The impact of public R&D investments on patenting activity: technology transfer at the U.S. Environmental Protection Agency

ABSTRACT

This paper presents estimates of the impact of public R&D on patenting activity at the U.S. Environmental Protection Agency (EPA). Using a time series of public sector agency data, we estimate the per-capita R&D elasticity of new patent applications using a knowledge production function framework model that is an expanded version of what other scholars have used with private sector data. New patent applications are an important step in the technology transfer activities of a federal agency. We estimate this elasticity to be about 2.0. This elasticity value represents an initial estimate of the impact of EPA’s R&D investments on its technology transfer activity.

KEYWORDS:

• R&D
• patents
• technology transfer
• knowledge production function
• Environmental Protection Agency

JEL CODES:

• O33
• O38
• O44
• C51

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 As stated in the FTTA, a CRADA is

“any agreement between one or more Federal laboratories and one or more non-Federal parties under which the Government, through its laboratories, provides personnel, services, facilities, equipment, or other resources with or without reimbursement (but not funds to non-Federal parties) and the non-Federal parties provide funds, personnel, services, facilities, equipment, or other resources toward the conduct of specified research or development efforts which are consistent with the missions of the laboratory … ”

2 A public effort to assess the economic contributions of federal R&D is certainly not new. See, for example, the National Academies (2011).

3 See, https://www.nist.gov/tpo/return-investment-roi-initiative/unleashing-american-innovation-symposium.

4 The relevant literature on CRADAs is summarized in Chen, Link, and Oliver (2018).

5 The technical name of this Act is the Clean Air Amendments of 1970.

6 The technical name of this Act is the Federal Water Pollution Control Act Amendments of 1972.

7 Other legislations include the Environmental Quality Improvement Act (1970), the Ocean Dumping Act (1972), the Coastal Zone Management Act (1972), the Marine Protection Research and Sanctuaries Act (1972), the Endangered Species Act (1973), the Marine Mammal Protection Act (1972), the Deepwater Ports and Waterways Safety Act (1974), the Fish and Wildlife Coordination Act (1974), the Wild and Scenic Rivers Act (1976), the Water Resources Planning Act (1977), the Water Resources Research Act (1977), and the Environmental Education Act of 1990 (Wisman 1985).

8 The source for the Technology Partnership Office reports is each agency’s report, all of which are now, under the Technology Transfer Commercialization Act of 2000 (Public Law 106-404), prepared annually.

9 See, https://www.nist.gov/tpo/federal-laboratory-interagency-technology-transfer-summary-reports

10 An invention disclosure is generally an agreement between the federal agency or one of its laboratories and an employed scientist or researcher regarding the ownership of the invention. Generally, a scientist or a researcher completes an invention disclosure through the agency’s or laboratory’s technology transfer office as a first step for the office to consider patenting the invention. See Bradley, Hayter, and Link (2013) on disclosures at universities.

11 Of course, not all R&D-based knowledge manifests itself in new patent applications; such new knowledge can manifest itself in terms of publications or even, in the case of small entrepreneurial firms, in terms of internal secrets (Hayter and Link 2018).

12 More generally, the production function for patent applications may be unknown, so that estimating equations such as (3) and (4) cannot be derived. In this case, it may be of interest to estimate a semi- or non-parametric regression model for patent applications. This would likely require a larger sample size. In this paper we therefore maintain the conventional parametric framework.

13 While we use a time series of patent application data for a single unit (agency), much of the empirical literature on patenting activity is based on analyses of panel data sets. With a panel, one could use the Poisson fixed or random effects estimator of Hausman, Hall, and Griliches (1984). Cincera (1997) and Blundell, Griffith, and Windmeijer (2002) discuss panel data estimators that relax some of the parametric assumptions underlying the Poisson and negative binomial models. More recently, Charlot, Crescenzi, and Musolesi (2015) have proposed a semi-parametric generalized additive model to analyze regional knowledge production functions in Europe.

14 This test is equivalent to a t-test in a regression of the residuals on lagged residuals and covariates. See, for example, Godfrey (1988) and White (1992).

15 Our elasticity estimate is about four times of that presented by Czarnitzki, Kraft, and Thorwarth (2009, 142), using a model similar to that in equation (2), but without disclosures (D), for a sample of private sector Flemish firms.