Introduction
The U.S. government is a global leader in research and development (R&D) investment. The rapid development of COVID-19 vaccines is one of many notable examples of economically critical innovation that was facilitated by federal support (Kates et al. Reference Kates, Cox and Michaud2023). In Fiscal Year (FY) 2021, the federal government spent $190.2 billion on R&D, which corresponds to about 0.7 percent of U.S. Gross Domestic Product (Anderson and Moris Reference Anderson and Moris2023). Most federal R&D spending (62 percent) was awarded to private and non-profit firms, universities and colleges, and other entities outside of the federal government. State governments committed almost $2.5 billion to R&D in FY21, with about $612 million supported by federal grants and the balance funded by own-source revenue and other non-federal sources.Footnote 1 Although total state R&D spending is dwarfed by that of the federal government, it grew at a 3.9 percent inflation-adjusted annualized rate from 2006 to 2021, compared to just 1.0 percent at the federal level.Footnote 2
R&D encompasses systematic and creative activities that help to advance human knowledge or involve the application of existing knowledge to solve problems (National Science Foundation 2018; OECD 2015). Investment in R&D leads to innovations that generate significant social returns but are difficult for private firms to capture (Nelson and Romer Reference Nelson and Romer1996). In other words, private firms are unlikely to engage in the welfare-maximizing level of R&D investment without government support. Nelson and Romer (Reference Nelson and Romer1996, p. 9) argue that “encouraging R&D investment is as important as investing itself.”
This raises a question as to how the United States can grow overall investment in R&D, including from subnational governments. Public funding arrangements for R&D projects in the United States reflect the complex nature of fiscal federalism. R&D is often a joint federal-state enterprise, with federal grants directly supplementing state expenditure on R&D or indirectly subsidizing it via pass-through intergovernmental transfers. The large literature on U.S. fiscal federalism demonstrates that federal grants influence the spending decisions of recipient governments. Economic theory predicts that federal grants will largely displace (“crowd out”) the spending of recipient governments on the subsidized function. Several prior empirical studies find support for this hypothesis, particularly in the context of state R&D expenditure (Wu Reference Wu2009; Wu and Merriman Reference Wu and Merriman2017). In contrast, many others find that intergovernmental grants spur recipient governments to spend more than they otherwise would, in effect “crowding in” additional spending by recipient governments (Hines and Thaler Reference Hines and Thaler1995; Inman Reference Inman2008).
Previous research does not consider whether federal grants affect state R&D spending beyond the academic sector. This paper seeks to close that gap by examining the relationship between federal grants and state expenditures for R&D performed by state government agencies separately from academic institutions. This is an important question as states directed almost $674 million in FY21 R&D expenditures toward state agencies, such as their departments of agriculture, transportation, and natural resources, representing about one-quarter of total state R&D expenditure. State agencies were the second most significant performer of state R&D after academic institutions, which received about 38 percent of total state R&D spending in FY21.
We argue that federal R&D grants to state agencies may elicit a distinct state spending response from those to academic institutions. We suggest this occurs because state R&D allocations are influenced by the geographic scope of the benefits that derive from R&D projects. We hypothesize that state agencies primarily engage in applied R&D projects that lead to operational improvements with localized benefits to state residents. By contrast, academic institutions pursue basic or specialized R&D that yields broad societal benefits, such as medical research. We evaluate these hypotheses by applying computational text analysis methods to federal R&D grants awarded between 2017 and 2019. This analysis shows that federal R&D awards to state agencies are concentrated in applied fields of inquiry – regulatory policy, transportation, and environmental management – that primarily yield localized benefits. By contrast, federal R&D awards to academic institutions are concentrated in human health and medicine.
We then test whether these distinctions meaningfully influence state allocations for R&D at state government agencies and academic institutions using expenditure data compiled over 2006 to 2021 by the National Science Foundation (NSF). We find that direct state spending on state agency R&D increased by $0.35 on average for each dollar of federal support using a traditional state-by-year fixed effects model, which rises to $0.39 when we apply an instrumental variables approach. We also find that federal R&D grants to academic institutions have no significant impact on state allocations for R&D at these institutions. This result contrasts with Wu (Reference Wu2009) and Wu and Merriman (Reference Wu and Merriman2017), which both find federal support for academic research effectively crowds out state spending. Regardless, all these empirical results are consistent with the hypothesis that state funding decisions in response to federal grants are influenced by the geographic scope of the benefits derived from the public good.
