Why Average Effects Can Hide Different Borrower Outcomes

TLDR: The central lesson from heterogeneous effects microfinance research is that credit does not produce one uniform result. Modest average effects can coexist with gains for some borrowers, little change for others and harmful repayment pressure for another group. The useful question is not simply whether microfinance works, but for whom, for what purpose, under which terms, in what market and over what period.

That distinction matters for lenders, funders, regulators and borrowers. A working-capital loan offered to an established market trader before a predictable sales period is economically different from the same-sized loan offered to a first-time entrepreneur facing uncertain demand. Averaging their outcomes may be necessary for estimating a program’s overall effect, but it can obscure the mechanisms that explain why their experiences diverge.

What heterogeneous effects mean in microfinance

Microcredit refers specifically to small loans. Microfinance is broader and can include savings, insurance, payments and other financial services. Financial inclusion is broader still: it concerns whether people can access and use appropriate, affordable financial services that meet their needs. Evidence about expanded access to microcredit should not automatically be generalized to every financial product.

An average treatment effect compares the average outcome for a group offered or receiving an intervention with the average outcome for a suitable comparison group. Depending on the study, the outcome might be business profit, revenue, investment, household consumption or another measure. It is a population-level estimate, not a prediction that every borrower will experience that result.

A heterogeneous treatment effect means the estimated effect differs across people, businesses, locations, products or circumstances. For example, borrowers who already operated businesses before gaining access to credit might respond differently from people attempting to start businesses. Differences may also arise by sector, income, local competition, repayment schedule or exposure to shocks.

Three stages should remain separate. Access means a product is available. Take-up means an eligible person chooses to use it. Realized benefit or harm concerns what happens after that decision. A program can expand access substantially while attracting relatively few borrowers, and borrowers who participate can still experience sharply different outcomes.

Why a modest average can conceal important outcomes

Suppose ten enterprises receive comparable loans. Two use the funds to buy fast-selling inventory and increase profits. Five make no durable gain after financing costs. Two divert funds to urgent household expenses and struggle with repayments. One closes after an unrelated shock. The average effect may be close to zero even though the loan mattered greatly—positively or negatively—to several households.

That does not make the average meaningless. Program-level decisions still require an estimate of what happened across the eligible population. But the average cannot, by itself, explain the distribution of outcomes, identify the mechanism or establish that a product is suitable for a particular applicant.

The influential 2015 overview of six randomized evaluations of microcredit expansions found a broadly similar pattern of modest rather than transformative average effects. Readers can review the six-study overview published by the American Economic Association. This challenged universal claims about rapid poverty reduction without establishing that credit never helps.

A 2019 Bayesian hierarchical analysis subsequently combined seven randomized microcredit-expansion experiments to examine variation across studies. It reported larger, but more variable, estimated effects among households with prior business experience than among those without it. The analysis also indicated that economic features of the interventions accounted for more cross-study variation than differences in study protocols. The seven-experiment analysis is especially useful because it treats variation as a substantive question rather than statistical noise.

Prior business ownership is not a universal targeting rule. Existing firms vary in profitability, demand, management, household claims on revenue and vulnerability to shocks. The evidence instead points toward a mechanism worth investigating: an operating enterprise may have suppliers, customers, skills and a known use for additional capital that a first-time entrepreneur has not yet developed.

A five-part framework for interpreting different outcomes

A practical way to open the credit “black box” is to examine who borrows, how the product works, what it finances, where it is used and when outcomes are measured. CGAP presents these dimensions as central to understanding inclusive-credit impacts.

These dimensions interact. A seasonal farm investment may need months before generating revenue; weekly installments can force the borrower to repay from another income source or sell output too early. An urban retailer turning inventory every few days may find frequent installments easier to match with cash flow. Product fit therefore depends on the timing and reliability of income, not only the borrower’s stated occupation.

How borrower situations can produce different pathways

Working capital for an operating business

An established shop may know which goods sell, how quickly inventory turns and what margin remains after transport and spoilage. Credit can relax a real constraint if additional stock produces enough cash before installments fall due. But borrowing still fails to help when demand is saturated, the margin is too thin or household withdrawals consume the restocking money. Researchers evaluating this pathway need measures beyond loan receipt, including inventory, margins, cash flow and business continuity. Our guide to measuring microfinance and business growth explains why revenue alone is not enough.

Starting a new enterprise

A first-time entrepreneur faces both a capital constraint and an information problem. The borrower may not yet know the level of demand, operating cost, suitable price or time required to break even. A loan supplies funds but does not automatically supply customers, technical skills or a viable business model. Training or market access may sometimes matter more than additional debt, although their usefulness also varies.

