Promoting Inclusive Markets and Financial Systems
The Impact of Microcredit on the Poor in Bangladesh: Revisiting the Evidence
With corrections from Pitt (2011), we improve our replication of Pitt and Khandker (PK, 1998), the most influential study of the long-term impacts of microcredit. Better replication strengthens our doubts about PK’s causal claims. Problems include: an arbitrary and influential imputation for the logarithm of the treatment when zero; the absence of a discontinuity asserted as a basis for identification; likely endogeneity in the quasi-experiment as a result; econometric evidence of such endogeneity; and instrument weakness arising from PK’s disaggregation of borrowings by gender.
The PK results are robust to fixes for most of these problems. But the instrument weakness is apparently making the estimator sensitive to outliers as well as bimodal: there is a negative-impact mode in addition to the positive one reported in PK. Dropping the 16 highest observations of the outcome of greatest interest, household consumption, eliminates the bimodality and any finding of positive impact. Our experience demonstrates the value of replication, and of sharing of data and code to facilitate it. It shows that sophisticated estimators can obscure specification problems, thus why simpler, typically linear, techniques should be used as checks. And it highlights difficulties of non-experimental evaluations
To access this document, please click here.