This paper is organized as follows. The next section provides an overview of state-federal research programs. We also discuss the prior literature on federal grant programs, which helps to motivate the conceptual framework that shapes our hypotheses and subsequent analysis. We then turn to the data, methods, and findings of a topical modeling analysis of federal R&D grants awarded to academic institutions and state agencies. We similarly describe the data, methods, and results of quantitative analysis of the responsiveness of state R&D expenditures to federal grants. We conclude with a summary of our key findings and the broader implications of this research.
Background
Public funding for R&D in the United States
The United States utilizes a blended system for publicly funding science and research, although the federal government plays the largest role overall. According to the NSF, about one-third of federally funded research is performed internally by federal agencies or federally funded research and development centers (FFRDCs). The remainder is performed externally, primarily by businesses and academic institutions. The federal government utilizes a variety of funding mechanisms for R&D performed externally, including contracts, grants, and cooperative agreements. The latter two are the most frequently used mechanisms to support R&D performed by state governments. Project grants are typically awarded through a competitive selection process to support programs or activities that meet specific criteria with minimal federal involvement. Cooperative agreements are similar but entail substantial federal involvement in implementation. Some grants and cooperative agreements include mandatory matching requirements, while others do not or permit voluntary cost sharing. Additionally, a smaller portion of federal funding is allocated based on predefined laws or statutory formulas, such as formula grants, which still require a formal application but are not awarded competitively.
There is a wide variety of federal R&D funding targeted toward states, and the specific funding mechanisms and requirements vary by program. For example, the federal government requires no state-level match of federal funding for the National Institute of Justice (NIJ) Research, Evaluation, and Development Project Grants.Footnote 3 Although cost sharing is not explicitly required, the NIJ states that they may be included to strengthen the application. By contrast, the Metropolitan Transportation Planning and State and Non-Metropolitan Planning and Research Program is a formula grant that entails mandatory matching requirements, typically set at 80 percent federal and 20 percent state funding, though states or other participating institutions may contribute additional funds if they choose.Footnote 4 Finally, the National Estuarine Research Reserve System is a federally funded, state-managed network of 30 estuarine reserves across the United States (General Services Administration 2025c). Under this program, the federal share cannot exceed 70 percent of the total project cost; states must support at least 30 percent of the cost and may opt to provide a larger share.
These examples illustrate the diverse nature of research performed through federal-state partnerships and the heterogeneity in their underlying funding arrangements. Although some programs entail minimum cost-sharing requirements, state governments possess considerable latitude to determine their level of financial commitment in federal R&D grant programs. Given this flexibility, it is plausible that state governments will commit additional state funds to R&D programs that are perceived to generate significant local benefits. Likewise, states may strategically withhold voluntary funding from R&D programs that are perceived as unproductive.
Fiscal federalism
The United States is a federal system in which the national, state, and local governments jointly supply public goods. Oates (Reference Oates1972) argues that the federal system of government combines the strengths of unitary and fully decentralized governments. A centralized approach to public finance brings about stability and equitable development, while a decentralized model encourages responsiveness and innovation in the public sector.
A distinguishing feature of fiscal federalism is the allocation of public goods at various levels of government. Olson (Reference Olson1969, p. 483) introduces the principle of “fiscal equivalence,” wherein economic efficiency is achieved by ensuring “a match between those who receive the benefits of a collective good and those who pay for it.” Mikesell (Reference Mikesell2017, p. 644) highlights the correspondence principle, which similarly suggests that “the critical factor in identifying the level of government that should provide (but not necessarily produce) a public service is the range of benefit spillover.” Both principles suggest a hierarchy for the provision of public goods and services: those that yield benefits with a limited geographic extent, such as local parks and fire protection, are most suitable for provision by local or state governments, while pure public goods, such as national defense, should be provided by the central government. The knowledge and innovation resulting from R&D, as noted by Stiglitz (Reference Stiglitz, Kaul, Grunberg and Stern1999), also share the characteristics of pure public goods, even in the presence of patents.Footnote 5 This implies that national governments should assume full responsibility for R&D investment.
