influence of social factors and luck. The article demonstrates a novel approach to the analysis of normally unobserved human sorting processes in a real-world context.