Schooling and Labor Market Consequences of School Construction in Indonesia: Comment

TL;DR

This paper revisits Duflo's 2001 study on Indonesia's schooling expansion, correcting data issues and testing robustness, finding some persistent impacts on schooling and wages, though with weak overall evidence.

Contribution

It provides a critical reanalysis of Duflo's original estimates, addressing data errors and specification sensitivity to clarify the true effects of the schooling expansion.

Findings

Some impact estimates are robust to revisions.

Returns to schooling are weakly identified.

Persistent 'fingerprints' suggest causality from expansion to wages.

Abstract

Duflo (2001) exploits a 1970s schooling expansion in Indonesia to estimate impacts on schooling and labor outcomes, as well as the returns to schooling. I correct data errors, adjust for potential sources of bias, follow up later in life, and check sensitivity to two specification choices that are not explained in the original text and that are applied to some regressions and not others. Headline estimates from the original, 1995 follow-up are generally not robust to the revisions. Returns to schooling are weakly identified. But certain "fingerprints" from Duflo (2001) specification checks--impacts appearing only in the birth cohorts or at the grade levels where expected--recognizably persist. E.g., impact estimates from kink-based specifications that focus on timing by birth cohort are consistently positive, if not always at conventional significance lev-els. The best explanation, though weakened, is causation from the schooling expansion to individual schooling attainment to wages.