Mikesell (Reference Mikesell2017) also discusses the subsidiarity principle, which emphasizes the devolution of certain functions to lower levels of government.Footnote 6 The rationale behind this principle is that subnational governments can often better identify and match public spending to the preferences of their residents than the national government. Extending this principle to R&D, a state government may be better positioned to identify and support research projects that generate significant but geographically concentrated benefits. Further, subnational governments often have an interest in promoting research and innovation as an economic development tool (Colombo and Martinez-Vazquez Reference Colombo and Martinez-Vazquez2020). This suggests that some state-level involvement in R&D investment could be economically efficient, especially if state spending is augmented by federal grants.
The influence of federal grants on state and local spending choices is a major topic of inquiry within the field of fiscal federalism. Most of these studies test whether federal grants encourage recipient governments to spend more of (i.e., “crowd in”) their own-source revenue on the subsidized function or reallocate it elsewhere (“crowd out”). Contrary to economic theory, which suggests that federal grants should fully crowd out the recipient government’s spending, a large body of empirical work finds that federal grants encourage additional state spending in several functional areas (Hines and Thaler Reference Hines and Thaler1995; Inman Reference Inman2008).Footnote 7 However, consistent with theory, other papers present compelling empirical evidence that federal grants crowd out in state spending in specific functional areas, including welfare assistance (Craig and Inman Reference Craig, Inman and Rosen1986), highway aid (Gamkhar Reference Gamkhar2003; Knight Reference Knight2002), and educational funding (Fisher and Papke Reference Fisher and Papke2000). The scholarly consensus is that the direction and magnitude of the effect of federal grants on states is contingent upon the type of grant, the functional or policy area in question, and other contextual factors.
In the domain of R&D, prior research suggests that federal grants crowd out state spending. Wu (Reference Wu2009) tests whether changes in federal R&D funding under the NSF’s Experimental Program to Stimulate Competitive Research (EPSCoR) influenced state R&D spending between 1979 and 2006. Using a fixed effects model, he finds that a marginal dollar of federal EPSCoR grants reduce state support for academic R&D by $0.50 per capita. Wu and Merriman (Reference Wu and Merriman2017) also assess the impact of federal support for R&D at public universities between 1985 and 2012. Their main result indicates that a one-dollar increase in federal support for academic R&D resulted in a 15-cent decline in state funding for public universities. Additional analyses suggest this substitution effect is largest during periods of slow federal funding growth.
In summary, the extant literature suggests that federal funding crowds out state R&D expenditure. We argue that this prior work fails to consider the distinct nature of R&D performed by academic institutions and state agencies. In the following subsection, we develop a conceptual framework that suggests federal R&D grants to state agencies may induce additional state research expenditures to those entities even as they crowd out state dollars at colleges and universities. We then find empirical support for this conceptual framework through a qualitative analysis of federal R&D grant awards to states and universities, as well as the hypotheses it generates through a quantitative analysis of state R&D expenditures.
Conceptual framework
We postulate that state agencies and academic institutions pursue distinct streams of R&D that reflect the goals and incentives of both types of organizations. State agency R&D is focused on process improvements that are directly beneficial for the performing agency and its stakeholders. As a result, state agency R&D is usually applied in nature. Though there are a wide variety of R&D projects and sponsors, common domains include public health (health departments), civil engineering (transportation departments), community development and planning (economic development agencies), and water or land management (environmental agencies). In some instances, this research may have broader applicability outside of the sponsoring state but is unlikely to lead to patents as state agencies rarely possess any means to pursue them.
By contrast, state-supported universities and colleges engage in R&D with the primary goal of advancing institutional prestige through broad knowledge creation. Academic institutions also leverage R&D to generate patents which in turn deliver royalty income that advances their institutional mission (Coupé Reference Coupé2003). Given these incentives, universities prefer to specialize in biomedical, chemical, computing, and engineering research. Knowledge generated in these fields often yields broad positive externalities, and only a portion of their total societal benefit is realized at the state level.
