Continuing the quest for evidence of enough return on investment in IGAs to increase household income.
My last post (#26) reviewed the conclusions of the AIMS Project, especially the seminal paper by Sebstad and Cohen, which concluded 12 years ago that microfinance is more likely to help the poor manage risk (through consumption-smoothing and asset accumulation to protect against financial shocks) than to significantly increase their households’ incomes and consumption.
Continuing to draw from a paper written a decade ago by my Freedom from Hunger colleagues Barbara MkNelly and Mona McCord, let’s see what the research of Mark Pitt and Shahidur Khandker had to say about the effects of microcredit on incomes and consumption.
The first round of field research (funded by the World Bank and reported in a 1998 paper in the Journal of Political Economy) focused on the household and intra-household impacts of the Grameen Bank and similar credit programs in Bangladesh. A methodological contribution of this study was that the analysis was able to separate the estimates of the impact of borrowing by men and by women. This study was also noteworthy for the design and econometric techniques that were used to minimize the potential biases and confounding influence of self-selection and endogeneity of study variables. (Or so it seemed to those of us who could not understand the very complex methodology.)
The study concluded that program participation had a positive effect on household expenditures, asset accumulation, self-employment, children’s schooling, food consumption and contraceptive use. Credit provided by the Grameen Bank had the most significant impact on variables associated with household wealth, women’s power, girls’ and boys’ schooling, women’s labor and assets and total household expenditure.
However, the effect was greatest when women were the program participants, even in terms of raising household expenditures. A dollar loaned to women raised household expenditures by a greater absolute amount than did a dollar loaned to men. Thus, it seemed that the gender of the borrower was also very influential on the ultimate outcomes. Pitt and Khandker concluded that program participation benefits the poor, especially women and children.
But that was not the end of the study and its analysis. David Roodman tells the story well in Due Diligence (pp. 160-65).
Jonathan Morduch reanalyzed the same data with a somewhat different method and got very different results. In response, or perhaps as planned all along, Khandker led a second round of field work to study how changes in microcredit use over the 1990s affected changes in household spending over the same years. His results seemed to more or less confirm the very positive results from the earlier study and refute Morduch’s analysis. Then Roodman joined forces with Morduch to reanalyze the data from the second round and got the opposite result from Pitt and Khandker. They found that Pitt and Khandker had pushed the complex statistical procedure they adopted to the point where it became unstable, tending to flip from very positive results to very negative results due to small changes in the data.
The conclusion? Don’t trust a non-randomized econometric study until you can replicate it and scrutinize it on your own computer, as Roodman did. Good luck with that!
Here’s the real conclusion according to Roodman: “These studies can show correlations but cannot credibly prove causation.” Morduch and Roodman are not denying the validity of the correlations between microfinance use by women and improvement of their household income, only that microfinance participation may not be the driving cause of improvement. A different methodological approach is needed to “prove” causation.
Don’t be too soon discouraged. Let’s keep looking – in the next posts.