Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
AS SES S ING THE IMPACT OF FEDERAL INDUSTRIAL RES EARCH AND DEVELOPMENT COP END ITURE ON PRIVATE RESEARCH AND DENIES OPMENT ACTIVITY Frank R. Lichtenberg* Graduate School of Business Co lamb ia Univers ity and National Bureau of Economic Research A number of studies have attempted deco determine how federal expenditures for research and development (R&D) performed in industry have affected the quanti~cy of private inures vent in R&D . The s tudies have analyzed the statistical relationship be~cween federal and L company R&D spending at the five, industry, and aggregate levels. lathe possibility that priorate decisions to invest in R&D may be influenced by federal industrial R&D spending is regarded as worthy of careful investigations in part because numerous studies have demonstrated that priorate R&D expenditure has ~ significant, posi~ci~re imp ace on Coccal factor productivity (TEE). Although most analysts have failed to find evidence of a similar direct productivity impact of federal R&D expenditures, 2 federal R&D nevertheless may have a considerate, e indirect impact (positive or negatived on IFP if it influences private R&D investment decis ions . Most9 if not all, of the econometric studies of the effect of federal expenditures on private R&D resource allocation are "reduced-form" is nature. They a~ctempt to determine the direction and magnitude of the effect without providing an explicit theoretical basis for a link between the two. The purpose of this paper is to fondue ace, and to review or develop evidence concerning, several maj or hypotheses that could account for the existence of such a link. The implications of these hypotheses are investigated in the context of a model predicated upon the assumption that priorate sponsors of R&D allocate R&D resources optimally, that is, they employ R&D resources until the point at which the expected marginal re turns to R&:~) input equal ache marginal cost . This ~ equilibria " * Financial support from ache National Science Foundation (aria grant PRA 85-12979 deco ache National Bureau of Economic Research) and from the MacAr thus Foundation Faculty Research Program at Columbia Unifiers ity is gratefully acknowledged. Donald Siegel provided capable research assistance. The remaining errors are Chose of the author. · 113 -
FIGURE 1 Supply and Demand Model of Privately Financed R&D Output Quar.ti Be termination and Price De te Ruination $ P* S - - - - - - - - 1 - - - - ,,, !Q* Quantity of privately financed Red) output - L 1 Lo -
mode 1 o ~ priorate R&D resource allocation is sketched in ache next section. In that section, three main hypotheses that could account for the existence of a relationship between federal and private R&D expend) Cure are offered ~ The three hypotheses, referred deco as "crowding out, n "spill overs, ~ and "demand pu119 n are examined in detail in subsequent sections. The section devoted to the demand-pull hypothesis is the longest and the only one containing new econometric evidence. Lee emphasis on that h~rpo~chesis reflects ache belief that insufficient attention has been paid deco ache role of federal demand for goods and serrices in stimulating pr~ra~ce R&D investment. As a result, estimates of the relationship between federal and private R&D have been misincerpre~ced as reflecting solely the effects of federal R&I) on the supply (cost) of privately produced innovations . The econometric evidence presented sugges ts an alternative interpre~ca~cion. The final section provides a summary and ten~cati~re cone lus ions . THE MARKET FOR PRIVATE R&D OUTPUT Economists engaged in the study of technical change sometimes find it useful to represent research and development as a production acti~ri~cy, an activity in which the services of R&D inputs (for examples scientists and engineers, Rho plant) are employed deco produce improvements in products or processes, which in turn constitute the output of R&D3 The measurement of R&D output, of course, poses maj or difficulties ~ In the absence of dla~ca on ache value, in cons Manic dollars' of improvemen~cs in products and processes, chose interested in measuring R&D output are confined to the analysis of " indicators" o f R&D output, such as patent counts e While economists are fully cognizant of ache serious problems associated with measuring R&D outpu~c (and, hence, witch measuring productivity of R&D inputs ), ache concept is both well defined and useful enough to play a key role in this analysis. We representation of R&D as an activity devoted to the production of R&D output is useful because it enables one deco apply ~ appropriately, the author