State legislatures are responsible for determining state appropriations for academic and non-academic R&D. Legislators are assumed to be rational actors that are motivated to deliver benefits to their constituents to improve their prospects for re-election (Shepsle and Weingast, Reference Shepsle and Weingast1995). State legislatures should therefore appropriate funds for R&D, whether performed by state agencies or external entities, such that the marginal dollar delivers the greatest net benefit to state residents. While this framework predicts that states will dedicate own-source revenue toward R&D, the response to federal R&D grants will vary by performer. Federal grants for state agency R&D should stimulate additional direct state spending owing to the localized nature of its benefits. By contrast, federal grants are expected to displace state expenditure on academic R&D, as states (and their residents) cannot easily capture the economic benefit of that spending.Footnote 8
Qualitative analysis of federal R&D awards
We assess the validity of the key assumptions of our conceptual framework by applying computational methods to textual data in recent federal R&D awards to state agencies and academic institutions. This exercise aims to identify common themes that are contained within the text which will enable us to gain a better understanding of the nature of R&D performed by each type of institution. Computational text analysis has grown in popularity in recent years as it is efficient means to analyze abundant textual data in social sciences research (Anastasopoulos et al. Reference Anastasopoulos, Moldogaziev and Scott2020; Baden et al. Reference Baden, Pipal, Schoonvelde and van der Velden2022; van Atteveld et al. Reference Van Atteveldt, Welbers and Van Der Velden2019).
Data and methods
We obtain data on all federal awards for R&D from FY 2017 to FY 2019 reported in the Award Data Archive on USASpending.gov.Footnote 9 The dataset contains information on all grants awarded by the federal government each fiscal year, including their title, type, the recipient organization, and amount of federal funds supplied as well as total funding from non-federal sources.Footnote 10 Although the dataset does not identify the subject area of R&D grant, we are able to infer the subject area from the title of the award, which also serves as our unit of analysis (also referred to as the “document”). Grimmer et al. (Reference Grimmer, Roberts and Stewart2022) note that the unit of analysis for textual analysis must be both manageable computationally and relevant to the research question. Grant titles fulfill both conditions because they are available in digitized format, formed in reasonable length of words, and contain information that identifies the general topic and scope of research project undertaken by grant recipients.
We completed several standard preprocessing steps prior to analyzing the federal award dataset which are summarized in Appendix A. We then apply two standard analytical techniques to the preprocessed dataset: term frequency and topic modeling. Term frequency simply reports how many times a specific term appears throughout the text. While this method is useful for identifying prominent words that appear in grant titles separately awarded to state agencies and academic institutions, it does not offer any context as to the meaning of those words. For example, the word “program” may refer to research on the effectiveness of a public program or the development of a new computer program. Likewise, the word “development” has myriad meanings in grant titles. Topic modeling accounts for this shortcoming by examining a collection of common terms within and across grant titles.
We perform this analysis using Latent Dirichlet Allocation (LDA). LDA is an unsupervised topic modeling technique that identifies common themes within qualitative data (Blei et al. Reference Blei, Ng and Jordan2003). During the training phase, LDA examines R&D grants titles, seeking out words or phrases that commonly appear within and across grant titles. The algorithm then generates a set of topics and associated keywords. LDA does not define the nature of the topic; rather, it is the task of the analyst to infer the topic based on the highlighted keywords.Footnote 11
Results
We begin by examining the frequency of terms that appear in federal R&D awards. Figure 1a and 1b show the terms that appear most frequently in federal R&D awards to academic institutions and state agencies, respectively. There is a great deal of overlap between these entities, with root terms such as “research,” “program,” and “develop” appearing among the ten most common terms, though their specific rankings vary to a modest degree. We also observe notable distinctions, which become more pronounced further down the list. For example, the terms “biomed,” “cancer,” “mathemat,” “neurosci,” and “comput,” appear frequently in academic R&D awards but are relatively uncommon in state agency awards. By contrast, the terms “state,” “plan,” “control,” “community,” “highway,” and “metropolitan” appear frequently in state agency R&D grants.
(a) Most common terms appearing in FY 2017–2019 Federal R&D grant titles to academic institutions. (b) Most common terms appearing in FY 2017–2019 Federal R&D grant titles to state agencies.
