Within-classroom grouping based on friendship network: Lessons from a field experiment in Bangladesh primary schools

We investigate an effective way of grouping students within classroom by randomly allocating students into small groups within classroom based on students’ friendship network. Once grouped, students were given a week to complete joint Mathematics homework. We find that students (1) met more often with their group members, (2) studied longer hours together, and (3) performed better in Mathematics examination, when they have more friends in their group. Our findings suggest that a number of reciprocal friends and followers are more predictive of test score than a number of friends named by individual students (i.e. outgoing friendship links).