In almost every news story about the impact of micro-credit, one will find an introductory paragraph mentioning the Bangladeshi entrepreneur Muhammad Yunus and his Grameen Bank. Mr Yunus received the Nobel Peace Prize in 2006 for the novel approach taken by his Bank in tackling endemic poverty. The logic of microfinance is simple: giving very small loans to the poorest of the poor – to those whom otherwise would deemed too risky by conventional banking and loan institutions – will help them escape the crippling cycle of poverty and allow them to earn a better standard of living for themselves.
As the Nobel Peace prize committee wrote:
“Lasting peace cannot be achieved unless large population groups find ways in which to break out of poverty. Micro-credit is one such means. Development from below also serves to advance democracy and human rights.”
The Grameen Bank microcredit model has subsequently been replicated around the globe with around 7,000 microcredit institutions serving over 25 million poor clients worldwide.
From a theoretical standpoint, this logic is intuitively appealing. Giving access to credit for the poor can spur economic development as more investments can now occur.
Echoing the writings of Jeffrey Sachs, poverty traps occur when individuals without any financial collateral (property, land, steady income) are deemed too risky to be granted credit.
Following a more Will Easterlian prescription, micro-loans are promoted as a superior alternative to foreign aid as they have embedded incentives.
Furthermore, is the borrowers themselves who know best how to use the money. They will be more motivated to invest the money optimally as the borrower must generate enough surplus to pay back the loan plus interest.
The microcredit model is appealing by virtue of its simplicity: access to credit can encourage entrepreneurs and business expansion, therefore stimulate employment opportunities – especially among the under-utilised female labour force in many developing countries.
It is also claimed that microloans can increase child school attendance at primary and secondary level, as the opportunity cost of keeping children out of school to work and earn for their parents income decreases.
As a result of this, many support the view that it could break the vicious cycle of poverty.
The empirics may not support the claims
While many parties paint microcredit is a panacea, conflicting reports have been recently emerging in the media and from academic sources. Many argue that in reality much credit is spent on consumption, rather than long term investment .
Over-indebtedness is a chronic problem in many countries, where the problem is not a shortage of credit to the poor, but an endemic surplus .
Finally, the issue of debt has been highlighted by stories of borrower suicides in India .
In conjunction with the shift from first generation charity microcredit programmes to second generation “for-profit” credit institutions, many ask if micro-loans could be causing more harm than good. In order to ascertain whether microcredit is an effective way to transform the lives of poor people and to help them escape the cycle of poverty, one must critically engage with the empirical record over the past three decades to see if this is true.
One such attempt to empirically scrutinise the effectiveness of microcredit was conducted by the World Bank, in partnership with the Bangladesh Institute of Development Studies (BIDS). From 1991 to 1992, these two organisations carried out large-scale household survey on microcredit. The survey covered 1,798 households randomly chosen from over eighty villages in Bangladesh; some with access to micro-credit and some villages with no such credit institutions available.
One caveat, when viewing results from surveys like this, is that it is difficult for individuals – especially those in communities with low financial or numerical literacy – to accurately recall their previous level of income a year ago compared to now. If asked “Did your income increased since last year?”, individuals honestly may not be able to remember if their income did in fact rise.
Additionally as one analyst argues, “borrowers could lie [and] want to say yes in order to endorse the microcredit programme, to prevent other members in their loan group from thinking they could have future repayment difficulties” (Goldberg, 2005: 15). The large-scale survey method on self-reported income change is not an infallible instrument.
From these above mentioned World Bank survey results, the Grameen Bank in Bangladesh extracted one of its most widely articulated statistic: that the average household income of Grameen Bank members is about twice as high as the target group in the control village, and 25% higher than the target group non-members in Grameen Bank villages (from Chemin, 2008: 463).
