Title and Abstract
Reassessment of classic case studies in labor economics with new instrument-free methods
For two classic case studies in labor economics, it is shown how recently developed instrument-free methods yield remarkably narrow asymptotically valid confidence intervals for regression coefficients. These methods achieve set-identification through adopting credible ranges for the correlation between endogenous regressors and model errors. They also provide more decisive evidence on (in)validity of exclusion restrictions than Sargan-Hansen over-identification tests. By simulation it is demonstrated that the popular (though yet scarcely examined) Sargan-Hansen tests will often not detect serious instrument invalidity, whereas instrumental variables based inference (also identification robust variants) suffers severely, even from mild invalidity, especially for weak instruments.
Kiviet, J. F., and S. Kripfganz (2020). Reassessment of classic case studies in labor economics with new instrument-free methods.
Jan F. Kiviet
University of Amsterdam;
University of Exeter