Working Papers
The Impact of Microcredit on the Poor in Bangladesh: Revisiting the Evidence - Working Paper 174 (Revised 2011)
6/18/09
Download PDF
This version 12/16/2011
The idea that access to small loans can help poor families build businesses, increase their income, and escape poverty has blossomed into a global movement. Its appeal is manifold. It is at once radical in its suggestion that the poor are creditworthy and conservative in its insistence on individual responsibility.
But how robust is the evidence that microcredit improves the lives of the poor by, for example, increasing or stabilizing household consumption? In this working paper, senior fellow David Roodman and NYU's Jonathan Morduch reevaluate an influential study by Mark Pitt and Shahidur Khandker using Roodman’s cmp program, and show that it does not convincingly rule out reverse causation. A positive association between microcredit and household spending, for example, may merely indicate that richer families borrow more. With these studies in doubt, solid academic evidence that microcredit reduces poverty is even scarcer than previously understood. For non-experimental methods to retain a place in the program evaluator’s portfolio, the quality of the claimed natural experiments must be high and demonstrated.
The first version of this paper, released in 2009, attempted to replicate three studies done using the same data from Bangladesh. In 2011, Mark Pitt identified two important discrepancies in the replication of the most influential of the three, the one by Pitt and Khandker. The 2011 version incorporates Pitt's corrections, gains a deeper understanding of the original paper's statistical problems, and, because of the complexity of the analysis, focuses just on it.
Related documents:
- A point-by-point reply to Pitt's 2011 critique of the 2009 version of this paper.
- The 2009 version.
- A complete copy of Morduch's 1998 paper, "Does Microfinance Really Help the Poor: New Evidence from Flagship Programs in Bangladesh."
Files needed to generate the results in the revised version of the paper:
Data sets
- Roodman & Morduch HH 2.zip—Stata file with one observation per household and survey round
- Roodman & Morduch ind 2.zip—Stata file with one observation per individual and survey round
- variable list.xls and variable list.pdf—describe the variables in these data sets
- RM db 2.zip—Microsoft SQL Server database that transforms the primary survey data into the Stata data sets. Also contains Morduch’s (1998) data set, and a partial data set obtained from Mark Pitt in 2008. Requires SQL Server (free version) to view
- Pitt data set 2011.zip—Stata data set posted by Mark Pitt in 2011
- PK 98 results 2.xls—Spreadsheet to transform raw regression result files into displayed tables. If prompted by Excel, data connections (embedded in last two tabs, to regression output files) can be left disabled.
Stata .do files
- RDD 2.do—Figure 1 in response to Pitt and Figures 1--3 in revised working paper
- PK reproduction 2 routines.do—core routines for data preparation and estimation, called by next two files
- PK reproduction 2.do—all tables except Table 8 in the revised working paper
- PK bimodality.do—Table 8 and Figures 4--7 of revised working paper
- ivreg2 version 3.0.06
Files needed to generate the results in the 2009 version of the paper:
Primary data not available at the World Bank site, obtained from Mark Pitt in the late 1990s
- CPIYEAR.DTA—Stata file with Consumer Price Indexes by month and region of Bangladesh
- LANDREL1.DTA—Details of relatives’ landholdings
- WGHT123.DTA—Sampling weights
Data sets
- Roodman & Morduch HH.zip—Stata file with one observation per household and survey round
- Roodman & Morduch ind.zip—Stata file with one observation per individual and survey round
- variable list.xls and variable list.pdf—describe the variables in these data sets
- RM db.zip—Microsoft SQL Server database that transforms the primary survey data into the Stata data sets. Also contains Morduch’s (1998) data set, and a partial data set obtained from Mark Pitt in 2008. Requires SQL Server (free version) to view
- Pitt data set.xls—Data set provided by Mark Pitt
Stata .do files
- Compare R&M to Pitt.do—correlations in Table 1
- RDD.do—Figures 1, 2, and 8 and fuzzy RDD regressions reported in text
- Borrowing-consumption lowess regressions.do—Figures 3-7
- PK 98 reproduction.do—replication of Pitt and Khandker 1998
- Estimate from Pitt data set.do—checks reported in text
- M 98 reproduction.do—replication of Morduch 1998
- K 05 reproduction.do—replication of Khandker 2005
- cmp on Pitt simulated data set.do—Table 11, appendix