Notes: Both plots reflect the most common terms that appear in federal R&D grant titles as reported on USASpending.gov. Term counts are scaled logarithmically, where each unit increase corresponds to a tenfold increase in magnitude.
As noted previously, word frequency alone provides an incomplete picture as to the nature of the R&D performed by the two types of recipients. Topic modeling provides stronger descriptive evidence about the research performed by both entities. Table 1 contains the results of the LDA analysis for both academic institutions and state agencies.Footnote 12 Panel A suggests that academic R&D is dominated by medical research; seven of the ten most prominent topics contain at least one term related to human health or medicine (e.g., “human,” “diabet,” “biomed,” “mental,” “neurosci,” “allergi,” and “cancer”). The only three exceptions – Topics 3, 6, and 9 – are closely related to geoscience, agriculture, and mathematics.
LDA topic analysis of FY2017–2019 Federal R&D grants to academic institutions and state agencies
Notes: Results above reflect the ten most prominent topics identified in federal R&D grant titles awarded to academic institutions (Panel A) and state government agencies (Panel B) in FY2017–2019 using the LDA package for R. While there is no strict standard that prescribes how many words should be presented in each topic, we chose 5 words per topic for academic institutions and 6 words per topic for state agencies based on the median length of words across grants titles in our dataset. See Appendix Figures A3 and A4 to view the distribution of term length in the titles of federal grants awarded to academic institutions and state agencies, respectively.
By contrast, panel B suggests state agencies engage in federally supported R&D with a stronger local orientation. Topics 2 and 4 both contain terms directly associated with transportation. Topic 1 refers to studies related to specific regulations (i.e., “acts”) that may have local implications. Topic 3 focuses on food and drug administration, which is closely tied to public health. Topics 5, 6, and 8 include terms related to improvement, protection, prevention, and control. Topics 7 and 9, respectively, include terms related to environmental resources and to agricultural and coastal management. In contrast with the topic analysis of federal grants to academic institutions in Panel A of Table 1, terms related to medical research appear only sporadically in two of the nine topics in Panel B.
The computational text analysis of federal R&D awards supports the assumptions of our conceptual framework. The results show that academic institutions specialize in R&D with diffuse social benefits, particularly medical research. By contrast, state agencies primarily engage in funded research related to regulatory policy, transportation, or environmental management that closely align with their agency missions. While this R&D likely generates benefit to state residents, it is unlikely to generate the same widespread benefit as the research pursued by academic institutions.
Quantitative analysis of state R&D spending
We next consider how federal funding influences state-level allocations for R&D across each type of institution through regression analysis of grant and expenditure data. Given the localized nature of state agency R&D, we anticipate that additional federal grant dollars will crowd-in state funding. By contrast, and consistent with prior research, we expect that federal funding will crowd out state support for academic R&D. This section describes the data and methods used to test this hypothesis, as well as our empirical results.
Data
The NSF’s Survey of State Government Research and Development is the principal data source for our analysis of state R&D expenditure. The survey sample includes all state agencies and dependent entities for the 50 states and the District of Columbia. It collects data on state expenditures on R&D completed by state agencies, academic institutions, and other external entities such as private firms based on uniform definitions.Footnote 13 The NSF reports the dollar value of state and federal spending on R&D performed by state agencies and state expenditures on academic institutions.Footnote 14 We separately obtain federal spending on R&D performed by colleges and universities at the state level from the NSF’s Survey of Federal Funds for Research and Development. One noteworthy limitation of the NSF’s state academic R&D expenditure measure is that it does not separately delineate the source of funding. That implies that some portion state expenditure on academic R&D is supported by federal pass-through funds. Using the available data, we estimate that about 17.4 percent of state academic R&D expenditure was backed by federal funds in fiscal year 2021.Footnote 15 However, the NSF reports whether state agency R&D was supported by the federal government or state directly, so this is not a limitation of the state agency R&D expenditure measures.
We retrieved and compiled the NSF survey data into a 50-state panel spanning fiscal year (FY) 2006 to 2021, which reflects the earliest year the survey was conducted by the NSF.Footnote 16 We report the average academic and nonacademic real spending per resident at the state level over 2006 to 2021 in Appendix Figures B1 and B2, respectively. Ohio was the median state for academic R&D spending over this period, with about $2.43 per resident over this period. Montana was the median state for state agency R&D expenditure over this period, allocating about $0.50 per resident.
