Endogenous Social Interactions: Which Peers Matter?

This paper compares endogenous social interactions models to determine which one fits the classroom reality best. The analysis uses data from German 9th-graders and considers the effect of the best and worst peers scores, the peers sum and peers average scores on own achievement. Although each model seems plausible when estimated separately, comparison and a selection test point to the classmates average model, meaning that group-based policies are effective. The worst peers model comes second, followed by the best peers and the sum of peers models. Examination of different-ability students responses to increases of average peer achievement reveals either competition for the first place or last-place aversion. Conditional on own course preferences, own and peer characteristics, spillovers transmit only through cognitive ability. Therefore, regrouping on the basis of characteristics such as immigration background is obsolete. Policies should aim at low achievers in small enough classrooms because only then single-student influences can change the social norm. By improving the average through the worst, the best become even brighter.