Seasonal livelihoods and farming

Farm and seasonal businesses often spend well before they earn. Loan maturity, grace periods and installment timing therefore become central design features. Weather, crop disease and price changes can affect many borrowers simultaneously, making individual effort an incomplete explanation for repayment outcomes.

Emergencies and consumption smoothing

A borrower may describe a loan as “for business” while household funds remain fungible. If medical costs, school fees or food needs are urgent, money intended for inventory may cover those expenses. Credit can help a household manage timing, but it does not create the future income needed for affordable repayment. Savings, insurance, emergency support or flexible payments may fit some risks better. World Bank and CGAP materials likewise place credit alongside savings, insurance and payments rather than treating it as a standalone solution to poverty or vulnerability.

Refinancing and accumulated obligations

A new loan used mainly to repay an old one can temporarily relieve pressure without improving the household’s underlying cash position. Repeat borrowing is not inherently harmful—healthy businesses may borrow repeatedly—but refinancing cycles warrant examination of total debt, payment-to-income burden and whether principal is actually declining. For a broader discussion, see the site’s overview of microfinance pitfalls and borrower precautions.

Why subgroup findings require restraint

A reported difference between subgroups is not automatically a reliable targeting rule. Small subgroups produce imprecise estimates. Testing many categories increases the chance of finding an apparently unusual result by coincidence. A credible interpretation should ask whether the subgroup was specified in advance, whether the difference is statistically distinguishable, whether there is a plausible mechanism and whether it appears in other settings.

Researchers should also distinguish variation across studies from variation among borrowers within one study. A result from one city, lender or product may reflect local demand, eligibility rules or loan terms. It may not transfer to another market simply because the borrowers share a demographic label.

Gender illustrates the problem. A loan issued to a woman does not by itself prove that she controls its use, keeps the resulting income or gains household decision-making power. Those are separate outcomes requiring direct measurement. At the same time, it would be equally careless to assume women-owned enterprises are uniformly constrained or unprofitable. Household bargaining, sector concentration, care obligations, asset ownership, mobility and local norms can all shape outcomes.

Rural-versus-urban comparisons also need context. Rural borrowers may face seasonal income and long distances to markets, while urban borrowers may face intense competition and high operating costs. Geography is often a label for several mechanisms, not the mechanism itself.

A practical decision test for lenders and funders

Before targeting credit at a subgroup or scaling a product, decision-makers should test the proposed pathway rather than relying on a positive average or an appealing borrower profile.

1. Identify the viable use. What specific activity will generate or protect enough value to justify borrowing?
2. Map the cash flow. When will money leave the household, when can revenue reasonably arrive and how variable is it?
3. Calculate affordability under stress. Include the total financing cost and model weaker sales, delayed harvests or emergency spending—not only the expected case.
4. Check existing obligations. Review formal and informal debts, guarantees and refinancing rather than assessing one loan in isolation.
5. Compare alternatives. Ask whether savings, insurance, payments, grants, emergency assistance or nonfinancial business support addresses the underlying problem more safely.
6. Validate the subgroup claim. Require a plausible mechanism, adequate sample, uncertainty estimates and evidence from relevant products and markets.
7. Monitor adverse outcomes. Track arrears, asset sales, skipped essentials, repeat refinancing, business closure and distress—not only repayment rates.
8. Measure at useful intervals. Observe investment, repayment-period pressure and post-loan outcomes so that short-term strain is not hidden by a single endpoint.

Borrowers can apply a simpler version: identify the source and timing of repayment, calculate what remains after household essentials, and ask what happens if expected revenue arrives late. If repayment depends on taking another loan, selling an essential asset or assuming unusually strong sales, the product may not fit the underlying need.

The takeaway: replace a universal verdict with a conditional one

Evidence on microcredit expansions does not support a simple story of universal transformation, but neither does a modest average prove that every borrower receives no value. Heterogeneous outcomes are the expected result when borrowers, enterprises, products, markets and shocks differ.

The responsible next step is to make the conditions explicit. Ask who has a credible use for capital, whether the terms fit the resulting cash flow, what risks could interrupt repayment and which alternative service might work better. Then measure both gains and distress over an appropriate period. That approach turns heterogeneous effects from a statistical complication into a practical guide for designing safer, more useful financial inclusion.

References

1. Six Randomized Evaluations of Microcredit: Introduction and Further Steps – American Economic Association
2. Understanding the Average Impact of Microcredit Expansions: A Bayesian Hierarchical Analysis of Seven Randomized Experiments – American Economic Association
3. Opening the Black Box on the Impact of Inclusive Credit | CGAP Research & Publications
4. Opening the black box: What we now know about credit and poverty

What Does the Evidence Say About Microfinance and Women’s Economic Agency?