We compile several variables that reflect each state’s economic and fiscal capacity to support R&D from 2006 to 2021. This includes state Gross Domestic Product (GDP), personal income per capita, and state population, which were each obtained from the U.S. Bureau of Economic Analysis. State unemployment rates were collected from the U.S. Bureau of Labor Statistics. State tax revenue, intergovernmental revenue, and total expenditure were compiled from the U.S. Census Bureau’s Annual Survey of State Government Finances. We estimate the annual share of the state population age 25 and older that holds a college degree using the Integrated Public Use Microdata Series (IPUMS) of the Current Population Survey. Finally, we compute each state’s total annual expenditure on colleges and universities per enrolled student as reported by the National Center for Education Statistics.Footnote 17
Next, we assemble several variables that reflect the political ideology and partisan political leanings of state governments. The first of these is a binary variable equal to one in each year that the state’s governor is a member of the Democratic Party.Footnote 18 We also compute the percentage of seats in both the upper and lower houses of each state legislature held by members of the Democratic Party.Footnote 19 The data for all three variables was prepared by the Council of State Governments (2025) and aggregated over the 2006 to 2021 period by the University of Kentucky Center for Poverty Research (2025). Lastly, we incorporate two additional measures of each state’s upper and lower house ideology based on the median legislator in each chamber as prepared by Shor and McCarty (Reference Shor and McCarty2011) and updated through 2021 by Shor (Reference Shor2025).Footnote 20
We also compile several additional measures related to state R&D capacity and determinants of federal R&D funding based on previous analyses by Wu (Reference Wu2013) and Wu and Merriman (Reference Wu and Merriman2017) that demonstrate their relevance. First, we obtain data on science and engineering doctorates awarded in each state by year from the NSF’s Survey of Earned Doctorates. Second, we measure annual federal spending on R&D activities conducted by federal entities, such as FFRDCs within each state’s borders from the NSF’s Survey of Federal Funds for Research and Development. Third, we determine the share of total seats held by each state on the majority party of the Congressional House and Senate Appropriation Committees using data compiled by Stewart and Yoon (Reference Stewart and Yoon2017).
Finally, we prepare two additional variables that reflect each state’s underlying administrative capacity to draw federal grants. Administrative capacity broadly refers to the institutions, resources, and personnel that enable governments to successfully implement and execute public policy. Prior research by Howell and Magazinnik (Reference Howell and Magazinnik2020), Aldag (Reference Aldag2025), and others suggest the importance of subnational administrative capacity in the application and receipt of federal grants. We follow the approach of Moffitt et al. (Reference Moffitt, Willse, Smith and Cohen2023) to operationalize each state’s underlying administrative capacity in our quantitative analysis. Specifically, we draw on data from the U.S. Census Bureau’s Annual Survey of Public Employment & Payroll (ASPEP) to compute the total number of administrative workers employed by the state governments in each year. Our first measure of state government capacity considers the number of full-time equivalent (FTE) workers employed in financial administration or other government administration roles. The second measure captures the number of FTE state higher education workers employed in non-instructional roles. Although imperfect, these variables represent the best available proxy measures of state administrative capacity that are available on a state-by-year basis.
The complete summary statistics for the quantitative dataset are reported in Appendix Table B2. We next describe the empirical strategy for analyzing these data.
Methods
Our baseline analysis makes use of a fixed effects regression model to evaluate the impact of federal R&D grants on state allocations for R&D. The principal advantage of this model is that the state fixed effect absorbs all time invariant factors that influence R&D spending at the state level. Inclusion of year fixed effects similarly absorbs any national year-specific shock to state spending patterns. We also separately control for other time-varying factors, such as state fiscal or research capacity that may influence state R&D spending. The regression model takes the following functional form:
$${y_{st}} = \alpha + \beta {F_{st}} + X_{st}^{\rm{'}}\gamma + {\eta _s} + {\lambda _t} + {u_{st}}$$
where
${y_{st}}$
is log-transformed R&D spending in state s and fiscal year t;
${F_{st}}$
is log-transformed federal support for state R&D activities;
${X_{it}}$
is a vector of economic, fiscal, and political covariates specific to state s in fiscal year t;
${\eta _s}$
and
${\lambda _t}$
are state and year fixed effects; and
${u_{st}}$
is the idiosyncratic error term.Footnote
21
The model coefficient
$\beta $
reports the impact of a one percent increase in federal support for state R&D on state R&D spending.
