In February 2025, USAID was effectively shut down. Unsurprisingly, most of the attention focused on the immediate humanitarian consequences: health programs, food aid, and emergency response. But something else was happening that received far less attention: major disruptions to science funding and scientific collaboration in low- and middle-income countries (LMICs).

Many programs aimed at supporting researchers in LMICs were paused or cancelled. And major science funders in the U.S. also shifted toward domestic priorities, making international collaboration and support for foreign science more difficult. Given the central role the United States plays in the global scientific ecosystem, these changes have had, and continue to have, far-reaching implications.

As policymakers, funders, and researchers consider what comes next for science in development, a fundamental question arises: what does the evidence actually tell us? Before deciding how to invest scarce resources, we need to understand the returns to supporting science in LMICs, which approaches to strengthening scientific capacity have proven effective, and where the greatest opportunities for impact lie. Just as importantly, we need to understand the limits of our current knowledge and where critical evidence gaps remain.

In this post, we take stock of the economic evidence base. We review what we know about why science in LMICs matters, where scientific activity is taking place, and the policy and funding levers that appear to support the growth of scientific capacity. (Stand by for a future blog where we will discuss what we still really need to understand better.)

Why science in LMICs matters

Science has long been recognized as a key engine of economic and social progress, driving improvements in productivity, health, and technology. But we are learning that investing in science in LMICs can generate especially large returns.

A good example of the research demonstrating this is an evaluation of the rollout of the Embrapa program. In the 1970s, Brazil invested heavily in agricultural R&D through a network of regionally distributed research centers. After the rollout of these public research centers around the country, Brazil transitioned from being a major food aid recipient to one of the world’s largest agricultural exporters. The figure below comes from the paper showing this rapid advance in agricultural productivity in Brazil over the years.

But was this actually because of investments in these R&D centers or something else?

The authors of this paper carefully estimate the causal impact of the rollout of Embrapa centers. As a first step, the authors find that Embrapa scientists were significantly more likely to study their centers’ own local ecological conditions rather than conditions elsewhere. This redirected research effort toward problems most relevant to local farmers in each municipality with an Embrapa center.

So did this increase in research output towards localized problems impact agricultural productivity?

They find that it did. Places with an Embrapa center experienced increases in agricultural productivity, but not only that - the authors also show that places with ecological conditions more similar to those near Embrapa centers experienced larger productivity gains, even when those places were geographically distant from the labs themselves.

Using this ecological exposure advantage of places that did not have an Embrapa center, but benefited nevertheless, the study estimates substantial gains in agricultural productivity. Overall, Embrapa’s research investments are estimated to have increased aggregate agricultural productivity in Brazil by roughly 110 percent, with a benefit-cost ratio of about 17 to 1.

And this pattern is not unique to this program. Across health, agriculture, and scientific publishing, we repeatedly see evidence that relatively modest investments in LMIC science can generate outsized returns, particularly in comparison to investments in science and innovation in higher-income countries, like the U.S. The figure below illustrates additional evidence from prior studies showing the returns per dollar invested in neglected disease R&D and agricultural productivity (top panel - evidence from here, here and here), as well as comparative estimates of the cost of producing an academic publication in LMICs versus the U.S. (bottom panel - evidence from here and here.).

Note. These figures should be taken with a grain of salt, as cross-study comparisons are difficult due to differences in methodologies, outcomes measured, and the scope of R&D returns considered. Nonetheless, they suggest (1) the value of developing more comparable and credible estimates of returns across settings and (2) that even if investments in LMIC science had half the return we see here, they would be a good investment.

Not only are there high returns overall, but science that takes place in LMICs is also important for solving local problems that might not be solved otherwise. Many LMIC challenges are context specific and emerging research has found that what gets researched depends on where researchers are located. Which implies that fewer researchers in some locations would result in less research on issues affecting those places.