believes ~ a standard economic model of the determination of ache quantity and price of an indus~cry's output deco investigate ache effects of federal R&D on equilibrium priorate R&D output. In the standard textbook model, as depicted in Figure 1, the equilibrium quantity and price of an indus~cry's output is determined by the intersection of the industry supply and demand schedules. The supply schedule is derived from the marginal cost schedules of ache individual producers . The position ~ intercept) of ache supply schedule depends on both input prices and input productivity; increases in input prices shift the supply schedule up, and increases in input productivity have the opposite effect. Offend the slope of Ache supply schedule is hypothesized to depend on the time horizon, with supply assumed to be less elastic (steeper) in the short run than in the long run (in which case it may be perfectly elastic?. 1 15
The primary reason for a less than perfectly elastic supply of output in the short run is a Less than perfectly elastic supply (that is, a short ° run immobility) of inputs . This paper proposes Deco analyze the impact of contract R&D expenditures on private R&D activity (both output and input) within ache context of an equilibria model of industry output determination. It is assumed that private sponsors of ~ produce Rho output until the point at which the marginal cost of improvements in products and processes eq,'= Is the (-expected) marginal retunes . Lee demand 5 or marginal willingness to pay, for privately produced R&D output is assigned to be a nonincreasing function of output. The marginal cost, or supply, Of n irmova~cions " (R&D Otltpu~C), on the ocher hand, is assumed to be a strictly increasing function of output. The latter assumption may be justified on both theoretical and empirical grounds . First, an ~ intangible ~ input critical to ache production of industrial innovations - ~ the underlying scientific and technological "knowledge base"--is, arguably, "quasi-fixed, n meaning that it is more expens Eve to increase the stock of knowledge quickly than slowly. Thus, increases in the employment of conventional R&D inputs, such as scientists and laboratory equipment S are likely to be subj ect to diminishing .returrts . Second, evidence will be presented ehat the supply of conventional R&I) inputs is less than perfectly elastic in tone short run. Additional ~ indirect) evidence of a positively sloped supply schedule of R&D output is found in the work of Nadiri and his colleagues. Their econometric studies suggest that the level of R&D activity responds quite sluggishly deco its hypothesized de~cerm~nants; the low observed ffspeed of adj,'-ct:ment" of R&D spending to its long- run equilibrium level. is consistene with the hypothesis of sharply increasing short-n~n marginal costs of producing R&I) output. Shifts in the supply of and the demand for priorate R&D output result in changes in the equilibrium quantity and price of R&D output. The next three sections outline in detail echoes hypotheses chat imply that changes in federal R&D expenditure either cause shifts in the pri~ra~ce R&D supply schedule, or are correlated witch (but do not cause) shifts in the demand for privately produced improvements in products and processes. Hence, each of the hypotheses court account for Ache existence of a correlation between federal and prince R&D. The first two hypotheses, referred to as "crowding out A and "spi3~lo~rers," may be regarded as instances of Ache general phenomenon of n economies ~ or diseconomies ~ of index try growth,. a phenomenon perhaps first identified (or at least labeled) by the eminent 19~ch-century economist Alfred Harshall. Economies (diseconomies) of growth are said to occur when expansion in the total output of an industry per se results in reductions (increases) in each producer's unit cost of production According deco the crowding-out hypothesis, increases in federal R&D expenditure impose "external" (pecuniary) diseconomies on private R&D sponsors. Spillo~rers, on ache other hand, may be interpreted as "internal" ~ technological) economies in the priorate production of R&D Output generated by increased federal R&D expenditure. - 115