We separately estimate the model above with two dependent variables. First, state spending on R&D conducted at academic institutions (i.e., colleges and universities) within the state. Second, state spending on R&D performed by agencies of the state government. The
${F_{st}}$
term corresponds to federal support for each type of R&D activity in each of the two analyses. Both regression models are estimated with standard errors clustered at the state level to account for potential serial correlation in the dependent variable (Cameron and Miller, Reference Cameron and Miller2015).
We also make use of an instrumental variables (IV) model to account for potential endogeneity between state and federal spending decisions. Knight (Reference Knight2002) argues that studies that consider the impact of federal grants on state spending decisions may suffer from omitted variable bias arising from political variables, such as political representation on Congressional appropriations committees. Wu (Reference Wu2013) finds that state scientific research capacity is the strongest predictor of federal support for R&D, as well as some evidence that appropriations committee representation impacts federal allocations.
Reflecting prior research, we select instruments that are correlated with federal R&D funding but are uncorrelated with state R&D spending decisions in both models. In the state agency model, we select (1) log-transformed federal funding for R&D conducted by the federal government in each state, (2) the state’s share of the Congressional House Appropriations Committee, (3) the total number of doctorates awarded in science and engineering fields, and (4) the number of state workers employed in financial or other administrative positions as instruments for log-transformed federal funding for state agency R&D activities. The number of science and engineering graduates is likely endogenous to federal academic R&D funding given that many graduate student researchers are supported by grant funding. We therefore do not incorporate it into the analysis of academic R&D spending. Instead, we select an alternative set of instruments, which include (1) log-transformed federal funding for R&D conducted by the federal government in each state, (2) the state’s share of the majority party that controls the Congressional House Appropriations Committee, and (3) the number of public college and university workers employed in non-instructional positions. Ordinary Least Squares (OLS) regressions confirm that each instrument is statistically associated with federal R&D grants across the two models, which suggests they are valid instruments in the first stage of an IV two-stage least squares (2SLS) model.
Results
Table 2 reports the estimation results for state agency R&D spending. The results in column 1 correspond to a parsimonious model specification that includes only state fiscal and economic controls. State fixed effects are added to the model in column 2, as are year fixed effects in column 3. Our preferred state-by-year fixed effects specification in column 3 indicates that a 1 percent increase in federal R&D grants to state agencies results in a 0.35 percent increase in state support for R&D at state agencies. This estimate is statistically significant at the 99 percent confidence level, as are the other OLS estimates of similar magnitude in columns 1–2.
Federal support on state agency R&D spending estimation results
Notes: Statistical significance is indicated by ***(p < 0.01), **(p < 0.05), *(p < 0.1). Robust standard errors are clustered at the state level.
The IV 2SLS estimation results are also reported in columns 4–6 of Table 2. The baseline IV model in column 4 indicates that a 1 percent increase in federal R&D funding results in a 0.14 percent increase in state support for R&D at state agencies, though this estimate is not statistically significant. The inclusion of state and year fixed effects in the IV model increases the estimated effect to 0.39 percent, which is marginally significant at the 90 percent confidence level. Overall, the estimation results across the three IV specifications match those of the OLS estimator and are generally consistent in magnitude though less precisely estimated. The results suggest that federal R&D grants to state government agencies crowd in state-level funding.
We next turn to the analysis of state allocations for R&D at academic institutions in Table 3. The OLS results in columns 1–3 indicate that additional federal support for academic R&D has essentially no effect on state allocations for academic institutions. All three estimates are small in magnitude and statistically insignificant. The IV model results in columns 4 and 5 similarly suggest that federal R&D funding has no significant or practical effect on state-level allocations. Though the magnitude of the estimated effect in the state-by-year IV specification reported in column 6 is larger than the others, it remains statistically indistinguishable from zero (t = −0.1; p = 0.92). Together, the results in Table 3 suggest that federal grants for academic R&D had no discernable impact on state R&D allocations between 2006 and 2021. We conclude by briefly summarizing our findings and discussing the implications of these results.