One of Caroline’s papers looked at scientific responses to the 2014 Ebola outbreak in West Africa. Scientists in the affected countries were significantly more responsive during the outbreak than otherwise similar scientists elsewhere. In large part, this was because they had access to the local knowledge, patients, samples, and institutional context needed to do the work quickly.

We are also beginning to accumulate evidence that policymakers, medical professionals, and even patients often place greater trust in locally generated evidence, or evidence that is about populations similar to them, rather than evidence produced elsewhere. So building scientific capacity in LMICs is not just about producing knowledge, it’s also about making knowledge usable.

And finally, science in LMICs produces spillovers that benefit everyone. Research conducted in LMICs contributes to disease outbreak control, agricultural resilience, and climate adaptation globally. Knowledge generated anywhere can diffuse everywhere.

The scale of the imbalance

Despite the clear benefits of LMIC science, global science remains extraordinarily concentrated, and global scientific output remains highly unevenly distributed. The figure below shows that publications per capita are concentrated in high-income countries, and limitations in data availability mean that scientific contributions from many LMICs are difficult to measure accurately.

The same is true for R&D spending.

But one thing that became very obvious while preparing for our May 2026 workshop is how difficult it is to even quantify development-related scientific investment (although we provide some estimates below). Government data on R&D spending in LMICs is difficult to find, as is foreign aid and philanthropic spending in LMIC science and innovation. USAID, for example, reported around $252 million in R&D spending in FY2023. But that number is almost certainly incomplete. The agency spent more than $270 million on evaluations alone that year, some of which arguably qualify as research activity (and some of which may have been reported as part of the $252 million in R&D expenditures - we just don’t know). And even where spending data exists, we often do not know how much of that spending actually supported LMIC researchers directly versus supporting research conducted elsewhere. This data gap matters because it makes serious prioritization extremely difficult.

What seems to matter

Given that right now increasing donor funding for science might be challenging, we can turn to considering the most effective levers for promoting science in LMICs.

Broadly, we organize the evidence on these levers into a few categories: human capital, networks, funding, and institutions/infrastructure.

Human capital

A growing body of research suggests that training and mentorship programs that aim to develop human capital in LMICs matter a great deal for science conducted in these contexts.

One example is a paper studying the Structural Transformation of Agriculture and Rural Spaces (STARS) fellowship program, which pairs African scientists with mentors at U.S. universities. The authors of this paper compared finalists who participated with finalists who narrowly missed out due to mentor availability rather than applicant quality. They found meaningful increases in citations and longer-term scientific engagement among participants.

What stands out from this work is that the benefits are often not purely technical. Trainees consistently emphasize the importance of learning the “hidden curriculum” of science: navigating ethics approvals, publication processes, collaboration norms, grant writing, and professional networks.

Other studies exploring alternative U.S. fellowship structures for foreign scientists (like the Fogarty Scholars and Fellows Program) report positive results as well. Benefits don’t just accrue to the participants, though; they also spill over to the participants’ colleagues.

Networks

Scientific networks may be one of the most underestimated drivers of productivity.

There is growing evidence that internationally mobile scientists help connect researchers in their home countries to global scientific communities. These links facilitate collaboration, information flows, and access to opportunities.

For example, Caroline studies African scientists who participated in the NIH Fogarty AIDS International Training and Research Program (AITRP), which supported periods of training in the U.S. She finds substantial spillover effects on the colleagues of trainees who returned home after their training. These spillovers appear to arise because returning trainees acted as bridges between U.S.-based scientists and researchers in their home countries, creating new collaborative connections and closing previously disconnected parts of the scientific network.

Funding

The evidence on the impact of providing financial support directly to scientists, particularly those based in LMICs, remains surprisingly limited. Nevertheless, the existing studies suggest that grants and prizes can substantially increase scientific productivity, with effects that may be especially large in settings with lower initial levels of scientific capacity. For example, a study examining research grants in the USSR finds sizeable impacts on scientific output. As discussed earlier, these effects appear particularly large when compared with estimates of the returns to research funding in the U.S., suggesting that investments in under-resourced scientific systems may generate especially high returns.

