On the Sensitivity of Return to Schooling Estimates to Estimation Methods, Model Specification, and Influential Outliers If Identification Is Weak

We provide a comparison of return to schooling estimates based on an influential study by Angrist and Krueger (1991) using two stage least squares (TSLS), limited information maximum likelihood (LIML), jackknife (JIVE), and split sample instrumental variables (SSIV) estimation. We find that the estimated return to education is quite sensitive to the age controls used in the models as well as the estimation method used. In particular, we provide evidence that JIVE coefficients' standard errors are inflated by a group of extreme years of education observations, for which identification is especially weak. We propose to use Cook's Distance in order to identify influential outliers having substantial influence on first-stage JIVE coefficients and fitted values.