I e the supply-and-demand model of private R&D output de~er=~nation is useful for evaluating some of the possible impacts of federal R&D expenditure on private R&D activity, it is less well suited to analyzing others. Two features of federal industrial it&l) activity whose impact on private R&D it is difficult to assess ~ in this framework and perhaps in general ~ are its relatively high degree of concentration and its relative instability. As Table 1 indicates, federal R&D ascends deco be performed by a smaller number of large companies than privately sponsored R&D. Thus, increases in ache federal share of total industrial it&l) expenditure would be expected to result in an increased concentration of R&D resources. Table 2 illustrates the substantially greater volatility in the growth of federal Red) than in the growth of company it&l) between L956 and 1983. Two measures of volatility, the coefficient of variation and the range of the growth rate, are, respectively, 5 . 6 and 3 . ~ times as large in ache case of federal R&D as they are for company R&D. This contrast reflects the episodic nature of federal commitments to devote resources to industrial Red). A troubling possibility is that the relatively volatile pattern of federal investment destabilizes the market for private R&~) output by increasing uncertainty about the future course of technological change and, perhaps, as a result, depresses the quantity of private R&D investment in the long run CROWDING OUT lye crowding-out hypothesis is that increases in federal R&D activity increase ache private cost of producing R&D output by driving up the prices of specialized R&D inputs. The hypothesis is predicated on the assumption that federal and pri~ra~ce R&D employ similar types of resources, such as scientists and engineers, the aggregate supply of which is less than perfectly elastic. A graphic representation of ache crowding - out phenomenon is gird in Figure 2 . Suppose, for concreteness, that the graph represents ache market for R&D scientists and engineers. The supply schedule of scientists and engineers is labeled S. and the federal and priorate demand schedules are labeled DO and Dp, respectively. 'the market-clearing wage race W is determined by the in~cersection of ache supply schedule and the aggregate demand schedule D ~ the horizon~cal sum of OF and I):, I, and equilibrium employment in each of the two sectors* (L ~ and L p) is found by evaluating ache demand schedules at ~ . An outboard shift in DF would result in an increase in W and a reduction in L p. The magni~cude of these effects depends, of course, on the slopes of ache demand and supply schedules. Suppose that these schedules are linear, so that they may be written as follows: supply: L aS t. bSW federal demand: LF aF ~ by priorate demand: Lp ~ ap - bpU bS, bF, bp 2 ° - 117
TABLE 1 Comparison of the Extent of Concentration of Federal and Company R&D Funds, 19 81 Federal Funds Percent of ~co~cal R&D funds spent by companies witch: 25, 000 or more employees 83 Fewer than 1, 000 employees 3 Percen~c of total R&I) funds spent by 20 largest RED performers Source: . 68 65 6 41 Company Funds Research and Developmene in Industry, 1081 . NSF 83 - 325 . Washington, DC: National Science Foundation, 1983, Tables B- B-ll, B-7 9. ! 1 8 - - _,
TAB Lo: 2 Comparison of the Extent of Stability of the Growth in Real Federal and Real Company R&D Spending, 1956~1983 Growth Rate 0 f Real Federal ILL Growth Race of Real Company 8&1) Spending Mean · O 021 O 049 S tandard deviation . 074 . 031 Coefficient of variation 3 O 52 DO 63 )£irlim~ ~ ~ t35 - . 028 Maximum O 231 . 090 Range .366 .118 Note ~ Federal and company R&D funds deflated using GNP implicit pric deflator ~ .. Source: Research and DeveJopmenc in Indusery 1983 National Science Foundation, 1985. - 119 - Washington OC . e
FIGtJRE 2 Marker for Scientists and Engineers: Impact of Shift in Federal Demand from O~ to D F on Cage Rate and Private Employment of Sciert~cists and Engineers ~~ t 4, It \ · 1 OF l I ~ F - - - - \.~ Dp ~ S ~ \ D 1)' L'; Lp - 190
The equilibrium condition is L - LF + Lp. The maricet-cl~aring wage rat te is ~cher.efore ,,*_ aF + ap - aS and the effect of a unit shift in the federal demand curve on W is dW, 1 daF bS + bF ~ bp This effect will be larger, the smaller each of the parameters b. by and by is ~ that is, ache less elastic the supply and demand schedules are ~ . Unfortunately, very little is known about the values of the demand parameters by and bp. In view of the national security orientation of much federal R&D, and the limited incentives deco minimize costs provided by federal con~cracting methods (for example, cost-plus ), one might reasonably cony endure chat bF is fairly small, perhaps very small, re latitude deco bp . Me lack of info rmation about b is particularly unfortunate, since the effect of a shift in federal demand on private RED employment, A, is -bp. <pa . Although there do not appear to be any estimates of ache complete set of structural par~ne~cers (bp, by, be)' as discussed below, some information on the reduced-form parameter A* is available. If there were