This seems like an impressive figure, which adds substance to the theoretical argument that microcredit can increase income and help borrowers escape the cycle of poverty in the developing world. However, these large figures do not truly tell us if microcredit works. While the findings follow a sound theoretical rationale and intuitively make sense, they do not address the practical problems of their research design. It is very likely that there is a systematic difference between the people who volunteer to become members of the Grameen bank and the non-member individuals who choose not to opt in.
The self-selection bias: are people who choose loans qualitatively different to those who do not?
How can we say without reservation that Grameen members would not have otherwise increased their income in the absence of the bank microcredit option? Bangladesh has seen continual GDP growth since its independence  and maybe people who seek credit to start a business would have increased their income, regardless of whether there was microcredit institutions in their village/district.
Those individuals who chose loans could all possess some entrepreneurial spirit, ambitious work ethic or delayed gratification restraint that would have ensured their incomes rose under any circumstance as the Bangladeshi economy grew.
This possibility cannot be removed by merely comparing income levels of Grameen Bank members to non-members because the “entrepreneurial” variable is not equal in both groups and therefore could be the reason for any difference, rather than the difference being down to microcredit.
Khandker, a Bangladeshi economist and World Bank researcher, and his colleague Mark Pitt, an American professor in economics, conducted extensive analysis of the World Bank survey’s findings in 1998. In attempts to nullify the prevalence of self-selection bias, they employed the technique of comparing the income of micro-credit borrowers in villages with access to micro-credit institution, to the income of similar “comparison” individuals from control villages without access to microcredit. This design hoped to remove the self-selection bias. They hoped in this design, the analysis would capture the individuals whom the authors predicted would have been most likely to seek out credit if they had been given access to credit banks.
They used fixed effects or “instrumental” estimators to act as control variables for unobservable phenomena they determined would be related to the borrower’s selection of credit programmes. One such instrumental variable measured whether the individual in the control village owned less than half an acre of land (this is the cut-off of eligibility for many of the microcredit groups such as Grameen Bank and Bangladesh Rural Advancement Committee, who do not offer loans to people with more than half an acre of land). Therefore, this particular exogenous criterion of eligibility for individuals into a programme was used as a means of avoiding the pervasive self-selection bias when making comparisons.
The results of Khandker and Pitt’s study were resoundingly supportive of the benefits of micro-credit for alleviating poverty. They found that when a Bangladeshi woman borrowed 100 taka (around €1.20), she would subsequently generate a further 18 taka (22c) in terms of household income in the year with help of the loan (male borrowers generate on average 11%). Khandker (1998: 56) inferred from these figures that annually, Bangladesh’s micro-credit programmes could deliver 5% borrowers from poverty.
However, there were critics to the Khandker and Pitt research design. Most indicated skepticism about the use of the instrumental variable concerning cut off points. Angus Deaton laments how instrumental variables have “moved from being solutions to well-defined problems of inference to now being used as devices that induce quasi-randomisation” (Deaton: 2009: 2).
Researchers who employ instrumental variables often make assumptions about what they can operationalise to capture unobservable variables of interest, which other researchers frequently refute as erroneous. Finding an instrumental variable to measure likely borrowers is very difficult. In fact, Banjeree and colleagues (2015: 2) study found a very low r squared figure (less than 5% in a model of borrower correlates), demonstrating that it was difficult to predict characteristics of potential borrowers, even with a large list of variables.
American economist, Jonathon Morduch, found fault with the methodology of the Khandker and Pitt’s (1998) paper on the grounds of the instrumental variables used. Examining the instrumental variable that limited micro-credit eligibility to individuals with under half an acre of land, he argued that inconsistently using fixed effects estimators can exacerbate rather than minimise selection bias between groups under investigation. Morduch (1998) found – in over 30% of the cases – borrowers had well over this land ownership limit (one borrower held 13.4 acres of land). Meanwhile, all the “matched” individuals in the comparison group invariably had under 0.5 acres of land. Hahn (2001) calls analysis in which treatment groups do not depend on a particular criterion such as acreage a “fuzzy design” (2001: 202). To tackle this, Morduch employed a difference-in-differences approach comparing eligible to ineligible individuals in villages with access to microcredit against the difference between eligible and non-eligible in control villages. Against previous research findings, Morduch’s approach found no statistically significant differences in income or consumption as a result of access to microcredit. Morduch (1998) did however find support for Khandker and Pitt’s (1998) finding that borrowers with micro-credit experience statistically significant lower levels of consumption and labor supply volatility throughout the year.
