Both — Defense R&D flows disproportionately to places already ahead, but rigorous studies find it also causes further private innovation on top of that head start

Summary

Both are true, and the literature has actually separated the two effects. Places that get the most federal defense R&D money were, on average, already ahead in innovation capacity — the Pentagon and its agencies (DARPA, service labs) place research money where the scientists, engineers, universities and existing defense-industrial base already are, which is why California, Massachusetts, Washington and Texas dominate both defense R&D and patenting today. But peer-reviewed studies using quasi-experimental designs that isolate funding changes driven by national defense cycles (not by a place's own economic momentum) find defense R&D also causes additional private-sector R&D and patenting on top of that pre-existing base. The best state-level US estimate: a 1% rise in military R&D obligations to a state raises private R&D there by roughly 0.11-0.14% over 4-5 years (Pallante, Russo & Roventini 2023, Research Policy), with parallel gains in high-tech engineering employment. An industry-level OECD study with French firm microdata finds a larger crowd-in effect (Moretti, Steinwender & Van Reenen, Review of Economics and Statistics 2023): a 10% rise in government-financed R&D generates roughly 5-6% additional privately funded R&D, plus productivity gains and cross-border spillovers. So the honest answer is not "which one" — it is that selection explains most of the cross-sectional pattern you see today, and a real, smaller, causal effect explains part of the year-to-year change.

What the corpus could and could not answer directly

AskAmerica's research schema (NSF NCSES Federal Funds for R&D) carries federal R&D obligations by funding agency (so Department of Defense can be isolated — see below) but with no state/geography breakdown at all, and only 2 fiscal years loaded (FY2022, FY2024). fiscal.usaspending_by_state carries state-level federal obligations across 2010-2025 but with no awarding-agency dimension — defense spending cannot be isolated from the rest of federal spending in that table. The one table that could in principle link state geography to federal R&D dollars, research.research_herd_rd_by_state_year, is built on research.nsf_herd_by_institution, which failed to query at all during this session (HTTP 404 on the underlying parquet metadata file) and in any case only carries a single year (2024) of institution-level data — not enough for a before/after or panel test. There is no table in this corpus today that lets a caller build a genuine multi-year state-by-defense-R&D panel, which is exactly the design both key papers below use. This gap has been logged (see Data gaps below); this answer therefore leans on the published literature's own causal designs rather than attempting a weaker proxy regression in their place.

What the corpus could confirm directly: Department of Defense obligations were $56.1B of the $69.9B (80%) in total federal R&D obligated to business performers in FY2024 (NSF NCSES Table 7) — defense is overwhelmingly the channel through which federal R&D dollars reach private firms, which is why the "defense R&D and innovation" question is really a question about DoD specifically. And patent concentration today is extreme: California alone produced 45,442 of the roughly 100,000+ utility patents granted nationally in 2024 (USPTO PatentsView, assignee location) — about 4x the next-ranked state (Texas, 11,154) — consistent with the same handful of states leading in both defense contracting and patenting long before any current funding cycle, i.e., consistent with the "already ahead" side of the question.

The causal evidence: two designs built to separate selection from effect

Moretti, Steinwender & Van Reenen (2023), "The Intellectual Spoils of War? Defense R&D, Productivity, and International Spillovers," Review of Economics and Statistics. Industry-by-country panel across OECD countries plus French firm-level microdata. The endogeneity problem is named explicitly: a country/region that is already innovative may simply attract more government R&D, or a booming local economy may draw in both. Their instrument is predicted defense R&D built from national defense-budget swings (driven by geopolitical events, arguably exogenous to any one industry's or region's own productivity trend) interacted with each industry's historical exposure to defense contracting. Result: a 10% rise in government-financed R&D produces roughly 5-6% additional privately financed R&D, positive TFP effects, and cross-border/cross-industry spillovers — the crowd-in effect survives the IV correction, though the OECD/international-industry grain is coarser than "places" in the sense of US states or metros.

Pallante, Russo & Roventini (2023), "Does public R&D funding crowd-in private R&D investment? Evidence from military R&D expenditures for US states," Research Policy. This is the closest match to the literal question — a genuine US-state panel, 1968-2017. Same identification logic at finer grain: national military R&D obligations swing with geopolitical events, not with any one state's business cycle, and different states have structurally different sensitivity to those national swings (some states' defense-industrial base scales up and down faster with the national budget than others') — that differential sensitivity is the instrument. Result: a 1% rise in military R&D obligations to a state raises private R&D in that state by 0.11-0.14% over a 4-5 year horizon, with parallel positive effects on high-tech engineering employment — smaller than the OECD industry estimate, plausible given a narrower and more local unit of analysis, but still a real, positive, causally-identified effect net of the state's own prior trajectory.

Complementary national-level evidence: the Dallas Fed's "Returns to Government R&D: Evidence from U.S. Appropriations Shocks" (Fieldhouse et al.) studies congressional appropriations shocks to DoD, DOE, NASA, NIH and NSF R&D and finds nondefense R&D appropriation shocks produce a delayed, gradual, and eventually highly significant rise in business-sector TFP over 8-15 year horizons, with structural estimates implying returns of 140-210% and government R&D accounting for roughly one-fifth of postwar business-sector TFP growth. This is national, not place-based, but it corroborates that appropriation-driven (as opposed to selection-driven) R&D shocks have real long-run productivity payoffs, reinforcing that the effect found in the state-level defense study above is not a fluke of that one design.

Reconciling the two halves of the question

Selection and causation are not actually in tension here once separated by design: a simple correlation between "defense R&D received" and "current innovation output" mixes both effects together, and on its own cannot tell you which dominates — which is exactly why every rigorous study above throws out the naive correlation and instruments for the funding change instead. Doing so consistently finds a positive, but structurally smaller-than-naive, causal effect, meaning: (1) a meaningful share of the raw cross-sectional pattern (why California and Massachusetts look far more innovative than Mississippi or West Virginia) is indeed explained by where the scientists, universities, and existing prime contractors already were, not by the R&D dollars themselves; and (2) on top of that base, the R&D dollars still do real, measurable work — crowding in roughly a tenth to half again as much private R&D per unit of public defense R&D, depending on grain, with a multi-year lag before the effect shows up.

Every query behind this report

2 warehouse calls ran in this session, in order. Each is reproducible against the same snapshot.

SELECT pl.state_fips, COUNT(DISTINCT pg.patent_id) AS patents_2024
FROM patents.patent_grants pg
JOIN patents.patent_assignees pa ON pg.patent_id = pa.patent_id
JOIN patents.patent_locations pl ON pa.location_id = pl.location_id
WHERE pg.grant_year = 2024 AND pl.country_code = 'US'
GROUP BY pl.state_fips
ORDER BY patents_2024 DESC
LIMIT 10

SELECT "year", funding_agency, obligations_usd_million
FROM research.nsf_federal_rd_obligations
WHERE performing_sector = 'Businesses' AND (funding_agency ILIKE '%Defense%' OR funding_agency = 'All agencies')
ORDER BY "year" DESC, obligations_usd_million DESC
FETCH FIRST 20 ROWS ONLY