an estimate of the daF elasticity of private demand for scientists and engineers, one could, d'7* using the info relation about <~a ~ predict the effect of a federal demand shift on priorate R&D employment. The only evidence that sheds ~ rather dime light on ache issue of the responsiveness of private R&D activity to changes in the price of Red) inputs is from the literature on R&D tax credits. On the basis of his study of a large number of American firms Mansfield5 reports that "the R&D Ax credit has had only a modest effect on American firms' R&D spending.. In principle, one would expect any two events that have ache same impact on the after-tax price of R&D inputs to have similar effects on priorate R&D employment ~ Thus, the finding that (before- tax) priorate R&D spending is insens itive Waco the introduction of tax credits sugges sacs that priorate R&D employmen~e is insensitive to variation in the wage races of scientists and engineers. As noted above, reduced- form estimates of the effect of federal demand shifts on market wages of scientists and engineers have been obtained in several studies. In two separate studies, Richard 121 -
Freeman analyzed the behavior of the markets for physicists and for engineers and the role of federal R&~) spending in determining their starting salaries during the period from the ~ ate ~ 940' s to the early 1970' s . He found that federal R&D spending played a preeminent role in determining the starting salaries of both Occupations: Changes in R&D expenditures were the chief factor influencing [eng~sleers' star~cing] salaries during ache post-Uc~rld War II period.. Me upswings and downswings in salaries car be traced to changes in federal R&D policy. And, Changed demand, due primarily deco changes in federal §&1) spending, was the chief source of changes in physics salaries. ~ Emphasis in original ~ Freeman estimated the elasticity of engineers' starting salaries with respect deco federal R&D expenditure (both is constant dollars ~ to be ~ n the range . 26 to .41. In ache case of physicists, he estimated the federal R&D elasticity of starting galaxies by decree tearer. The estimated elasticities were in the ranges 0 34 deco .36 for individuals with bachelor's degrees, .54 to .56 for these with master's degrees, and .33 to 044 for those with Ph.D. 's. Freeman's estimates are summarized in Table 3. Federal obligations for R&D, in current dollars, have increased from $38 0 7 bills on in fiscal year 1983 ~co an estimated $520 3 billion in fiscal year 1985. This represents ~ nominal increase of about 30 percent or Lassoing a two-year increase in the price laurel of 10 percent or less ~ in increase in real federal R&D expenditure of at least 20 percent. Freeman's estimates imply that the federal R&D elasticity of both engineers' and physicists' starting salaries is in the neighborhood of one third or perhaps somewhat higher. An elasticity of one third would imply that the. growth in real federal R&D spending between 1983 and 1985 has increased real starting salaries by roughly 7 percent over two years (10 percent if we use the estimated growth in R&D obligations to industrial firms). Evidence about the effect of recent changes in the growth rate of defense spending generally (not just federal R&D spending) on the growth of engineers' and technicians' starting salaries has been obtained from college placement officers. Real defense spending increased rapidly beginning in about 1979 until this year when Congress and the Reagan Administraelon agreed to end the buildup by maintaining a zero race of real spending increase. College placement officers report that increases in starting pay for engineers and technicians, which "have ranged from ~ to 12 percent over the last half-d~-n years, are expected to average only 3 percent his year. n 122 -
D TABLE 3 Estimates of Elasticities of Real Starting Salaries of Engineers and Physicists with Respect deco Real Federal R&D Expenditures ~ estimates refer co the period from the late 1940 ' s to the early 1970 ~ s Es timated Range of Elasticity End neers . 26 ~ . 41 Physicists B . S O . 34 - . 3 6 ,Y . S O ~ S4 ~ . 5 6 Ph O D G ~ 33 ~ .44 Source: R. B. Freeman. "Supply and Salary Adjustments to the Changing Science Manpower Marice~c: Physics, 1948 - 1973, American Economic - Retries, Vol. 65, No. [(March 1975), pp. 27-39; and R.B. Freeman. "A Cobweb Model of the Supply and Starting Salary of New Engineers, " Industrial and Labor Relations Reviews, Vol O 29, No . 2 (January 1976 ), pp . 236 - 248 . ~ 123 _