Institutions and infrastructure

Lastly, the institutions, incentives and infrastructure supporting scientists appears to matter. A study from the Indonesian context shows that the introduction of publication-linked incentive systems shape the behavior of scientists in very predictable ways, and promote scientific output in the country. And although not evidence from LMICs, recent work has found that access to equipment and open data can dramatically change who is able to participate in science, which gives promising insight into what we might expect if these inputs are made widely available.

This evidence provides a strong starting point, and, if translated effectively into policy and funding decisions, could help unlock substantial scientific and societal benefits. However, we still have major gaps in our knowledge. Much of the existing economic evidence comes from a small number of programs, countries, and interventions, leaving important questions unanswered about what works, for whom, and under what conditions.

In our next post, we’ll explore these unanswered questions and outline a research agenda for understanding how investments in science can most effectively contribute to development, innovation, and problem-solving in low- and middle-income countries.

• You know how suggested you “go forth and export”? The Africa Jobs Fund now has a fellowship to support people building a career in manufacturing in Africa. I would totally do this if I did not have, erm, several other jobs.

• has ideas for metascience experiments.

• Boko Haram finds AI useful in planning terrorist attacks. The security impacts of AI will not only be felt in rich countries…

• The Centre for AI Security and Access summarizes AI philanthropy so far.

• From : data center groups (largely in the US) will consume as much power as entire African countries.

• A New York man was evicted from his apartment for owning too many books. I view this story as aspirational.

• As of 2010, no residents of Manhattan’s Little Italy were born in Italy.

• Chipotle is expanding to Mexico. Love a backported restaurant - there are several Korean restaurants that were started in LA and now have locations in Seoul.

• Three different people have sent me this ad for Kazakhstan:

Do not fear, the algorithm also served it to me.

A few weeks ago, we brought together people who had worked on international science in the U.S. government, researchers who study innovation and economic growth, and practitioners who design and implement programs to support science in low- and middle-income countries (LMICs) - some of our favorite thinkers and doers in science for international development. We met in Washington, D.C., to discuss what we know about science in LMICs: its returns on investment, what works in strengthening scientific capacity, and how those of us working from the U.S. and Europe can help advance it.

Even before the shutdown of USAID and the pulling back of other funding sources, science accounted for only a small share of U.S. government and philanthropic spending. And while evidence of the high returns to research and development (R&D) in low- and middle-income countries (LMICs) has grown substantially, our understanding of what works in strengthening scientific capacity has remained relatively limited. When Andrew worked at USAID, for example, there was little rigorous evidence to guide investments in science systems. With these existing challenges, and new resource limitations, we felt like it was a critical time to get together and try to make some progress on key questions that keep us all up at night around science for development.

In this workshop, we aimed to take stock of where the field stands today and explore what is needed going forward. Over the course of the day, we began with a presentation by Caroline and Liz Lyons (UC San Diego) about the evidence base for science in LMICs and evidence gaps. We then convened a panel of leaders implementing science-for-development initiatives: Fiona Moejes (Mawazo Institute - calling in from Kenya), Amy Jamison (Alliance for African Partnership at Michigan State University), Sasha Gallant (DIV Fund), Maya Ranganath (Center for Effective Global Action at UC Berkeley). They discussed a range of approaches, including a multi-university U.S.–Africa partnership, support and funding for women PhD students in Africa, scaled funding for development innovations, and training and fellowship programs for economists from LMICs.

Finally, we facilitated a theory-of-change exercise, asking participants to map pathways for strengthening science in LMICs and imagine what success might look like in the future. Some of the outcomes participants focused on included:

• Speeding up science funding and facilitating riskier bets

• Increasing the quality and quantity of LMIC publications

• Increasing the number of science graduate degrees in LMICs

• Ensuring that science investments support local livelihoods

The insights from these discussions and activities shaped the reflections that follow. We begin with a few high-level takeaways before turning to some more specific observations that jumped out during the workshop.

