Syndicate content

Add new comment

Martin, I do believe our overhauled replication of PK resolves any remaining mysteries. The study has many problems. The most serious appears to be instrument weakness caused by trying to separately instrument male and female borrowing. Combined with undiagnosed outliers, this makes the estimator radically bimodal and unstable. See http://blogs.cgdev.org/open_book/2011/12/bimodality-in-the-wild-latest-on-pitt-khandker.php Pitt's corrections to our initial attempt at replication allowed us this deeper insight. This instability most likely helps explain why leaving out one control variable led us to such different results in the initial attempt. The process of convergence to the truth has been awkward. But human beings are not perfect, and our openness in sharing data and code, created a feedback loop that helped us overcome our limitations. --David