One must take the results from the findings of researchers such as Khandker and Pitt, Murdoch (also Roodman and Murdoch (2009) and Chemin (2008) who use the same dataset) with caution, when they use a single sample of households from two surveys in Bangladesh to extrapolate to the entire enterprise of microcredit. It does not offer a full picture nor can it satisfactorily answer the question of whether microcredit alleviates poverty around the world in different settings. Filling this gap in 1998 and 1999, the American Agency for International Development went outside Bangladesh and carried out evaluations of the poverty-reduction effects of micro-loans in Honduras (Edgecomb & Garber, 1998) and in Mali (McNelly & Lippold, 1999). These studies were designed to pilot low-cost alternatives to the more labour-intensive and far more expensive evaluations of micro-credit previously employed in Bangladesh. Their methodology included comparing incumbent borrowers to both incoming borrowers and people who had dropped out of the programme. They argues that any difference between these two groups would illuminate the “impact” of the micro-credit. The findings found differences between the borrowers and non-borrowers, but they did not reach statistical significance in the Mali study. One caveat with this study was the very small sample sizes, especially compared to the large World Bank survey sample size in Bangladesh.
Naturally, the new methodology of the study was criticised, especially the operationalisation of the comparison groups. Dean Karlan, American economist from MIT, holds that the difference-in-difference methodology is a contentious research design, based on too many untested assumptions about the different groups. When examined empirically, the assumptions often do not stand to scrutiny. For instance, he argues this approach “assumes that individuals who drop out of credit schemes have mean income and consumption levels that don’t differ significantly from those who remain. This approach assumes that dropouts are not made worse from taking out micro-credit. Finally, this approach also assumes that when lending groups establish themselves they do not sort themselves by economic background” resulting in another unaccounted factor that could confound results. (Karlan, 2001: 78). Furthermore, erroneous estimates of impact could be generated at the village-level, as credit institutions often select less risky village to set up the scheme in the area and subsequently expanded to villages that often differ systemically. And “veteran” clients who chose to access credit earlier may differ systematically from those more caution borrowers who waited before joining. Thus any comparison of the two groups obfuscates this difference. Karlan argues that the American Agency for International Development studies using dropout cases as comparisons actually exacerbates these biases. Thus, choosing efficiency over accuracy with this novel research design, is too great a cost according to Karlan, as they do not generate reliable results concerning any “impact”.
Given the controversy to the various methods used to measure the comparison groups, Khandker (2005) and Khandker and Samad (2013) updated the previous research design, and added to the original World Bank survey data from 1992. The World Bank continued to track families for a remarkable twenty years with a new round of household surveys in 1999 and in 2010. Using this extensive panel data, Khandker and Saman (2013) did not include comparison groups based upon the heavily contested assumptions. Roodman disagrees and argued that “exploiting the panel dimension does not compensate for the lack of clearly exogenous variation in the treatment variable” (Roodman: 2009). He argued that systematic biases persist and causation cannot be extracted from the correlations that Khandker found.
Randomised methodologies: breaking away from traditional research design
As the main literature from the microcredit enterprise was subjected to years of debate, refutation and re-analysis, it was very clear that the methodologies used were unsuitable in capturing the real impact of microcredit on poverty reduction. The main problem with all previous studies was an over-reliance on descriptive statistics, impact evaluations anecdotal “proof” and most importantly, that too many assumptions about the people studied were being made. It was proving impossible to distill causation from correlations in the presence of complex endogenous variables, and extricate the actual impact of the microcredit,
Starting in the early 2000s, randomised methodologies departed from these previous methodologies that. To rectify the flaws of previous research design approaches, researchers began to conduct randomised control trials (RCT) to contribute rigorous evidence isolate true causality and discern whether microcredit is an effective way to transform villages and help individuals escape poverty. According to Esther Duflo, one of the leading researchers using RCT, along with Banerjee, she argues:
Creating a culture in which rigorous randomised evaluations are promoted, encouraged, and financed has the potential to revolutionise social policy nowadays just as randomised trials revolutionised medicine during the 20th century” (Duflo, 2005: 8).