Freeman also found that the proportion of students entering college who decide to enroll in engineering and physical science degree programs is quite elastic with respect to the relative or real scarring pay of current graduates of Chose programs. But it takes four years deco "produce n an engineering or science graduate - -more in ache case of advanced degrees. The quantity of graduates in any given year, therefore, is determined largely by federal (and other) demand for such graduates a number of years ago, and ~ in the short run, i ~ can be increased only slightly in response deco current demand shifts. To summarize, there is fairly convincing evidence that increases in real federal R&D spending raise the wage rates of graduating scientists and engineers in the short run. Because the supp ly of graduates is much more e las tic in the long run (wi thin three to f ire years ~ there it is in tise shore run, such short- rust salary increases are likely to overshoot the tong- run response to an increase in federal R&1:) spending ~ The supp ly and demand diagram of the market for priorate R&D output indicates that a one. time, permanent increase in real federal R&I) expenditure would result in a fairly immediate upward shift in the supply schedule, from SO deco S1 ~ see Figure 3~. As time elapses, and more students enroll in and gradua~ce from science and engineering programs, starting wages begin to fall, and the supply schedule shifts downward slowly (~co S2 and subsequently ta S3 ~ . If ache long-run supply of scientists and engineers were perfectly elastic, the supply schedule would retur`` eventually to its initial position SO. SPI=O\tERS The previous section cites evidence that increases in federal R&D activity may cause an at Least transitory upward shift in the supply schedule of private R&D output, due Deco induced increases in the prices of R&D input. But, it is hypothesized of Icon that federally supported R&D generates knowledge that can be exploited cheaply or costlessly by priorate sponsors of R&D in their search for new and imp roared products and processes. Federal R&D is postulated to add ~co the publicly available stock of scientific and ~ce~hnological knowledge - cap Incas, which is an impor~cant, if difficult to measure, input Deco the private production of knowledge. If federal it&l) does contribute deco the stock of intangible knowledge that is available and useful to priorate sponsors of R&D, then it might be hypothesized to increase ache productivity of conventional R&D inputs, such as the services of scientists and engineers and of R&D plant. The larger the stock of knowledge-capital, the smaller the quantity of conventional Rho inputs required by priorate R&1:) sponsors to produce marginal improvements in products and processes. In this framework, the existence of spillo~rers implies there increases in federal R&D activity would result in downward shifts of the supply schedule of priorate R&~) output. The increases in the productivity of R&D inputs aris ing from such knowledge sp illovers would tend to offset, and 124 -
IGliRE 3 Crowding Out in the Short Run and the Long Run 5 - - - - S 2 - s3 c ~0 ~ D Quantity of privately financed R&D output ~ 125 -
could evens dominate, ache increases in R&D input prices associated with crowding Out e (To dominate, the elasticity of priorate R&D producti~ri.ty witch respect to real federal R&D expenditure would have to exceed the elasticity of R&D input prices, which, in the case of physicists and engineers, has been estimated to be about one third in the short Nan. ~ While the crowding-out and spillover hypotheses have opposite implications for ache effect of federal it&l) spending on priorate R&D output,. they do not necessarily have opposite implications for its effect on real private R&D input or expenditures The csowd~s~g-out hypothesis implies ocher increases in federal support of R&D, by driving lip the prices of Rat) inputs, reduce the equilibrium quantities of both pri~ra~c. R&D input and priorate R&D output. Chile the spillover hypothesis implies that increased federal R&D has an un~^guously positive effect on real priorate Rho output, ache implied effect on R&D input is indeterminate, ~ priori Al~chc~ugh priorate sponsors of R&D will be stimulated by knowledge sptIlovers to produce ~ greater quantity of innovations, they need to employ a smaller number of scientists, engineers, aced R&t) plant per innovation. Because R&D output is so difficult to measure, researchers typical, y the statistical relationship between The discuss ion above suggests that a significant negative relationship between federal and AND incur done: no indicate necessarily Chat the crowding:- out confine themselves deco analyzing federal and private R&D inpu~c. finding private _ _ ~~ effect dominates Cue spillover