Project Details

Method: Quantitative

The overview of the project:

This research investigates the effects of social networks on the educational outcomes of students in Bangladesh. The experimental design involved grouping students within a classroom among grade-four students in rural primary schools. We randomize 150 schools into three different groups: (i) the pure random group (80 schools) where we randomly allocate students into groups; (ii) the friendship group (35 schools) where we group students based on their friendship; (iii) the pure control group (35 schools) where we did not group students at all. Our results suggest heterogeneous effects of grouping. We find that friendship groups perform better than pure-random groups in individual math test among the middle group ability. However, for very high and low ability students, belonging to a friendship group does not necessarily increase own math score. Interestingly, we find that female student's test scores improve significantly more than their male counterparts when grouped with their friends. We also find no evidence of outcome differences between students in the pure random group schools and those in the pure control group schools. We then study the importance of network centrality of a group on own outcomes among students from the pure random groups only. We show that, for the general knowledge test, the average eigenvector centrality of the group matters most for own grade while, it is the Katz-Bonacich centrality and the key-player centrality that mostly affect the math score. Finally, we test peer effects among students in the pure random groups and show that having being (randomly) assigned to a group with high grades does indeed increase own grade in math tests.