Note: This page summarizes the rationale behind a GiveWell grant to the University of Oxford. University of Oxford staff reviewed this page prior to publication.
Summary
In December 2024, GiveWell recommended a $491,700 grant to the University of Oxford to fund data collection for a large-scale randomized controlled trial (RCT) of GiveDirectly’s cash transfer program in rural Malawi.
This study will examine how the provision of $550 transfers (125% of the average annual household income in the study area) to 110,000 adults affects both recipients and the broader local economy, including potential spillover effects on market prices, business activity, and non-recipient households.
We are recommending this grant because:
- We think it is right to contribute funding to research that is directly relevant to our interests, and which could have broader benefits for the development community;
- By contributing funding, we can be an engaged research partner, with opportunities to shape the study design and to access findings early;
- We may have the opportunity to fund additions to the study that would be informative for us, such as measuring child mortality, or testing seasonal targeting or focusing on pregnant women;
- This grant addresses key uncertainties from our recent update to our cost-effectiveness estimate of GiveDirectly’s Cash for Poverty Relief program (more below), aligning with our interest in identifying cash transfer grantmaking opportunities that exceed our cost-effectiveness bar.
Our main reservations are that:
- The research team may not be wholly impartial towards GiveDirectly, although we believe we have mitigated this risk by setting conditions for our funding, such as our approval of the study’s pre-analysis plan;
- We think this study would likely have received this additional funding without us, since the research team has already secured most of its budget from another source.
Published: March 2025
Background
GiveDirectly’s Cash for Poverty Relief program involves sending one-off unconditional cash transfers of ~$1,000 to households living in poor regions of low-income countries in sub-Saharan Africa.1
In 2022, Dennis Egger, Johannes Haushofer, Edward Miguel, Paul Niehaus and Michael Walker published a study looking at the general equilibrium effects of GiveDirectly’s program in Kenya. Egger et al. found large and positive consumption “spillovers” (i.e., increases in consumption among non-recipients), which at face value would imply a multiplier effect of 2.5 to the program’s consumption benefits.2 While we adjust this multiplier downwards (for reasons we discuss here), this study formed a key input into our November 2024 update to our GiveDirectly cost-effectiveness estimate, in which we found GiveDirectly’s program to be around three to four times as cost-effective as we had previously estimated.3
We have a number of uncertainties about how best to translate the Egger et al. results into our cost-effectiveness estimate, including how we should expect the spillover effects to generalize in different program contexts, such as poorer and less economically integrated areas, and more saturated program designs (i.e., in which a larger proportion of the population receives the intervention).
At the time we updated our cost-effectiveness estimate, we were aware that a separate study team, led by Dennis Egger, had begun research in Malawi that would evaluate the inflationary effects of a universal GiveDirectly program, which could address the aforementioned uncertainty. This study team conducted a pilot evaluation to inform a large-scale randomized controlled trial (helping the researchers to understand what kind of demand shocks and inflationary effects to expect).4
The grant
This grant will provide partial funding for a large-scale RCT of GiveDirectly’s Cash for Poverty Relief program in Malawi. This study is the largest RCT implemented by GiveDirectly to date: it will provide transfers of $550 (125% of the average annual household income in the study area) to 110,000 adults in one district, Chiradzulu (i.e., to around half of the district’s total population).5 The 2022 Kenya study mentioned above, by contrast, provided transfers worth 75% of annual household income, and included limitations on who was eligible to receive the program.6 Combined, these differences mean that the Malawi program is approximately an order of magnitude larger in overall transfer amount relative to the size of the study area economy.
This study would test whether or not the multiplier found in Egger et al. 2022 can be replicated in a different program context, where people are typically poorer, and communities less integrated in trade. As mentioned above, it will also focus on a universal program with significantly larger transfers than the 2022 Kenya study, where only a subset of households were eligible.7
The study will look at the effects of the program on the local economy as a whole, including supply-side effects (whether demand shock causes disruptions or inflation, and various effects on business activity) and demand-side effects (whether customers reallocate their spending to markets with lower inflation). To measure this, the study design includes high-frequency market surveys to track expenditure, inflation, and market activity, and monthly surveys to monitor recipient spending patterns.8
This study will randomize cash transfers at different geographic levels, in order to generate variation in cash flows into markets, as well as randomizing whether sellers receive advance knowledge of the cash transfer, giving them time to prepare for an increase in demand for their products (which could reduce inflationary pressure).9
Budget for grant activities
GiveWell’s grant will provide funding for $491,700 of the total cost of the evaluation (i.e., excluding the cash transfers themselves) of $1,847,200.10 Our funding breaks down as follows:11
- $312,000 (63% of the grant): monthly surveys (including monthly household surveys to monitor recipient spending patterns, price surveys, and a market census and survey)12
- $135,000 (27%): baseline and endline enterprise surveys
- $44,700 (10%): University of Oxford indirect costs
Timeline
The study timeline has not yet been fixed, due to Malawi’s 2025 general election, which will require a break in the disbursement of cash transfers. Broadly, however, we expect the study to proceed in two phases, as follows:13
- Research pilot (February–September 2025): From February until June, GiveDirectly will provide transfers to 34,000 people, covering a third of Chiradzulu district. In this phase, the researchers will run a relatively large-scale research pilot, which will provide early indicative results on consumption, inflation, and potentially political outcomes and may inform subsequent study scale-up. However, this phase will likely not have sufficient sample size to provide results on spillovers or economic multipliers.