effect that is, that priorate R&D output declines in response to increases in federal R&D expenditure. The finding of a significant positive relationship between federal and private R&D input' however' would provide ~ fairly strong indication Chat the spillover effect dominates. To what extent has federal support of industrial R&D generated knowledge Chat is available and useful (in the sense of reducing R&D costs) deco primal e R&D sponsors? Stated differently, from the perspective of the priorate R&D sector, what is the value of the knowledge that has been produced by faderal.ly supported R&D? Unfortunately, i~c appears to be impossible to provide a general answer to this question on the basis of available information. Nor is it easy to imagine what kinds of data one could collect to provide a defini~cive answer. The remainder of this section considers several fragments of evidence that sugges tic chat the value of sp illovers genera~ced by the "average" dollar of federal tndustris1 R&D is relatively insigrtifican~c. But, Me quality of the evidence regarding sp illovers is, perhaps necessarily, inferior to ache quality of Me evidence regarding crowding out. Wi~chout doubt, some of ache most important technological. developments in the U.S. economy have been in areas in which ache federal government was the first maj or sponsor of industrial research. Host of the early research on computers, for example, was federally sponsored, and the entire course of development of the computer industry was fundamentally affected (no doubt accelerated) - 126 -
~ ~ government involvement. As is discussed at length in the next section, however, it is difficult Deco determine the extent Deco which it is direct government support of R&D, as opposed to government demand for the protucts to be developed by ache R&D, that influences the pattern of primate R&D investment. refile it usually is believed that federal R&D has a positive or, at worst, a zero spillover effect on priorate Ran, it is possible that in some instances the value of the spillover is negative For example, it is recognized that the design of most U. S nuclear power plants is based on systems developed originally for nuclear-powered Marry submarines. Perhaps to some extent because civilian power-plant designs were influenced heavily by an (unsuitable) technology developed originally for military purposes, the nuclear power industry has been ;~ financial disaster for public utilities, their customers, and the general publics While there appear to be cases of very large, positive (ant, perhaps, negative) spillo~rers arising from federal R&D, it is of interest to know the value of spy [lover resulting from the average do 1 lar o f federal R&D . S. ince Ache government does no ~ know, in advance, which proj ects or areas of research are going to yield substantial spillo~rers, federal R&D investment decisions should be based on the expected, rather Chart extreme, value of spillovers. Although there is no direct evidence on ache shape of the distribution of s?illo~rer values, it appears likely that this distr~bu~cion is highly skewed; that is, a very small number of federally supported R&D projects yield very large spillovers, and most federal R&D proj ects produce little or no ex~cesnal benefit. This con; ecture is based on findings from a number of studies that the distributions of both the (priorate ~ value of patented innovations and the value of patent protection are extremely skewed. Schanicerman and Pakes, for example, conclude from their analysis of the value of patent rights in the United Kingdom, France, and Ge.u~any during the post- 1950 period, that There is a dense concentration of patent rights with very little economic value.... Most of the value of the stock of patent rights is conc~n~crated in ache tail of the distribution (especially the upper fire percent).... The general picture of a sharply skewed distribution of Eve value of patent rights emerges clearly in all three countries. While the distribu~ctons of spillovers and of patent rights need not have the she shape, in the absence of evidence to the contrary, Ache presumption of a highly skewed distribution of spillo~rers appears warranted. Usually it is hypothesized that the probability that a dollar of federal R&D expenditure will generate spillovers (of given value) depends on the " stage " of R&D . Basic research represents the earl lest stage of R&D and is thought to offer the greatest prospects for generating external benefits. Applied research, which is - 127