Federal support on state academic R&D spending estimation results
Notes: Statistical significance is indicated by ***(p < 0.01), **(p < 0.05), *(p < 0.1). Robust standard errors are clustered at the state level.
Discussion and conclusion
The goal of this paper is to evaluate whether federal grants influence state allocations for R&D at state government agencies and academic institutions. We propose that state agencies specialize in R&D that generates localized benefits for states and their residents, while academic institutions pursue federal grants for research that predominately results in broad positive externalities. We find support for this claim through topic modeling of federal R&D awards to state governments and colleges and universities. This helps to inform a conceptual framework that assumes state policymakers rationally allocate state public funds towards R&D that generates the greatest local benefit for their constituents. This framework generates two testable hypotheses regarding the impact of federal grants on state R&D expenditures in line with the prior literature on fiscal federalism and intergovernmental grants. First, federal R&D grants awarded to academic institutions reduce own-state spending on academic R&D. Second, federal grants for R&D performed by state government agencies stimulate additional state expenditure.
Our analysis of federal and state R&D expenditure data from 2006 to 2021 partially supports the first hypothesis and strongly supports the second. Our preferred specification suggests that a one dollar increase in federal grants for state agency R&D results in a $0.35 increase in state support. By contrast, we find no economically or statistically significant changes in state R&D spending on academic institutions in responses to federal grants over this period. However, a modest share of state expenditures on academic R&D reported by the NSF appear to be supported by federal pass-through dollars. This makes it challenging to make inferences about the true level of state investment in academic R&D in response to federal support.
Overall, our results suggest that state spending responses to federal R&D grants is influenced by the ability of state governments to directly capture the intangible benefit that results from investment in R&D. This finding may help to inform the broader fiscal federalism literature, which finds that federal grants crowd in state and local spending in some contexts, but spur crowding out in other circumstances.
This study also seeks to shed light on an often-unnoticed source of knowledge production in the U.S. economy: state governments. Though total state R&D spending is dwarfed by that of the federal government, it grew at a much faster rate over the 2006 to 2021 period that is the focus of our study. State R&D spending continued to grow through 2023, the most recent year that NSF data are available. Notably, state R&D expenditure rose to just over $3 billion in FY23, an 11 percent annualized nominal increase from $2.5 billion in FY21. The NSF only attributes $130 million of the $563 million increase in state R&D expenditures to growth in federal support. This raises an interesting question of how states will respond to the federal government’s newfound interest in fiscal retrenchment and disinvestment in science and R&D.
The results of this study suggest that the federal government could expand total public investment in R&D by strategically allocating more federal research funding to state agencies. However, we do not argue that state agency R&D is inherently more valuable than academic R&D. Academic institutions are best equipped to undertake research in biomedicine, artificial intelligence (AI), and other advanced technologies that help ensure U.S. economic competitiveness in the coming decades. As such, university-based research continues to merit strong federal support, even if states show a lower propensity to contribute their own fiscal resources to those projects. Given the distinct nature of academic and non-academic governmental R&D, the opportunity costs of any change in federal funding priorities must be carefully considered.
By contrast, R&D performed by state agencies is most directly related to public welfare. For example, state public health agencies were assigned a leading role in managing the impacts of the COVID-19 pandemic. Greater investment in R&D at state public health agencies could yield improvements in vaccine delivery or quarantine strategy that carries the potential to mitigate future pandemics, as well as the resulting economic disruptions. Future inquiry on this topic should seek to quantify and measure the cumulative research impact of state government agencies and identify their comparative advantage in the U.S. R&D pipeline.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0143814X25100913.
Data availability statement
Replication materials are available in the Journal of Public Policy Dataverse at https://doi.org/10.7910/DVN/HHA3QM.
Acknowledgements
The authors thank Josh Ryan and three anonymous reviewers for thoughtful comments and suggestions, as well as seminar participants at the Indiana University O’Neill School and attendees of the Association for Budgeting and Financial Management and Association for Public Policy Analysis and Management annual conferences. The authors have no conflicts of interest to declare.
Open access