Concerning the microcredit enterprise, randomised control trials ensure that there is true randomness in the provision of credit-institutions at the village-level. In terms of research questions, these studies mostly focus on finding answers for questions such as: does the investment of borrowers from microcredit institutions translates into increased business profit, household income or increased consumption? (which are often treated as proxies for rises in one’s standard of living). If these trends are observed, this would support the idea that micro-credit realistically and substantively contributes to poor people’s capacity to escape the cycle of poverty they would otherwise perennially find themselves in.
Criticisms levelled against the randomised control trial paradigm often focus on the short time horizons under investigation (see Deaton, 2009). Due to financial constraints, issues of practicality or ethical concerns (Blattman, 2004), studies operate in short term as the conditions must be controlled and outcomes measured. Many effects of micro-credit – such as long-term returns generated from investing in education or machinery assets – will not be reflected in studies that depend on the treatment and control group staying isolated for specific periods of time to get “uncontaminated” results after a specific time span. Contingent on the possibility that researchers can ensure the two groups under investigation remain different, future studies can examine other important correlates of micro-finance that may not manifest for years, including diffuse spillover effects and externalities.
In attempts to address this criticism, Banerjee, Duflo, Glennerster and Kinnan (2014) added follow-up investigations a year after they originally conducted a randomised control trial to investigate the impact of micro-credit in India. What differentiates methodology of randomised control trials from those employed in previous studies is that the analysis is not comparing borrowers to non-borrowers. Rather, because the areas which offer micro-loans are randomised, they compare all those surveyed in treatment areas – which encapsulate both individuals who did not borrow as well as those who did – against all borrowers and non-borrowers in control areas.
Spandana, an Indian micro-credit charity foundation operated in fifty two randomly selected villages. Results from a randomised control trial study in 2007 showed that 27% of households in villages with access to Spandana reported taking micro-credit. In controlled villages, 18% of the households availed themselves of other credit. They followed up the study one year after the micro-credit programme began and measured an increase in small business investment and profits of pre-existing businesses. They found, however, that consumption did not significantly increase. Also in this follow-up study, they found that access to microcredit had no statistically significant effect on household expenditure in the treatment village, compared to the control village. As David Roodman summarised these results, as succinctly as humanly possible:
“In 2007-–08, one year after Spandana began operating in some areas of Hyderabad, among households that had lived in their area for at least three years and had at least one working-age woman, those in Spandana areas saw no changes in empowerment, health, education, and total spending, on average, that were so large as to defy attribution to pure chance, compared to those in areas Spandana would soon expand into (as distinct from areas Spandana avoided for having too many geographically transient workers).
Got that? It is a mouthful of qualifiers”.
When articulated this way, one can understand the criticisms that randomised control trials suffer from external validity. Yes, it is an interesting finding that contradicts the more optimistic assertions of previous research by Khandker and colleagues, but is it possible to extend these results to all poor people’s experiences with micro-credit around the globe?
One positive result that they found was credit access did increase expenditure on assets such as sewing machines and new business ventures, while cutting back on “temptation goods” such as gambling and tobacco. This demonstrates that microcredit can have positive effect. But again, we cannot generalise these results too much. The paper lacks a tenable theory as to why these trends in consumption differed between groups. If future studies want to replicate these findings in different contexts or at different times, they will need to formulate sound theoretical rationale for this finding to be generalisable beyond rural India and useful to practitioners and policy-makers interested in consumption patterns of poor borrowers in their own country’s context.
