Evaluating the magnitude and the stakes of peer effects analysing science and math achievement across OECD

What follows is an exercise aimed at estimating peer effects' impact on science and math test scores of secondary school students surveyed in 1995 by the International Education Agency across OECD countries. It is also to discuss their importance for educational policy, particularly regarding the highly sensitive issue of ability-grouping. Using this unique international database. This study assesses the magnitude of the peer effect relative to more traditional inputs. Referring to education policy stakes, we control for the presence of increasing or decreasing return. This study also checks for cross effects in order to determine whether peer effects matter more to low or high SES pupils, and whether their final impact on achievement is affected by the underlying level of heterogeneity within the group. Using a methodology, which a priori accounts for the clustering of the data within countries and schools/classrooms - i.e. fixed/random effect or hierarchical model - our analysis indicates that peer effects are strong determinants of both math and science achievement relative to individual SES and other school inputs. The presence of increasing of decreasing returns is not obvious. But we find systematic evidence that low-ability pupils are more sensitive to peer group characteristics. By contrast, this study also find that - for a given level of the peer effect - higher heterogeneity comes at a certain cost. In brief, these results provide no systematic evidence regarding grouping policies.