When Public Research Pays Off Depends on What You Measure, Two Landmark Studies Show

Two academic studies of public research investment—one tracking the Apollo space programme across the entire United States economy, the other following laboratories struck by fires, floods and structural failures—arrive at a conclusion that complicates decades of policy assumptions: the economic returns to public science depend heavily on what kind of knowledge is produced, how the money is spent, and the industrial context in which it lands. Neither study settles the general question of what public research returns, but each demonstrates what becomes measurable when the identification problem is solved, and in both cases the answer challenges a prevailing assumption.

The analysis, published in EPJ Research Infrastructures by Fiona Larner, Alexander Whalley and Stefano Baruffaldi, argues that large-scale research and technology infrastructure decisions should be evidence-informed, yet the long-term, economy-wide effects of these investments remain difficult to establish. The evaluation literature has identified the constraints: attribution is hard, comparison groups are rarely available, and the interval between investment and measurable effect often exceeds the evaluation window. Big Science is typically evaluated by funding bodies, often in five- to ten-year windows, and commonly restricted to capturing the value the investment brings to a single member state. Some academic evaluations adopt considerably longer horizons, and the horizon matters—the lag between procurement and a supplier’s patenting response can exceed a standard evaluation window, so a short study may record a null that reflects timing rather than the absence of an effect.

What has been missing, the authors argue, is not recognition of these problems but settings in which they can be overcome. Historical episodes occasionally supply variation in research funding or research capital that is large, abrupt and unrelated to expected returns—the conditions under which causal effects can be credibly identified. Academic research is well placed to exploit these settings because it operates on longer timescales than commissioned evaluation allows, is not tied to the reporting cycle of a particular investment, and can pursue questions wherever the available variation leads.

The macroeconomic setting is the Apollo programme. Launched in 1961, it produced a mass expansion of publicly funded R&D reaching one percent of GDP at its peak, over roughly 14 fiscal years, at an estimated cumulative cost of about US$152 billion in 2026 dollars. For comparison, approximately US$34 billion in 2026 dollars was spent on the Manhattan Project over 5 fiscal years. What makes Apollo analytically useful is not its scale alone but its origin: the expansion was driven by political rivalry rather than by expectations of economic return, and it was concentrated in places and technologies for reasons unrelated to their growth potential. That combination supplies the variation ordinary evaluation settings lack, allowing researchers to follow funded areas over the long term and weigh what happened against what would have happened without the investment.

The findings are striking. The changes attributable to NASA funding included an increase in employment, scale of output and capital, some of which had a long-term effect. However, there was an absence of productivity spillovers where the impact of newly developed technologies would be expected to be clearest. In the Moonshot era it was claimed that NASA research investments had a return of 7 to 1 for every dollar spent—a figure comparable to the ratio determined through evidence-based analysis by the UK Government’s Department for Science, Innovation and Technology, which found an average long-term return of £8 in net economic benefits for every £1 of civil public R&D investment. But NASA’s investment did not yield this return. Kantor and Whalley calculated a multiplier of 0.3 during the space race, and 0.4 following it. The authors caution that these NASA and DSIT estimates should not be interpreted as measures of an identical outcome, since they differ in scope, methodology and the categories of benefit included; their divergence nevertheless illustrates how strongly estimates of public research returns depend on what is measured, over what period, and against which counterfactual.

The explanation lies in the nature of the investment. NASA’s spending was intended to exploit or accelerate technological developments already underway within US firms. Rather than generating new ideas and innovations that would increase productivity—and thereby yield higher returns—it primarily stimulated additional manufacturing activity among existing companies. In such applied, mission-oriented investment, technological change and innovation tend to be incremental, while a substantial proportion of expenditure is directed towards manufacturing. This is not to say that mission-oriented procurement never generates innovation: a quasi-experimental study by Castelnovo, Clò and Florio finds that firms increase patent applications by approximately 10 percent relative to a control group following procurement, with the effect persisting and growing over subsequent years. The contrast with the Moonshot findings is instructive rather than contradictory—patenting and productivity are distinct outcomes, and the two settings differ in scale, era and the maturity of the supplying industries. DSIT’s 8-to-1 return, meanwhile, relates to the whole R&D pipeline and all sectors within it, from basic discovery research through to applied research.