- Scale-up (October 2025 onwards): After Malawi’s general election, GiveDirectly will scale up to the remainder of Chiradzulu district, and potentially beyond, if they receive additional funding. In this stage, the researchers will roll out the full study design, which will be powered to detect spillover and multiplier effects.
The study has been designed so that data from both the pilot and scale-up regions can be combined into one coherent study and paper.14
The case for the grant
We are recommending this grant because:
- Most importantly, funding this grant allows us to contribute to the research around unconditional cash transfers. GiveDirectly is an important organization for GiveWell (since we have historically used our estimate of the cost-effectiveness of its program as our “bar” for comparing other programs)15 and for the broader international development ecosystem (since it has sparked a conversation around development effectiveness through the use of cash as a “benchmark”, as well as about beneficiary preferences). As we mention below, we think it is likely that this study would have received this remaining funding without us; however, we think it is right for us to contribute funding for evidence generation that is directly relevant to our work. Also, given the broader importance of GiveDirectly, we think that this funding constitutes a public good, signalling our interest in advancing collective knowledge on unconditional cash transfers.
- By contributing funding, we expect we will become more engaged with this study than we otherwise would have been. One of our conditions for recommending this grant is that we will view and approve the study’s pre-analysis plan, which gives us an opportunity to weigh in on the study design and potentially contribute to improvements in the research overall. We also expect to receive the study’s findings earlier than we otherwise would have done, though, as mentioned above, we have not yet received a finalized timeline for this grant.
- We may have the opportunity to fund additions to the study that would be informative for us. While we have not yet received cost estimates for these, we understand the researchers are open to measuring additional outcomes (such as child mortality), or randomizing additional elements that could improve cash transfers’ cost-effectiveness (such as providing the transfer during the “lean”, i.e. non-harvest, season, or targeting the transfers to pregnant women).16 Collecting data on child mortality could help us to triangulate other sources of data that we more typically use for this, which could be informative for our grantmaking more generally.
- We view this grant as “on strategy” for GiveWell’s livelihoods grantmaking portfolio. We are currently engaged in conversations with GiveDirectly about ways to increase the cost-effectiveness of their programming, such that GiveDirectly could clear our cost-effectiveness bar and receive direct grant funding from GiveWell. We view this study as directly relevant to this longer-term area of focus, given that it addresses key uncertainties from our updated cost-effectiveness estimate of GiveDirectly, specifically around spillover effects in poorer, less-integrated areas.
Risks and reservations
Our main reservations about this grant are:
- We are not sure whether the researchers will be wholly impartial towards GiveDirectly. Dennis Egger, who worked closely with GiveDirectly in the past, has recently received a grant to establish a Centre for Macro-Experimental Development at the University of Oxford, and our assumption is that this center would benefit greatly from a strong working relationship with GiveDirectly. We think it’s possible that this incentive could bias the researchers towards GiveDirectly within this study. However, we think there are positives to having the same lead principal investigator for this study as for the Kenya study, such as consistency in outcome measurement, and the maintenance of institutional knowledge. To mitigate against the risk of bias, we communicated with the researchers that our decision to recommend funding is contingent upon the following four conditions, which they accepted:17
- Disbursement of the grant funding will follow after GiveWell’s approval of the study’s pre-analysis plan;
- The researchers will publish their data and code, as they did for the Kenya study;
- GiveWell will likely ask a third party to replicate the findings of this study (as we did for the Kenya study when we revisited our GiveDirectly cost-effectiveness estimate);18
- No-one employed by GiveDirectly will be on the study’s research team.