directed toward specific obj ectives, ~s less likely to do so, and development expenditures are characterized by the lowest rate of spillo~rer. Me distribution of 1981 indus~cri~l R&D expenditure, by source of funds, is presented in Table 4. As the table indicates, less than one sixth of all federal R&D funds was devoted to research as opposed to deve lopmen~c, and only 2 percent was allocated to teas ic research. Additional evidence on the distribution of federal R&D by stage is available from the Federal procurement Data System (FPOS and is presen~ced in Table 5. According to those data, during the period 1979-1985, less than 12 percent of ache value of R&D contract actions for which a n stage n of R&D was reported was for research or explora~cory development. (According to FEDS definitions, some applied research may be included in the "exploratory development" category. ~ Advanced deve:1 opment, which involves the development of hardware taco test a design concept, and two subsequent phases of the development process accounted for about 84 percent of the federal R&D funds classified by stage. Less than one fourth of one percent wa devoted Deco commercialization, n the process of ~crans£erring a new or improved technology from engineering de~relopmen~c into the competitive market place . " Evidence on contractors' election not to acquire title to inventions that they have produced under government con~cract and on the licensing of goverrment-owned patents does not support the view that spillo~re~s tend to be substantial. lathe St~cement of Coverr~en~c Patent Policy issued by President Nixon on August 23, 1971, contained the following provision: Where ache purpose of the ~ government R&D ~ contract is to build upon existing knowledge or technology, to develop info~^a~cson, products, processes, or methods for use by the Government, and the work called for by the contrac~c is in a field of technology in which the contractor has acquired technical competence (demonstrated by factors such as know-how, experience, and patent position) directly related to an area In which the contractor has an es cabl ished nongo~rernmenta] commercial post Lion, Ache contractor shall normally acquire the principal or exclusive rights tbrou~3out the world in and to any. resulting innovations e ~ Emphasis added] Thus, ache go~rernmen~c agrees to grant contractors title deco innovations in situations where techno] oilcan spillo~rers (from governmen~c contrac~c research ~co commercial activi~cies ~ are most likely to occur. (Pses~ably, ache government agrees to this in order deco provide contractors witch the incentive to deploy research resources most effectively and to report research findings deco the sponsoring agency.) Butt as the data in Table 6 indicate, ache majority of cor`-cractors decline to exercise their right to acquire circle to innovations. During the period 1963-1972, contractors reported a total of 41,401 "invention disclosures Deco the Army, Navy, and Air Force; the sponsoring agencies acquired title to 27, 834, or about wo thirds, of the inventions, in almost every instance because the con~cractor elected not to claim title. Evidently, contractors feel
:.A=L_ ~ Distribution of 1981 Industrial R&D Expenditure o f R&~) and S ourc e o ~ Fiends Federal Funds ., by " S Cage ComcanY Funds Basic research 2.09' 3 .796 Apply ed research 14.3 23. 6 D eve lopment 8 3 . 7 72 . 6 Source: Research and Developmenc in Induse~, 2981. NSF 83 - 325 . Washington, DC: National Science Foundation, 1983. - L29 o
TABLE 5 Distribution of Value of R&D Contract ~Actions, by " S tage " of arc, Fiscal Year 1979-Fiscal Year 1985 (dollars in billions) Research Exploratory development Advanced development Engineering detre lopment Operational systems development Management and support Commercial ization Not Repor ted To Cal Source "Federal Proctsremen~c Data System. " author . _ 1 30 $ 3.7 ~ ~ 9 14. 7 400 9 12.5 3.2 0.2 21 ~ 9 $103 . O Data tape prepared for Eve
TABLE 6 Ember of Contractor Invention Disclosures Rece inured by De Agencies and Number to which Government Acquired Title, 19 6 3 -19 72 tense (1) (2) (3) Contractor N',~nber of Contractor Invention Invention Disclosures - Disclosures deco which Go~rer~ent ( 1 ) + ~ 2 ~ Agency Received Acquired Treacle (percent) Army 10,152 5, 269 5~. 9 Marry 12, 788 9, 6S7 7S.5 Air Force 18, 461 ~ 2, 908 69 . 9 To~cal 41, 401 27, 834 67 . 2 Source: Annual Report on Governmene Patent Pol icy. Washington, DC: Federal Council on Science and Technology, December 1971 Ad December 1972 combined. - 131 . -
