Better off on their own? How peer effects determine international patterns of the mathematics gender achievement gap

This paper applies recent spatial regression techniques in peer effects estimation to a sample of 33 countries in the IAE's TIMSS 2015 study in order to quantify the gender achievement gap in eighth grade mathematics. Based on an education production function setting and controlling for the mediating influences of student-, parent-, teacher-, and school-level factors, a significant, but small gender achievement gap amounting to 10% of a country-level standard deviation is confirmed in the standard linear model. I model endogenous and exogenous peer effects based on the workhorse linear-in-means model as well as on homophily-based dyadic relations between students, both with controls for group unobservables. The results show that the effect size increases to 12-14%, depending on the underlying socio-matrix. However, the partitioned impacts suggest even larger effects when considering spillovers within the classroom. These could amount to as much as 38% of a standard deviation, depending on the underlying dyadic peer weights, with the linear-in-means model possibly overstating the magnitude of classroom externalities.