Efficient peer effects estimators with group effects

We study linear peer effect models where peers interact in groups and individual’s outcomes are the group mean outcome characteristics. allow for unobserved random effects as well observed fixed effects. The specification is part motivated by moment conditions imposed Graham (2008). show that these can be cast terms of a model they lead to class GMM estimators with parameters generally identified long there sufficient variation size or types. also our contains Quasi Maximum Likelihood estimator (QMLE) model, Wald (2008) within Lee (2007) special cases. Our identification results extend insights how assumptions about effects, certain forms heteroscedasticity used overcome reflection problem identifying QMLE accommodate additional covariates valid situations large but finite number different sizes Because general based procedures, using instruments other than binary indicators estimation straight forward. accommodates level spirit Mundlak Chamberlain offers an alternative specifications. This feature significantly extends applicability Graham’s strategy assignment may not correlation controlled observable Monte-Carlo simulations bias decreases size, increases size. prove consistency asymptotic normality under reasonable assumptions.