TLDR

The microfinance women empowerment evidence supports a careful conclusion: traditional microcredit can expand access to finance and may help some women and households, but average results are generally modest rather than transformative. A loan issued in a woman’s name does not establish that she chose to borrow, controls the money, manages the enterprise, keeps the profits, or gains influence over household spending. Those are separate outcomes and must be measured separately.

High-quality evidence does not justify the simple claim that microcredit either empowers all women or does nothing. Results differ across borrowers and settings. Existing business experience may matter, as may local markets and household circumstances. The practical lesson for lenders, funders, and researchers is to measure control, benefits, and risks alongside outreach and repayment.

Access to finance is not the same as economic agency

Microcredit, microfinance, financial inclusion, and women’s economic empowerment are related concepts, but they are not interchangeable. Microcredit means small loans, usually offered to people underserved by conventional banks. Microfinance is broader and can include savings, insurance, payments, and other services. Financial inclusion concerns whether people can access and use suitable formal financial services.

Women’s economic empowerment goes further. It concerns a woman’s practical ability to make and act on economic decisions, control resources and income, manage risk, and influence decisions affecting her livelihood. A financial product can contribute to that ability, but product access is an input rather than proof of the final result.

This distinction matters because programs often report readily available administrative indicators: the number of women borrowers, loan amounts, repayment rates, account openings, or transaction counts. Those figures can show reach and use. They usually cannot reveal who decided how a loan would be spent, who performs the work, who controls revenue, or who absorbs a loss.

What randomized and pooled studies find

Randomized evaluations are useful because they can estimate the average causal effects of expanding access under the conditions studied. They still do not answer every question. Results may conceal large differences among participants, and a study designed around business investment or consumption may not measure control over money well.

A 2015 overview brought together six randomized evaluations of microcredit conducted in six countries across four continents. Its overall reading was that results were modestly positive, not transformative. That does not mean every outcome was zero or every borrower had the same experience. It means the experiments did not support a broad expectation of dramatic average changes in household welfare or business outcomes from standard microcredit expansion alone. The six-country randomized-evaluation overview provides the study-level context.

A 2019 Bayesian hierarchical analysis synthesized seven randomized microcredit experiments. It concluded that average effects on household business and consumption outcomes were unlikely to be transformative, while finding substantial heterogeneity. Prior business experience emerged as a plausible source of variation: households with previous business experience showed larger but also more variable effects.

That finding is important for program design, but it should not be converted into a universal rule that experienced entrepreneurs will always benefit. It suggests that the same loan can interact differently with an operating business, a new venture, household consumption needs, and local demand. Credit relaxes one constraint—liquidity—but does not automatically create customers, profitability, skills, mobility, bargaining power, or protection from shocks.

Readers interested in the wider poverty debate can place these findings alongside the broader evidence on microfinance and poverty. The central issue here is narrower: even a positive business or consumption result does not necessarily show that the woman named as borrower controlled the decision or retained the benefit.

Does microcredit increase household bargaining power?

The strongest supplied evidence directly addressing household control comes from a 2014 Campbell systematic review and meta-analysis. The review examined microcredit’s effects on women’s control over household spending in developing countries. Its higher-quality evidence did not show a consistent causal effect. The Campbell systematic review on women’s control over household spending also noted that many studies failed to measure control over loan use empirically.

For randomized studies in India, Morocco, and South Africa, the pooled standardized mean difference was -0.007, with a 95% confidence interval from -0.041 to 0.027. In practical terms, the average estimated effect was effectively zero, and the interval was narrowly distributed around zero.

This is not proof that no woman in those countries benefited. An average can combine positive, negative, and negligible individual experiences. Nor does it establish that every microfinance model has the same effect. It does show why a general statement such as “lending to women increases their household power” goes beyond what this pooled causal evidence supports.

Control over household spending is also only one dimension of agency. A woman might gain authority over business purchases but not household expenditures, or she might control daily sales while another household member decides how profits are used. Conversely, she might gain greater household influence without running a business. Studies and program reports should identify which dimension they measured instead of placing all of them under a single empowerment label.

Why repayment and women borrower counts are incomplete indicators

A strong repayment rate answers an operational question: were scheduled obligations paid? It does not identify who generated the repayment money, whether the investment was profitable, or what sacrifices a household made to avoid default. Repayment can come from business revenue, wages, savings, another loan, asset sales, or transfers from relatives. Without further measurement, the lender cannot infer agency from repayment alone.

Women borrower counts have a similar limitation. They are relevant measures of outreach, particularly where women face formal barriers to finance. But the borrower recorded in a management system may not be the person who selects the investment, controls the financed asset, or decides how earnings are spent. A program can therefore reach many women while producing uncertain changes in enterprise control or bargaining power.

