Preference or opportunity? Why do we find more friendship segregation in more heterogeneous schools?

TL;DR

This paper investigates how perceived friendship segregation in heterogeneous schools may be influenced by modeling biases, showing that actual preferences for same-ethnicity friendships are consistent regardless of school composition.

Contribution

The study demonstrates through simulations that exponential random graph models can misrepresent homophily preferences depending on school composition, highlighting potential interpretative biases.

Findings

p* coefficients are larger in schools with more minorities despite constant homophily preferences

Modeling biases can lead to overestimating segregation in heterogeneous schools

Simulated data confirm that actual preferences remain unchanged across different compositions

Abstract

Research on friendship networks in schools suggests that heterogeneity increases homophily preferences. We argue that this may be a misleading interpretation of the coefficients of the exponential random graph models (p*) that are used to model the network data. If students wish to avoid having no friends at all, then minority students may appear to be willing to integrate more with the majority in a more homogeneous school, even if the preference for having same ethnicity friends has the same strength in all schools. We use a random utility model of network formation to study in computational experiments the effects of the preferences that drive network choices. We generate simulated networks for different school compositions but with the same parameter for a preference for homophily. We estimate with a p* model on these simulated data the coefficients for the effect of same ethnicity on (simulated) friendship choice. The tests confirm that p* coefficients for the effect of same ethnicity are much larger in populations with a large minority, although actual homophily preferences are the same across all simulated school compositions.