Impact of microcredit in Eastern Indonesia : dealing with selection bias

Abstract

The popularity and increasing controversy around microcredit has given rise to the need for rigorous evaluation of its welfare impact. We collected pre- (299) and post-treatment (209) survey data to determine the impacts of a group loan with individual liability on indicators of household welfare in NTT province, Indonesia. Because of the lack of statistical significance of impact estimates due to the short follow-up period of only one year, we focus disproportionally on methodological issues. The main challenge of an evaluation of non-randomized programs, as in our setting, is endogenous treatment selection. We focus on nonparametric methods when dealing with attrition and selection bias. We propose a novel nonparametric test of instrument validity in a general recursive model. In contrast to existing overidentifying restriction tests for linear IV models such as the well-known Sargan test, our test is consistent even when none of the instruments to be tested is valid and can thus be applied when only one instrumental variable is available. This and other tests failed to refute the validity of our new instrument, a dummy indicating whether the household has at least one treated or previously treated acquaintance, when a wealth index was the outcome. Our main finding is that take-up of TLM's group loan has a negative short-term impact on household wealth, indicating loan-induced distress sales of assets.