This paper identifies the effect of neighborhood peer groups on childhood skill acquisition using observational data.We incorporate spatial peer interaction, defined as a child's nearest geographical neighbors, into a production function of child cognitive development in Andhra Pradesh, India. Our peer group definition takes the form of networks, whose structure allows us to identify endogenous peer effects and contextual effects separately.We exploit variation over time to avoid confounding correlated with social effects. Our results suggest that spatial peer and neighborhood effects are strongly positively associated with a child's cognitive skill formation. Further, we explore the effect of peer groups in helping to provide insurance against the negative impact of idiosyncratic shocks to child learning. We find that the data reject full risk-sharing, but cannot rule out the existence of partial risk-sharing on behalf of peers. We show that peer effects are robust to different specifications of peer interactions and investigate the sensitivity of our estimates to potential misspecification of the network structure using Monte Carlo experiments.
This paper identifies the effect of neighborhood peer groups on childhood skill acquisition using observational data.We incorporate spatial peer interaction, defined as a child's nearest geographical neighbors, into a production function of child cognitive development in Andhra Pradesh, India. Our peer group definition takes the form of networks, whose structure allows us to identify endogenous peer effects and contextual effects separately.We exploit variation over time to avoid confounding correlated with social effects. Our results suggest that spatial peer and neighborhood effects are strongly positively associated with a child's cognitive skill formation. Further, we explore the effect of peer groups in helping to provide insurance against the negative impact of idiosyncratic shocks to child learning. We find that the data reject full risk-sharing, but cannot rule out the existence of partial risk-sharing on behalf of peers. We show that peer effects are robust to different specifications of peer interactions and investigate the sensitivity of our estimates to potential misspecification of the network structure using Monte Carlo experiments.