General reflections

This was a truly impressive group of people - intellectually curious, deeply committed, and eager to find ways to strengthen science in LMICs. Despite having no budget to support travel, participants came from across the U.S. and the United Kingdom, and our panelist Fiona Moejes would have joined in person from Nairobi had travel disruptions not intervened. There was a shared sense that this is an underappreciated, but critically important, topic. We were already convinced of the need for more evidence and ability to better translate evidence to decision-makers, but this further solidified that belief.

It is important to acknowledge who was not in the room. Given the constraints of a small budget and a Washington, D.C.-based meeting, we were unable to engage many of the leaders and practitioners working on science systems within LMICs themselves. Our hope is that we can hold future convenings that place greater emphasis on regional perspectives and locally grounded challenges and opportunities.

Some more specific thoughts from participant discussions:

1. Evidence on what works to improve science in LMICs (and its importance) exists - but it is fragmented, under-visible, and weakly connected to decision-making

One theme that emerged in our discussions was that while some evidence exists on the importance of science in LMICs, and on what works to strengthen it, it is fragmented, under-visible, and only weakly connected to decision-making.

Existing evidence suggests a high return on investment to LMIC science (look for more on this in an upcoming blog!). For example:

• Research into local environments and factors in LMICs – such as agriculture, which varies widely by climate – can boost productivity massively. For example, Embrapa, Brazil’s agricultural research program, increased aggregate agricultural productivity in Brazil by 110%, with a benefit cost ratio of 17.

• R&D in LMICs can also have a much higher return on investment than in the US, with lower research costs and more low-hanging fruit to pick.

• Scientists from LMICs that are trained in the US produce spillovers, increasing their local colleagues’ productivity as well.

But we need more social scientists to study this and more evidence on how best to support LMIC science (funding, human capital, institutional strengthening, etc.).

We have noticed that awareness of the existing evidence base was surprisingly low, even among participants who deeply engage in science, innovation, and development. Relatedly, we came away with the sense that there are few institutions or systems focused on synthesizing evidence about science and innovation in LMICs and translating it for policymakers, funders, and practitioners. There is no think tank for innovation in LMICs, for example. Even for decision-makers who want to use evidence, finding the relevant research can be difficult.

This reinforced our view that strengthening the evidence ecosystem requires more than producing new research (which is important too!). It also requires investing in the institutions, networks, and intermediaries that can help existing evidence inform decisions.

2. We should design programs that are both useful and which we can learn from

Many programs already exist that could help us learn more about how to strengthen science in LMICs. For example, there were four different fellowship programs represented at the workshop, alongside a range of other initiatives focused on funding, training, and scientific capacity building.

But it is a challenge that many programs are not designed to be effectively studied. For example, they might have a very small number of participants, their selection approaches may be inconsistent, or they simply might not collect needed data.

We see an opportunity to address this in two ways. First, by evaluating existing programs more systematically, including exploring ways to combine data across similar initiatives when individual programs are too small to study on their own. Second, by designing new programs with learning built in from the start. Such programs can simultaneously strengthen science in LMICs and generate evidence that helps improve future investments.

3. Context matters more than we might account for

A strong theme that emerged was that “what works” in strengthening science in LMICs is highly context dependent: an approach that is effective in one country may not transfer well to another. For example, evidence suggests that fellowships for women may be less effective in countries with low gender equality, even within the Global South. Relatedly, there was discussion about how definitions of “impact” themselves can vary across settings, raising the question of whether we are consistently using the right measures.

This has several implications. First, because one of the issues is a lack of data and understanding of cross-country differences, it suggests a greater need for comparative and descriptive work alongside causal methods such as randomized controlled trials. Second, it points to the importance of deeper understanding of institutional differences across countries and contexts, rather than assuming interventions will generalize easily across LMICs. Finally, it underscores the importance of rethinking how we define and measure outcomes when evaluating programs or assessing the return on investment in this space. In particular, it highlights the value of working more closely with LMIC partners to define what impact looks like in different contexts.

