Despite decades of impact evaluations and evidence generation, the financial inclusion sector still cannot say with confidence whether a given financial service will meaningfully help clients, have little or no impact, or cause harm. Addressing this knowledge gap has been the central aim of CGAP's Financial Inclusion 2.0 initiative.
Nowhere is this variability of impact more visible than in microcredit. CGAP's Impact Pathfinder (an evidence platform designed to shed light on financial inclusion’s impact on development outcomes by synthesizing more than 400 credit-focused studies) and the recent focus note built on it both show how wide that range is. Results run from transformative to detrimental depending on who borrows, how the credit is structured, and the context it is delivered in.
However, impact evaluations that primarily examine averages – as much of the early randomized controlled trial (RCT) literature on microcredit did – do not always capture the full story. Sub-group analysis within individual studies and more recent meta-analyses, such as Rachael Meager’s Bayesian hierarchical analyses, show that those averages can hide important variation. For instance, Meager’s finding that stronger gains solidified among borrowers who already had business experience.
However, individual studies are not always able to explore all the angles that explain the variability of impact given sample size limitations. This leaves a need for more nuanced insights that stakeholders, such as financial service providers and funders, require to make well-informed decisions. A key question that CGAP sought to explore was whether existing data could be used in new ways to find answers to the lingering question of what drives this impact variability —i.e., the heterogeneous impact question.
The Credit's Heterogeneous Impact Pathways Study (CHIPS) data set, developed by CGAP together with Innovations for Poverty Action (IPA) and the Global Poverty Research Lab (GPRL) at Northwestern University, picks up on this challenge. While CGAP’s Impact Pathfinder and focus note draw on published findings from a wealth of studies, CHIPS takes the next step by harmonizing the underlying microdata from 29 RCTs – covering more than 115,000 individuals across 18 countries – into a single data set. This gives researchers sufficient statistical power to examine how factors like gender, livelihood, household composition, and rural versus urban location shape the impact of microcredit.
The data set is a publicly available resource meant for any interested researcher. On its own, it does not generate new findings; rather, it opens new opportunities for researchers to ask and answer more targeted questions about microcredit impact heterogeneity than any single evaluation could support.
What the development process did reveal, however, was three important implications that can inform the direction of a future research agenda on when, how, and for whom microcredit works to ensure positive customer outcomes.
1. Harmonizing existing data can advance our understanding of why impacts differ
CHIPS demonstrates that data harmonization efforts across RCTs can generate new insights into how microcredit affects different customers under different circumstances.
CHIPS offers sufficient observations to support robust analysis on the impact of microcredit on specific customer outcomes: income, value of assets, and employment status. It also captures specific contextual factors that may drive how microcredit impacts these outcomes, including business ownership, whether a business operates in agriculture, household size, gender, and rural or urban residence.
This makes it possible to explore questions such as: How does microcredit affect the incomes of rural women engaged in an agribusiness relative to their male counterparts? Does household size influence the returns to credit? Do impacts differ between agricultural and non-agricultural businesses?
Researchers and funders can use CHIPS to investigate these and many other questions about who benefits most from access to credit.
2. Harmonization efforts reveal important financial inclusion questions that remain understudied
The harmonization process also revealed significant gaps in the existing research landscape.
Many outcomes that are highly relevant to policymakers and practitioners are assessed too infrequently to support meaningful cross-study analysis. These include women's economic empowerment, food security, psychological well-being, and savings accumulation. Furthermore, none of the eligible studies for CHIPS collected data on outcomes related to climate adaptation and resilience.
Similarly, important contextual variables are often missing, including education levels, participation in social protection programs, and proximity to markets or infrastructure. Information on credit product features—such as interest rates, loan terms, and repayment structures—is also rarely available, constraining analysis of how product design affects customer outcomes.
These gaps point to clear priorities for future research. By collecting a broader set of outcomes and contextual variables in their impact evaluations, researchers can generate evidence that can be leveraged beyond the context of an individual study and be more useful for policy and product design.
3. New analytical methods can unlock more value from existing data
Recent advances in impact evaluation and data science create new opportunities to leverage harmonized data sets such as CHIPS to further explore the variable impact question – opening the potential for new insights on the impact of microcredit.
One approach is the use of Bayesian Hierarchical Models (BHMs), which help estimate average treatment effects while assessing how results vary across study settings. For example, researchers can examine the average effect of microcredit on income or consumption and quantify how those effects differ across countries and populations. Data sets like CHIPS offer the opportunity to scale up studies like the previously mentioned Meager studies.
More recent machine learning approaches provide an additional avenue for exploration. Generic Machine Learning (GML) methods can identify patterns of heterogeneous impact without requiring researchers to specify in advance which variables are most important. Instead, the data itself can help reveal where impact occurs and which factors are most predictive of positive or negative outcomes.
These methods offer a powerful complement to traditional approaches when analyzing harmonized data sets and can help uncover insights that may otherwise remain hidden.
Turning evidence into better financial inclusion outcomes
CHIPS represents an important step toward understanding not only whether microcredit works, but for whom and under what conditions. The data set provides a valuable foundation for the next generation of rigorous research on financial inclusion, while also highlighting critical evidence gaps that deserve greater attention. In doing so, it advances the core aim of CGAP's Financial Inclusion 2.0 initiative: shifting the sector's focus from measuring access to understanding outcomes.
The opportunity now lies with the broader research, donor, and practitioner community. By using CHIPS, expanding data harmonization efforts, collecting richer contextual and outcome data, and applying innovative analytical methods, researchers can explore questions of impact heterogeneity, which have remained unanswered for too long. Answers obtained promise to be relevant for policymakers and financial service providers to design better program interventions and financial products. Together, we can move beyond measuring average impacts and toward designing financial services and policies that consistently improve the well-being of the people they are intended to serve.
Explore the CHIPS dataset, contribute new evidence, and help advance a more nuanced understanding of what makes microcredit work.
Add new comment