The implication is that, if the objective is economy-wide productivity growth, the rationale for public investment may be stronger for basic research than for the mission-oriented research exemplified by NASA during the Space Race. A parallel contrast emerges within the cost–benefit tradition: applying a common analytical framework to two facilities built on the same underlying accelerator technology, one study assesses the Large Hadron Collider as fundamental physics infrastructure while another assesses the CNAO hadron-therapy centre as applied medical infrastructure. In the applied case, benefits accrue predominantly to patients and depend heavily on the services delivered to users, rather than on the wider spillovers that dominate the basic-science assessment. Basic research is the domain in which spillover effects are likely to be greatest, particularly where the work concerns areas in which private-sector investment is limited.

The microeconomic study tackles a different question: what physical capital—tools, laboratory space, equipment and infrastructure—actually contributes to scientific output. Its value is difficult to observe directly, because laboratories that hold more of it also differ in funding, seniority and field. What is required is variation in capital that arrives independently of these characteristics. Laboratory disasters supply it: fires, floods and structural failures destroy research capital for reasons unconnected to the productivity of the scientists affected. The setting is constructed to preserve that independence—events are restricted to those isolated at the laboratory level and the tools within it, excluding disasters that damaged roads, utilities or the surrounding institution, since wider destruction would confound the loss of research capital with disruption to everything else. Data were collected principally through news articles, supplemented by surveys sent to laboratory leads and merged with publication data.

The result is a long-term loss of roughly 20 percent in publications for scientists in laboratories hit by disaster, an effect that is stronger if the quality of publications is accounted for. The loss is driven by the destruction of specialised capital—highly specific equipment that serves explicit research objectives, addresses questions not answered before, requires substantial time investment to develop, and embodies substantial tacit knowledge, making it non-transferable between scientists. Generic, standardised and replaceable capital, by contrast, can be restocked, and in some cases temporary access to unaffected laboratories is available on a solidarity basis, so only very short-term negative effects are seen where disasters affect only generic capital. Specialised capital, once lost, can hardly be repurchased on a market and requires new investments of time in often uncertain processes, so recovery is less likely. Yet laboratories established for longer can recover: the loss of physical capital creates an opportunity for researchers to change research direction, breaking path dependency in trajectories once considered optimal but favoured primarily because of vested investment in specific, non-liquidable assets. For more modern laboratories, which have made recent specific investments in cutting-edge topics, changes in direction are less likely to be beneficial.

The policy implications are concrete. Economic effects depend on the type of knowledge produced, the composition of expenditure and the industrial context, so evidence from one programme may not transfer to another. Where mission-oriented investment is expected to generate wider economic spillovers, that expectation should be evidenced rather than assumed—in the Apollo case, spillovers were weaker than the standard basic-research rationale would predict, though the Italian procurement study finds positive innovation effects in a different context. The value of research infrastructure includes not only financial cost but the time and knowledge embodied in tools and equipment, so consideration for the opportunity cost of time investment needs to join monetary expenses to fully capture the economic value of R&D infrastructure. The authors recommend retaining and opening up administrative data, matching evaluation horizons to the expected lag between investment and effect, and being clear on whether the case for investment rests on the mission itself or on expected economic spillovers. They also point to a knowledge gap: Kantor and Whalley find relatively modest economic effects from the Space Race, yet in agriculture they find substantial productivity effects from early twentieth-century public research investment, particularly through gains to nearby farmers—evidence that returns depend on the setting, and that the case for any particular investment has to be made rather than presumed.

Subject of Research: Economic returns to public research investment, examined through the Apollo space programme and laboratory disasters

Article Title: Evidence for Science Policy: What Can Public Research Do for Innovation and Growth?

Article References: Larner, F., Whalley, A., & Baruffaldi, S. (2026). Evidence for Science Policy: What Can Public Research Do for Innovation and Growth?. EPJ Research Infrastructures, 10(1), Article 30. https://doi.org/10.1007/s41781-026-00185-4

Image Credits: AI Generated

DOI: 10.1007/s41781-026-00185-4

Keywords: public research funding, Apollo programme, NASA, economic growth, productivity spillovers, research infrastructure, laboratory disasters, specialised capital, mission-oriented research, basic research, science policy, evaluation methods