4. The customer for this evidence is likely changing

An issue that came up repeatedly was the question of who the “customer” for evidence on strengthening science in LMICs actually is. There is growing and increasingly rigorous evidence on the importance of science in LMICs, and on approaches to building scientific capacity, that could inform decisions by governments, donors, and implementers. But do they want the information? Will this change decisions and behavior? And can we be more specific about who the customer is for the evidence?

Potential users of this evidence include LMIC governments, think tanks, NGOs, and universities, though the form and framing of evidence likely needs to be tailored to their specific needs and contexts. More innovation and science policy think tanks in LMICs themselves are warranted! One model that could work to build these is partnerships between institutions in high-income countries and LMICs to generate and disseminate evidence. For example, Berkeley’s Center for Effective Global Action has been effective in partnering with LMIC universities on economics research and capacity strengthening, and a similar approach could be applied to building knowledge on what works in strengthening LMIC science systems.”

Foreign donors and implementers are another key audience. We’re in a new age of development. The US government and other high-income countries are increasingly transactional and the arguments for foreign aid that were used before may not work now. While it may be unrealistic to expect major near-term U.S. government engagement on science in LMICs, European and Asian governments, multilateral institutions, and regional development banks continue to invest in science cooperation. For these, having access to better evidence on what works is especially important.

International philanthropy is also likely to play a growing role in funding LMIC science, particularly given the massive potential expansion of philanthropic funding in the U.S. Historically, philanthropic organizations have made substantial investments in LMIC science - even when the evidence base was relatively limited (for example, through funding the Green Revolution or founding the Consultative Group for International Agricultural Research). Philanthropy should draw on evidence about what works in LMIC science and help support the development of that evidence base itself.

5. LMIC scientific capacity still matters - even with AI

While working on the theory of change for human capital, we asked whether the need for improved scientific and technical capacity in LMICs might be reduced or even eliminated by advances in AI.

We think that is unlikely. Many of the most important challenges facing LMICs are deeply contextual, persistent, and not easily generalized, including complex physical data collection (soil samples, output from crops grown in different micro climates, blood draws, water samples) and few other researchers have studied them. While AI may automate certain tasks and reshape parts of the service sector (e.g., through the automation of call centers), it is unlikely to reduce the need for locally grounded research and scientific capacity.

The core ingredients of scientific systems (human capital, funding, networks, and institutions) will still be required in a world with powerful AI. If anything, they may become more important for ensuring that AI is used effectively in diverse local contexts.

It’s really hard to know how AI will affect LMICs. Development is hard to predict, and it’s difficult to know how AI will affect LMICs, development trajectories, or the returns to R&D. This uncertainty itself suggests a need for economists and other researchers to think more creatively about how technological change interacts with scientific capacity and development outcomes.

6. Diffusion from high income countries isn’t going to solve LMIC problems (but it can help)

A question raised by a participant was whether it might be more effective to incentivize top researchers in the U.S. and other high-income countries to work on problems relevant to LMICs. As acknowledged in the question (and written about by Caroline with Matt Clancy here), a scientist’s geography influences what they study - and HIC researchers don’t tend to study the things most important to LMICs.

While research conducted in high–income countries ultimately improves lives around the world (e.g., through basic research that feeds into new technologies, biomedical research that leads to new drugs), that doesn’t negate the need for local research in LMICs.

Incentivizing U.S. researchers to focus on LMIC issues would be great, but also is likely to leave three gaps still: 1. It can be very expensive 2. Many findings are not easily translatable and 3. Having local intellectual talent is a national sovereignty and security issue.

Final thoughts - and more to come

We’re convinced, even more so now, that we need more evidence and that we need to better disseminate it. We’ll share more thoughts - including a summary of Caroline and Liz Lyons’ evidence presentations - soon.