The Grant Is Not the Treatment: Why Dose Matters
This series examines PIPE, São Paulo’s long-running public R&D grant program for small and medium-sized firms, and asks a deceptively simple question: does giving a small firm a larger subsidy reliably produce more innovation? The answer matters to anyone deciding how scarce public money should be allocated. A grant can remove a financing barrier, but a firm also needs people, equipment, knowledge, and managerial capacity to turn that money into research and eventually into an observable result.
The evidence we will follow comes from firms that applied to the same PIPE calls, some of which received funding and some of which did not. That comparison makes it possible to study whether grants increase patenting. But the deeper question is about dose. The relevant contrast is not only funded versus unfunded; it is also a very small grant versus an intermediate grant versus a large one, each interacting with what the firm and its project can actually absorb. Understanding that distinction will make it possible to judge both the strength of the evidence and the limits of the policy lesson.
The program behind the question
PIPE is the Innovative Research in Small Businesses Program operated by the São Paulo Research Foundation, FAPESP. It was established in 1997 and began operations in 1999 to support scientific and technological research in innovative firms located in São Paulo. The program provides non-repayable R&D grants to firms with up to 499 employees. This is not a generic loan scheme or a tax deduction whose value depends on a company’s taxable income. It is direct public support for research projects proposed by smaller firms, with the firms retaining ownership of the resulting intellectual property.
That institutional setting makes the program a useful case for thinking about subsidy intensity. PIPE applications are assessed within the same calls, under the same broad eligibility rules, information conditions, evaluation criteria, and budget constraints. Technical merit and the feasibility of the proposed research project matter for selection; prior patenting is not part of the evaluation score. An applicant that is not funded therefore differs from a recipient in a setting where both entered the same contest, rather than being a company that never sought public research support at all.
The awards themselves are not evenly distributed. In the study’s observed data, most grants are at the lower end, while a long tail extends to approximately US$600,000. About 22% of grants fall between US$100,000 and US$200,000, 10% fall between US$200,000 and US$500,000, and fewer than 1% exceed US$500,000. The paper reports an average expenditure per firm of US$88,143.76 over 1999–2019 and an average benefit duration of 3.7 years. These figures describe the program’s historical distribution; they are not instructions about what every firm should receive.
The distribution matters because the policy question is not solved by proving that public support can help. If all grants were identical, an average funded-versus-unfunded comparison might be enough for a first judgment. But when awards vary widely, the average combines firms receiving different amounts for projects with different needs. A small award may leave a technically promising project unable to hire specialized staff or buy essential equipment. A much larger award may provide resources that the firm cannot coordinate or deploy effectively. The same word—“grant”—can therefore describe materially different treatments.
Why receipt alone is not enough
The usual public debate begins with a binary question: did the firm get the money? That question has a legitimate purpose. It asks whether the program made a difference compared with what would probably have happened without funding. For PIPE, the study finds that funded firms later file more patents than comparable non-funded applicants. But an average effect can hide the allocation problem that follows: if support helps, how much support is useful for which kind of firm and project?
A dose perspective treats grant size as part of the treatment rather than as an administrative detail. The mechanism is intuitive without being automatic. At a low dose, the grant may be too small to overcome the fixed costs of an R&D project. Research needs are often lumpy: one skilled researcher, one specialized instrument, or one external technical partnership may be necessary before a project can move forward. A little money can be real assistance and still be insufficient to change the project’s trajectory.
As the grant becomes large enough to cover those complements, the return to additional support can rise. The firm may be able to assemble a team, purchase equipment, contract for expertise, or sustain a project long enough for an innovation to emerge. In this range, an extra dollar is not merely added to a budget; it helps complete a combination of inputs that was previously out of reach.
The logic can reverse at higher levels. Small firms do not have unlimited managerial bandwidth, organizational routines, or experience with complex research programs. If new resources arrive faster than the firm can recruit, coordinate, learn, or make decisions, the additional money may not translate proportionally into additional innovation. That does not mean the money is wasted, or that large grants are intrinsically harmful. It means that the value of an extra unit of support depends on the firm’s ability to use it productively.
This is the core idea of absorptive capacity. A firm must be able to recognize useful external knowledge, assimilate it, and exploit it in its own work. In a small business, that capacity can be constrained by finance, skills, equipment, management, formal organization, or prior experience with R&D. The study uses absorptive capacity to interpret why an intermediate level of support might outperform both a very small and a very large one. It does not directly measure a chain from grant dollars to each of those internal capabilities, so the mechanism should be treated as a plausible explanation of the observed pattern, not as a separately proven result.
The distinction between receipt and dose also changes what counts as a responsible policy conclusion. “More support can help” is not the same as “larger awards are always better.” Nor does a diminishing marginal return to patenting mean that the firm has received no other benefit from a large grant. It may build capabilities, credibility, collaborations, or knowledge that the main outcome does not capture. The question is therefore not whether money has value in the abstract, but whether the next increment of support is matched to the project’s requirements and the firm’s capacity to deploy it.
This map is a conceptual guide, not a claim that the study directly estimates every arrow. It identifies the relationship the evidence is trying to clarify: how the amount of PIPE support interacts with a firm’s capacity and resources before an innovation becomes visible in the data.
What will count as evidence
The study’s main observable outcome is patenting, measured primarily through fractional patent applications at the firm level. Patent activity is a valuable, comparable signal of technological output, but it is not the same as innovation in its entirety. Small firms may improve a product, process, or service without filing a patent. The analysis also does not directly measure forward citations, patent-family size, or every other indicator of patent quality. A finding about additional patents should therefore not be silently expanded into a finding about sales, productivity, employment, or total social value.
Timing matters as well. R&D does not normally become a patent at the moment a grant is received. The study follows patent outcomes over a three-year window, a period intended to allow research investment to become observable. Its event-study evidence indicates that effects emerge gradually: post-treatment differences become statistically significant from the second year and are described as stabilizing between roughly 0.15 and 0.25 additional patents after four years. Judging a grant immediately after payment could therefore miss the process it was meant to support.
The paper’s baseline estimate gives the first part of the answer. In its doubly robust comparison, funded firms register an average increase of 0.155 patents relative to matched non-funded applicants, with a reported standard error of 0.050. The number is modest in absolute terms but large relative to the control group’s descriptive mean of 0.016 patents. It shows why public funding is worth taking seriously in this setting. It does not yet tell us whether every grant amount works equally well.
The deeper result is that the modeled response is not linear. The study finds an inverted-U relationship across the observed distribution of grant amounts: very small grants often do not provide enough scale, innovation gains strengthen through an intermediate range, and marginal gains decline at higher levels. The estimated maximum occurs at 59% of the study’s normalized intensity scale. The paper translates that point into approximately US$340,655.18 for its setting, while explicitly warning that neither figure should be transferred directly to another program or region.
That warning is essential. The 59% figure is a modeled maximum based on PIPE’s observed grant distribution and institutional context, not a legal threshold or a universal funding target. The study’s conclusion is conditional: receiving support helped the firms studied; the strongest modeled patent response appeared at an intermediate dose; and the location of that dose depends on the firm population, the projects, the program rules, and the outcome being measured. The capacity explanation is persuasive as an interpretation, but it remains distinct from the empirical estimate.
The next question is consequently concrete rather than philosophical: what features of PIPE and its data make this comparison meaningful? Before deciding whether the curve deserves confidence, we need to understand how applications, funding decisions, firm records, and delayed patent outcomes are connected. That foundation will show what the study can compare—and where its counterfactual still depends on assumptions.