- We think this funding gap would likely have been closed without GiveWell. The researchers have already raised the majority of the funding for this study (through a UK Research and Innovation Future Leaders Fellowship),19 and our general impression is that they are skilled fundraisers. Given this, we think it is likely that another funder would have contributed to this study in our absence. However, for the reasons mentioned above, we think it is right for us to contribute to this study despite this.
Plans for follow up
- Prior to the disbursement of funding for this grant, the researchers will submit a draft pre-analysis plan for GiveWell’s feedback and approval.
- We plan to have regular check-in calls around key study milestones, such as the launch of the random assignment and the completion of data collection.
- We’ve asked the researchers to notify us prior to making any decisions that could materially affect the design or implementation of the study.
Internal forecasts
For this grant, we are recording the following forecasts:
| Confidence | Prediction | By time | Resolution |
|---|---|---|---|
| 20% | The economic multiplier constructed as part of this study in Malawi is >2.5 (the same as the multiplier in the Kenya study). | December 2028 | - |
| 33% | GiveWell increases its average cost-effectiveness estimate of GiveDirectly to greater than 4x cash transfers.20 | May 2029 | - |
Our process
- Two GiveWell researchers spoke to Dennis Egger in August 2024 to discuss his research agenda, GiveWell’s main open questions about GiveDirectly, and potential ways that his research could address our uncertainties.
- In 2024, GiveWell ran a request for proposals (RFP) for research to inform our grantmaking (alongside Open Philanthropy).21 Dennis Egger submitted the proposal for this grant through the RFP in November 2024.
- We conducted a light-touch investigation into this grant opportunity in November and December 2024.
Sources
- 1
The transfer amount of ~$1,000USD is nominal, i.e., not adjusted for inflation.
- 2
“We find large impacts on consumption and assets for recipients. Importantly, we document large positive spillovers on non-recipient households and firms, and minimal price inflation. We estimate a local transfer multiplier of 2.5”. Egger et al., 2022, p. 2603.
- 3
For more on this update, see this blog post or the full report.
- 4
Unlike the general equilibrium RCT that this grant is supporting, this pilot evaluation did not include long-term follow-up, because the control group received the treatment after 6-8 months, and did not measure spillover effects or collect household data other than expenditures. “GD did a pilot in Khongoni TA in Malawi last year. Rollout was randomized at the GVH level and was universal. We collected a) monthly market prices, and b) monthly expenditure data. The problem with this setup: no long-term impacts (everyone is treated after 6-8 months), no spillovers, no household data except expenditure [...] this study was instrumental in informing the next phase (described in the proposal). It will also serve as input for our trader information intervention, i.e. it will tell us what kind of demand shocks to expect, and whether we can reduce the inflationary impacts.” Dennis Egger, email to GiveWell, November 26, 2024 (unpublished)
- 5
- “This project will study the largest Randomized Controlled Trial (RCT) implemented to date by the NGO GiveDirectly (GD). GD will provide universal unconditional cash transfers of USD 550 (125% of annual household income) to 110k adults in Malawi’s Chiradzulu district, representing 50% of the population and 60% of district GDP. This program represents the next phase in GD’s ambition to scale up to treating entire countries over the next decade.” Egger et al., “GiveWell/Open Philanthropy Research Proposal – High-Frequency Engel and Supply Curves in General Equilibrium: Experimental Evidence from Large Universal Cash Transfers in Malawi,” 2024, p. 1.
- For more information on why a transfer size of $550 was chosen for this geographical location, see this footnote to our November 2024 report on GiveDirectly’s Cash for Poverty Relief program.
- 6
“For the purpose of this study, to be eligible for transfers, households had to live in homes with thatched roofs, a simple means-test for poverty. In treatment villages, GD enrolled all households that met this criterion (“eligible” households) as classified by their field staff through a village census, and confirmed via two additional visits… Approximately one-third of all households were eligible.” Egger et al., 2022, p. 2607.
- 7
“However, GiveWell notes that important open questions remain… This study is designed to address some of these gaps. It will allow us to a) understand how the large spillover effects from Kenya generalise to universal (as opposed to targeted in Kenya) transfers, much larger transfers (125% of GDP as opposed to 15% in Kenya), and much poorer settings that are less integrated in trade; b) whether the inflationary impacts remain low, or whether they can be muted by very low-cost information treatments.” Egger et al., “GiveWell/Open Philanthropy Research Proposal – High-Frequency Engel and Supply Curves in General Equilibrium: Experimental Evidence from Large Universal Cash Transfers in Malawi,” 2024, p. 2.
- 8
“Our study aims to understand the effect of transfers at this scale and over time. On the supply side, does the demand shock cause disruptions or inflation? Or does it increase trader entry, mobility, imports, investment, hiring, or a reduction in excess capacity? On the demand side, do consumers re-allocate their spending to markets with lower inflation?
