Description:
In the research on peer effects, unweighted mean classroom performance is the predominant measure used in the estimation of education production functions. In this paper, based on the sociological concept of homophily, I introduce social network matrices that correspond to a weighting scheme for peers in the same class at school. Using spatial regression techniques, I confirm the presence of peer effects for the eighth grade population in the USA in the TIMSS 2015 student assessment. For students, the likelihood of cooperation increases conditionally on visible and non-visible characteristics, such as age, gender, migratory background, and attitudes towards scholastic achievement. This grouping behavior is found to affect the spillover effects of student variables, such as gender and language skills. The main findings are robust to various definitions of the social network matrix, as well as to the inclusion of teacher fixed effects.