In January 2015, the most recent series of randomised control trials studies examining the impact of micro-credit on 37,000 individuals in total, greatly contributed to the project of subjecting microfinance to empirical scrutiny. Randomised control trials are very expensive endeavours and only the big names in international development can realistically garner the funding through organisations such as J-PAL lab in MIT , which some commentators have called Freakonomics-style research . The lab claims to search for counter-intuitive findings to challenge received wisdom about development tool and policies.
The sheer breadth of the studies conducted by Banerjee and colleagues from 2003 to 2013 – including Bosnia & Herzegovina (Augsburg, De Hass & Harmgart), Mongolia (Attanasio, Britta, De Haas & Fitzsimons) Morocco (Crepon, Florencia, Duflo & Pariente), Mexico (Angelucci, Karlan & Zinman), Ethiopia (Tarozzi, Jaikishan & Johnson) and India (Banerjee, Duflo, Glennerster & Kinnan), in both rural and urban localities – goes far in mitigating the common criticism that randomised control trials cannot be externally valid; in fact most result patterns were consistent across markets and social landscapes in all six countries. At present, research teams are carrying out seven- and eight-year follow-up studies in India and Morocco to measure micro-credit’s long-run impacts.
One drawback of the RCT method is that it cannot comprehensively investigate the impact of micro-credit before the credit institution began this particular programme in the area (i.e. pre-trial influences). Lenders may carry out impact evaluation only after several years of being in operation, so there is no baseline against which researchers can compare their findings. Diminishing margins of return could be at play here and interfering with results. Additionally, because the mean differences between groups is observed at the village-level (the chosen level of randomisation), the results cannot readily discern individual variance at the sub-village level. Randomised evaluations of projects are useful for garnering a robust estimate of the average or mean impact of a treatment. However it cannot comprehensively answer whether, for example, micro-credit immensely benefits 5% and does not affect the other 95%? Or does it benefit 50% and greatly and unethically harm the other 50% of the treatment group?
The J-PAL results show that on average, in villages located in Morocco, India, Ethiopia and Mexico, only one household in five wanted microloans. This proved to be substantially lower than the projected estimates by the credit-lending institutions themselves. This finding deals a blow to the widely publicised assertion by microcredit advocates that if only all poor people had access to credit, it would transform the lives completely. In reality, most won’t avail of it. The results did find that some of the borrowers used the money to grow their own small businesses. However, this did not translate to significantly higher profits in any of the studies, compared to control groups. Furthermore, none of the trials found a significant difference between treatment groups and control groups with regard to average household income or education levels. In Bosnia and Herzegovina, microloans might have encouraged families to remove children from school to work in expanding businesses. Households taking credit in these treatment communities were actually 9% less likely to send their children to school.
The one positive difference, which was echoed in previous studies on credit mentioned above, finds that loans give borrowers decreased labour volatility and consumption volatility. The studies found that while people made the same money as before, they actually earned the money from different activities that they themselves chose. Therefore increasing their level of self-reliance, autonomy and flexibility (i.e. they could choose to work on something that suited their life better, such as set up a home business rather than work in a factory far from their home to earn an income) . These results reflect the previous findings by both Khandker and Morduch using Bangladeshi data. These results could be more assertive in claiming that microcredit offers greater security from external shocks during day to day living and offers greater freedom and self-reliance.
Randomised control trials focus on measuring quantifiable results – such as change in annual income and business profitability. At present, they have not investigated the more intangible results from microcredit in the developing world. One such widely advertised positive effect of microcredit is its way of empowering women in developing countries. Pitt, Khandker and Cartwright (2006) find that microcredit (through direct mechanisms and positive externality effects) positively affecting female’s level autonomy in economic decisions, promoting self-sufficiency, promoting family planning, and increasing their access to resources outside the household.
