Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
We develop Lagrange multiplier and likelihood ratio statistics to test hypotheses on subsets of the structural parameters in an instrumental variables regression model. The asymptotic distributions of these statistics are robust to instrument quality. A key assumption is, however, that the instruments are valid for the remaining endogenous variables. We show that the statistics lead to 95% confidence sets for the return on education in data from Card (1995) that are considerably different from the confidence sets that result from the 2SLS Wald statistic, which is the common statistic used in the literature. 2004 President and Fellows of Harvard College and the Massachusetts Institute of Technology.