This distinction does not make outreach or repayment unimportant. It makes them insufficient as impact measures. Responsible evaluation pairs portfolio indicators with questions about product choice, control, pressure, financial stress, benefits, and unintended harm. The same principle applies when considering microfinance risks and borrower precautions.

Why outcomes vary among women and places

The evidence gives good reason to expect heterogeneity, but less reason to declare one universal mechanism. Prior business experience is one plausible factor identified in the seven-experiment synthesis. Other features worth investigating include the purpose of borrowing, demand for the product being sold, access to assets, freedom of movement, household norms, private access to a phone or account, and responsibility for repayment.

These are decision factors, not proven explanations for every observed result. Their relevance must be tested in the specific program and geography. For example, private digital access could make it easier for one client to control transactions, while exposing another to monitoring or pressure. A profitable market opportunity could turn credit into useful working capital, while a saturated market could leave a borrower with debt and unsold stock.

Program design also changes what is being evaluated. A loan-only product is not equivalent to a savings group, insurance product, mobile-money account, digital loan, or package combining finance with training, grants, mentoring, childcare, or market access. Evidence about conventional microcredit should not automatically be presented as evidence for or against all of microfinance.

Where broader financial inclusion fits

Financial inclusion data help describe the infrastructure within which agency may develop. The World Bank’s Global Findex 2025 covers accounts, payments, savings, borrowing, and digital connectivity across 141 economies, drawing on data from 145,000 adults. It is valuable for understanding who has access to and uses financial services. The World Bank’s Global Findex database provides that descriptive context.

Findex indicators are not causal evidence that account ownership or digital connectivity changes household bargaining power. An account can create an opportunity for greater privacy or control, but researchers still need to observe who accesses it, who authorizes transactions, and whether its use changes decisions or economic outcomes.

The same caution applies to digital credit. Speed and convenience may reduce transaction costs, but they do not resolve questions about suitability, control, price, repayment pressure, or enterprise returns. Technology changes the delivery channel; it does not make agency automatic.

A practical test for women’s empowerment claims

Before labeling a financial program empowering, MFIs, funders, researchers, and readers should ask questions at each stage of the financial relationship:

• Choice: Did the woman decide to apply, understand the terms, and have a realistic option to refuse?
• Use: Who decided how the funds would be used, and did actual use match the stated loan purpose?
• Enterprise control: Who chooses suppliers, prices, inventory, working hours, and major business investments?
• Revenue: Who receives sales income, and who controls withdrawals, profits, and reinvestment?
• Household influence: Did authority over spending, saving, assets, or major purchases change?
• Risk: Who is responsible for installments, and what happens when revenue falls short?
• Economic results: Did profits, income, assets, resilience, or business survival improve after accounting for costs and debt?
• Durability: Do any changes persist beyond the loan cycle, and are they visible over an appropriate follow-up period?
• Distribution: Are average results hiding meaningful differences by prior business experience, income, location, household structure, or product design?

No single survey question will capture all of economic agency. Programs should combine administrative data with confidential client interviews and outcome measures suited to their theory of change. If the goal is enterprise control, measure enterprise decisions. If the goal is household bargaining, ask about specific decisions rather than using borrower status as a proxy. If the goal is durable income growth, follow income, costs, assets, and business continuity long enough to test it.

For educators, this layered framework can also improve discussions about impact evidence. A session built around organizing a microfinance awareness event on a college campus can ask participants to separate access, use, control, and economic gains rather than debating whether microfinance simply “works.”

The responsible conclusion

Traditional microcredit can widen access to capital, and some women—particularly in favorable business circumstances—may use it productively. But the evidence reviewed here does not support treating a woman’s borrower status, account ownership, or repayment record as proof of greater household power, enterprise control, or long-term income transformation.

The next step is to evaluate each program against the outcome it is actually expected to change. Track access, but also determine who controls the money. Track repayment, but also identify who bears the risk. Measure business activity, but also measure profits, decision-making authority, and durability. That approach replaces a broad empowerment promise with a more useful question: under what conditions does finance expand a woman’s real choices, control, and economic security?

References

1. pubs.aeaweb.org/doi/pdfplus/10.1257/app.20140287
2. Understanding the Average Impact of Microcredit Expansions: A Bayesian Hierarchical Analysis of Seven Randomized Experiments – American Economic Association
3. The Effects of Microcredit on Women's Control over Household Spending in Developing Countries: A Systematic Review and Meta‐analysis – Vaessen – 2014 – Campbell Systematic Reviews – Wiley Online Library
4. The Global Findex Database 2025