To answer these questions, cash transfers will be randomized at different geographic levels, generating variation in cash flows into markets. High-frequency market surveys will track expenditure, inflation, and market activity, while monthly surveys will monitor recipient spending patterns. Matching data across these surveys, we will construct short-run supply curves at the firm and market levels offering insight into supply impacts.
We will then test whether providing information to a randomly selected group of traders about future increased demand for their products reduces inflationary pressure. The market surveys will track price effects, and we track traders’ responses using a newly developed app at daily or weekly frequency. This will provide insight into how microenterprises respond to demand uncertainty, with implications for macroeconomic dynamics in low income countries.”
Egger et al., “GiveWell/Open Philanthropy Research Proposal – High-Frequency Engel and Supply Curves in General Equilibrium: Experimental Evidence from Large Universal Cash Transfers in Malawi,” 2024, p. 1. - 9
"Randomization of cash: Randomization will be across 3 different geographic units (villages, GVHs, and TAs) in order to create variation in exposure to treatment not only at the village level, but also at higher geographic orders, allowing us to estimate treatment effect spillovers across larger geographies". For full details of the study design, see the Egger et al.’s proposal.
- 10
See Egger et al.’s GiveWell Budget, November 23, 2024.
- 11
See Egger et al., GiveWell Budget, November 23, 2024.
- 12
Comprises:
- monthly high-frequency household surveys to monitor recipient spending patterns: $120,000
- monthly price surveys: $48,000
- a monthly market census and survey: $144,000
- 13
- 14
Dennis Egger, email to GiveWell, December 13, 2024 (unpublished)
- 15
For more on how we use cost-effectiveness estimates in our grantmaking, including how our (now-outdated) estimate of GiveDirectly’s cost-effectiveness acts as our funding benchmark, see this page. We are continuing to use this outdated estimate for now to preserve our ability to compare across programs, while we reevaluate the benchmark we want to use to measure and communicate cost-effectiveness.
- 16
“Child mortality data: I am very excited about collecting this in this study, too. I've been thinking about two ways about this:
a) We could, in our household sample, over-sample women of childbearing age, or women who are pregnant or planning to be pregnant at the time of baseline. This would then increase our power to look at behaviors surrounding pregnancy (pre- and post-natal care, labor supply, nutrition, birth outcomes and delivery modalities, etc.). It would of course come at the cost of representativeness. So, what this might involve is to aim for a slightly higher sample size, with a core sample, and a pregnant-women-sample.
b) For mortality outcomes: As mortality is a very rare outcome, we need a lot of sample size. So, we'd have to collect this during censuses (as well as during the surveys). This gets quite expensive (as you know from Kenya), as doing this properly will cost several 100k. I have begun thinking about potentially more efficient ways, e.g. by relying on community health workers to keep track of births and deaths and regularly check in with them. There is also a chance to get outcomes from clinics but will have to look more into this.
Bottom line: Given your interest and ours in this, we'll think more and do a bit of scoping to get a better sense of the various ways of doing this and what it would cost. I'd get back to you on this.”
Dennis Egger, email to GiveWell, December 7, 2024 (unpublished) - 17
Dennis Egger, email to GiveWell, December 7, 2024 (unpublished)
- 18
For the reanalysis we commissioned from independent economist and data scientist Michael Wiebe, see here.
- 19
“We have already raised an additional USD 1.4 million from the UKRI Future Leaders’ Fellowship awarded to Dennis Egger for setting up the Oxford Centre for Macro-Experimental Development. The fellowship will additionally pay for PI time, and the hiring of a research manager, post-doctoral researcher, and pre-doc ensuring substantial additional personnel, management, oversight, and research assistance. The fellowship lasts for 4 years and encompasses the broader project, and the setup of the RCT and data collection infrastructure for the next decade. As such, it cannot be used fully for fieldwork only, and needs to be complemented by additional funding. We ask for funding from GiveWell specifically for all price, market, and enterprise related activities of this next phase of the program to complement the core infrastructure. This ensures that GiveWell’s money will be spent highly cost-effectively, exclusively on fieldwork.” Egger et al., “GiveWell/Open Philanthropy Research Proposal – High-Frequency Engel and Supply Curves in General Equilibrium: Experimental Evidence from Large Universal Cash Transfers in Malawi,” 2024, p. 7.
- 20
“Cash transfers” here refers to our now-outdated, earlier cost-effectiveness estimate of GiveDirectly.
- 21
Once this process has concluded, we plan to publish a blog looking at how this went.