There have been contrasting findings with regards to this claim. Nilakantan, Sinha and Datta, (2013) investigated the impact of microcredit on India women’s level of empowerment and found that greater access to microcredit (operationalised as the duration of treatment in the study design) did not indicate greater empowerment effects on the economic decision-making. Such findings indicate that the dynamics between men and women within the household along traditional gender lines need to be fully addressed before microcredit can be fully effective for empowering women. Nilakantan et al. (2013) employed a pipeline research design, using prospective members who had signed up but yet to receive loans as controls for veteran members. As mentioned above about “difference-in-difference” approaches, flaws were criticised in this design. Randomised evaluations will need to be carefully planned to attempt answering these questions with more empirical precision. Nilkantan and colleagues acknowledge that the pipeline approach is less than optimal, but is suitable in situations where it is “not feasible to interview sufficient non-members in order to form a large enough comparison group, and when credit lending has already been going on for some time” (2013: 31). Both of these were happening in the Indian sample during the study, thus precluding randomised evaluations. With this design, ever present selectivity biases mean authors cannot discount the possibility that women who are already more empowered are more likely to access micro-credit in the first place, compared to women who lack this level of freedom.
In conclusion, despite decades of high hopes for microcredit as a developmental panacea to lift millions of people out of poverty – particularly women – the empirics thus far cannot offer resounding support for these claims. Based on Bangladeshi data, Pitt and Khandker (1998), as well as Khandker and Saman (2013) constitute the most widely disseminated research papers which provide substantial support for microcredit, in the form of significantly increased income. However, the prevalence of biases in these analyses requires wholly different research approaches. Using randomised control trials, researchers frequently fail to find statistically significant impacts of credit on the transformative variables that would lead to widespread poverty-reduction: namely increased business profits, increased household income and increased consumption levels. Furthermore, while female empowerment constitutes a cornerstone of the microcredit package, the research is yet to find evidence of empowerment “success stories”.
While empirical records do not support the view that microcredit is cure-all solution, the findings highlight how microcredit can enable poor people’s to make their own choices in how they work – whether through wage employment or self-employment – and more flexibility in how they can budget and spend their household income. While current RCTs have failed to find transformative effects of microcredit in helping people escape from poverty, research can still be carried out to find ways to improve microcredit in helping reduce consumption volatility, labour supply volatility and make credit more appropriate for heterogeneous client bases. Endeavours such as reducing borrower-lender information asymmetries and transaction costs (maybe through less frequent repayment schedules which require less administrative work) would be fruitful avenues for improving the impact of credit institutions and increase the time available to make investments before the first installment is due).
Too often microcredit is synonymous with microfinance; but the data shows that credit alone will not transform the lives of the poor. Micro-lenders can expand their services beyond microcredit with greater focus on borrowers’ needs. For example the Bangladeshi Rural Advancement Committee found that over 30% of late loan repayments were due to health shocks or after the death in a borrower’s household. The creation of micro-insurance for the poor could go a long way to reduce volatility, uncertainty and financial problems (such as resorting to risky activities or loan sharks) that poor people face.
Defenders of the microcredit enterprise who are skeptical of randomised control trials argue that no-one should anticipate that giving loans will act as a magic bullet to end poverty in a single generation. Future long-term or longitudinal studies with randomised evaluations appear to be the most robust and objective tool at the disposal of researchers to answer these questions and find out if micro-credit can alleviate poverty in the long-run or if they can conclude once and for all that micro-credit is not the panacea that was so widely touted since the 1970s.
Angelucci, M., Karlan, D., & Zinman, J. (2013). Win some lose some? Evidence from a randomized micro-credit program placement experiment by Compartamos Banco (No. w19119). National Bureau of Economic Research.
Attanasio, O., Augsburg, B., De Haas, R., Fitzsimons, E., & Harmgart, H. (2011). Group lending or individual lending? Evidence from a randomised field experiment in Mongolia.
