Replicating Replication : Due Diligence in Roodman and Morduch’s Replication of Pitt and Khandker (1998)
"The Impact of Microcredit on the Poor in Bangladesh: Revisiting the Evidence," by David Roodman and Jonathan Morduch (2011) is the most recent of a sequence of papers and postings that seeks to refute the findings of the Pitt and Khandke...
| Main Authors: | , |
|---|---|
| Format: | Policy Research Working Paper |
| Language: | English en_US |
| Published: |
World Bank, Washington, DC
2013
|
| Subjects: | |
| Online Access: | http://documents.worldbank.org/curated/en/2012/11/16985792/replicating-replication-due-diligence-roodman-morduchs-replication-pitt-khandker-1998 http://hdl.handle.net/10986/12111 |
| Summary: | "The Impact of Microcredit on the
Poor in Bangladesh: Revisiting the Evidence," by David
Roodman and Jonathan Morduch (2011) is the most recent of a
sequence of papers and postings that seeks to refute the
findings of the Pitt and Khandker (1998) article "The
Impact of Group-Based Credit on Poor Households in
Bangladesh: Does the Gender of Participants Matter?"
that microcredit for women had significant, favorable
effects on poverty reduction. In this paper the authors show
that these latest Roodman and Morduch claims are based on
seriously flawed econometric methods and theory and a lack
of due diligence in formulating models and interpreting
output from packaged software. On the basis of Roodman and
Morduch's preferred two-stage least squares regression,
an alternative calculation of the standard errors would lead
one to conclude that the problem with Pitt and Khandker is
that they underestimate the positive and statistically
significant effect of women's credit on household
consumption. As in their previous efforts, the methods of
Roodman and Morduch are shown to bias the findings in the
direction of rejecting the results of Pitt and Khandker. We
also further examine two aspects of our instrumental
variable approach that have been attacked by Roodman and
Morduch. The first is the validity of the exclusion
restrictions underlying the use of interactions between
program choice and the set of exogenous variables (including
the village fixed effects) as instruments. The second is the
application of the "one-half acre" program
eligibility rule. The authors show that identification does
not require both of these, and present new results dropping
each assumption in turn. The results originally reported in
the Pitt and Khandker paper hold up extremely well in this
new analysis. |
|---|