Unveiling homophily beyond the pool of opportunities

TL;DR

This paper introduces a new statistical methodology to accurately measure and infer choice homophily in social networks, accounting for structural factors and applicable to various network types, validated through synthetic and real-world data.

Contribution

The authors develop a robust, versatile framework using statistical network ensembles to quantify choice homophily, outperforming traditional metrics especially in imbalanced group scenarios.

Findings

Method accurately captures generative homophily even with additional tie-formation mechanisms.

Triadic closure has minimal impact on homophily inference in networks.

Preferential attachment does not affect the inference results.

Abstract

Unveiling individuals' preferences for connecting with similar others (choice homophily) beyond the structural factors determining the pool of opportunities, is a challenging task. Here, we introduce a robust methodology for quantifying and inferring choice homophily in a variety of social networks. Our approach employs statistical network ensembles to estimate and standardize homophily measurements. We control for group size imbalances and activity disparities by counting the number of possible network configurations with a given number of inter-group links using combinatorics. This method provides a principled measure of connection preferences and their confidence intervals. Our framework is versatile, suitable for undirected and directed networks, and applicable in scenarios involving multiple groups. To validate our inference method, we test it on synthetic networks and show that it outperforms traditional metrics. Our approach accurately captures the generative homophily used to build the networks, even when we include additional tie-formation mechanisms, such as preferential attachment and triadic closure. Results show that while triadic closure has some influence on the inference, its impact is small in homophilic networks. On the other hand, preferential attachment does not perturb the results of the inference method. We apply our method to real-world networks, demonstrating its effectiveness in unveiling underlying gender homophily. Our method aligns with traditional metrics in networks with balanced populations, but we obtain different results when the group sizes or degrees are imbalanced. This finding highlights the importance of considering structural factors when measuring choice homophily in social networks.