Augsburg, B., De Haas, R., Harmgart, H., & Meghir, C. (2012). The Impacts of Micro-credit: Evidence from Bosnia and Herzegovina (No. w18538). National Bureau of Economic Research
Banerjee, A. V., Banerji, R., Duflo, E., Glennerster, R., & Khemani, S. (2010). Pitfalls of Participatory Programs: Evidence from a randomized evaluation in education in India. American Economic Journal: Economic Policy, 1-30.
Banerjee, A., Cole, S., Duflo, E., & Linden, L. (2005). Remedying education: Evidence from two randomized experiments in India (No. w11904). National Bureau of Economic Research.
Blattman, C. (2008). Impact Evaluation 2.0. Presentation to the Department for International Development (DFID), London.
Chemin, M. (2008). The benefits and costs of microfinance: evidence from Bangladesh. The journal of development studies, 44(4), 463-484.
Crépon, B., Devoto, F., Duflo, E., & Parienté, W. (2011). Impact of micro-credit in rural areas of Morocco: Evidence from a Randomized Evaluation. MIT Working Paper.
Deaton, A. S. (2009). Instruments of development: Randomization in the tropics, and the search for the elusive keys to economic development (No. w14690). National Bureau of Economic Research.
Edgecombe, E., & Garber, C. (1998). Evaluación de impacto dirigido por practicantes: un ensayo en Honduras, Practitioner-led impact assessment: a test in Honduras. Assessing the Impact of Micro-enterprise Services (AIMS). AIMS.
Goldberg, N. (2005). Measuring the impact of microfinance: taking stock of what we know. Grameen Foundation USA publication series.
Hahn, J., Todd, P., & Van der Klaauw, W. (2001). Identification and estimation of treatment effects with a regression‐discontinuity design. Econometrica,69(1), 201-209.
Karlan, D. S. (2001). Microfinance impact assessments: the perils of using new members as a control group. Journal of Microfinance/ESR Review, 3(2), 75-85.
Khandker, S. R. (2005). Microfinance and poverty: Evidence using panel data from Bangladesh. The World Bank Economic Review, 19(2), 263-286.
Khandker, S. R., & Samad, H. (2013). Microfinance Growth and Poverty Reduction in Bangladesh: What Does the Longitudinal Data Say? (No. 16).
MkNelly, B., & Lippold, K. (1998). Practitioner-led impact assessment: A test in Mali. Washington, DC: AIMS.
Morduch, J. (1998). Does microfinance really help the poor?: New evidence from flagship programs in Bangladesh. Research Program in Development Studies, Woodrow School of Public and International Affairs.
Nilakantan, R. Datta, S. C., Sinha, & Datta, S. K. (2013). The Impact of Microfinance on Women Empowerment: Evidence from Eastern India. International Journal of Development and Conflict, 3, 27-40.
Pitt, M. M., & Khandker, S. R. (1998). The impact of group-based credit programs on poor households in Bangladesh: Does the gender of participants matter?. Journal of political economy, 106(5), 958-996.
Pitt, M. M., Khandker, S. R., & Cartwright, J. (2006). Empowering women with micro finance: Evidence from Bangladesh. Economic Development and Cultural Change, 54(4), 791-831.
Roodman, D., & Morduch, J. (2009). The impact of micro-credit on the poor in Bangladesh: Revisiting the evidence. Center for Global Development Working Paper, (174).
Tarozzi, A., Desai, J., & Johnson, K. (2013). On the impact of micro-credit: Evidence from a randomized intervention in rural Ethiopia.
 According to Rahman and Yusuf, “about a fifth of that economy was destroyed during the Liberation War of 1971, and severe dislocations caused at that time left Bangladesh on a slower economic growth trajectory for the following two decades. Then the economy accelerated sharply from 1990 with high GDP growth rate [and] Higher investment-to-GDP ratio means more capital in the economy” (p. 6) (http://www.hks.harvard.edu/fs/drodrik/Growth%20diagnostics%20papers/Economic%20growth%20in%20Bangladesh%20-%20experience%20and%20policy%20priorities.pdf