that most or the time it is not worth the trouble to exercise their rights, even if they have "an established nongovernmer~-cal commercial pos Option" in a relevan~c market . The relatively tow rate of licensing of unexpired government- owned patents suggests also that the inventions covered by the patents are not perceived as terribly valuable by private firms and individuals. Table 7 reports the number of unexpired go~rernmen~c- owned U. S . patents available for licensing and ache number licensed a' the end of fiscal year ~ 972, total and by selected agency of ownership. Slightly over 5 percent of all unexpired patents were licensed ~ the rate of licensing of govers~ment-os~ned foreign parents was about half as high, 2 ~ 7 percent) . Me rate of licensing was considerably higher for patents "owned" by civilian agencies than it was for those owned by defense agencies. The extent of spillovers arising from federal industrial R&D is reduced almost certainly by the secret or highly classified nature of Irleuch federally supported activity. Often9 a company's federally and privately funded R&D proj ects are performed arc different locations and in different adminis trative digit ions, in part due Deco the requirement to limit access to information about government research. To summarize the discussion of sp illo~rers, i ~ has been argued the ~ knowledge generated by federal R&D in~restmen~c could, in principle, be useful and available to private R&D sponsors, thus towering their private cost of achieving given improvements in products and processes and tending to offset possible crowding. out effects of federal R&D o Me fact that most federal industrial R&D is at the advanced development stage and highly classified suggests ~chat the extent of spillovers is not large, and the data on the distribution of property rights to contractor inventions and on the licens ing of government- owned patents tend to support that view . But, probably, the frequency distribution of spillover values is highly skewed, and it is extremely difficult to make j udgments about the mean of that distribution. file federal R&D may increase the private costs of product ng innovations by driving up R&~:) input prices and also reduce costs by increasing R&D input productivity, all R&D sponsors may not experience both effects, or they may not experience them to the same extent. Some of ache useful knowledge arising from federally supported R&D may spill over to, and reduce the costs of, foreign ~ for example, Japanese ~ R&D sponsors as well as domestic sponsors . Indeed, it is sometimes suggested that the Japanese are more skillful than Americans at adapting or transferring technologies, so ethic, conceivably, Japanese firms could realize greater benefits from U.S. go~rers~ment R&D than American firms do. In contrast, most of the burden of higher R&1) input prices resulting from increased federal R&D expenditure probably is borne by American it&D-performing companies. While increased U. S . government demand for scientists and _ , 3 2
- ~ ,, L I c U. S . Patents TAIL_ , Licensing of Government-Owned Patents by Federal Agency Sponsor- ng Research Underlying Patents, End of Fiscal Year 1972 (1) (2) (~) Number of Unexp ired Pates'cs Available for ~ 1 ) / ~ 2 ~ ~ icensing ~ number Licensed (I)ercenc) All agencies 23, 401 1, 238 ~ . 3 USDA i,444 i86 12.9 AEC 4 9 623 560 12. 1 NASA 1,903 107 5.6 Army 4,226 110 2.6 Navy 8, 372 73 0.9 Air Force ~ 1, 821 32 T. 8 II. Foreign Patents All agencies 5, 019 13S 2. 7 Source: Annual Repore on Government Percent Pal icy. Washington, DC: Federal Council on Science and Technology, December 1911 and December 1972 combined. - 133 -
engineers, plus higher U. S . salaries, might lead co increased net migration of such personnel from overseas, the resulting increases in foreign salaries are likely Deco be much smaller than those in U. S ~ salaries. Thus, the net effect of increases in federal R&D spending on ache priorate cost of innovation may be quite di-=erent in the case of U. S. and foreign companies. DEMAND PULL As noted earlier, changes ~ n the equilibrium quantity of priorate R&D output can result from shifts either in the supply (or cost) or in ache demand (or rewards ~ of producing innovations . The question of what fraction of the variation over time or across incus tries in the rate of innovative activity ~ s attributable to changes in demand versus supply continues Deco be debated hotly by analysts of technological change. Advocates of the "~cechnalogy-push" hypothesis, such as Nelson, ~4 attach greater importance Deco the role of supply factors in explaining variation in research activity. The 5 demand~pul' hypothesis, on the other hand' of which Schmookler was the maj or exponent, stresses the role of market opportunity ~ demand) in innovation. Kamien and Schwartz attempt to adj udicate the dispute by arguing that n the technology- push and demand- pull hypotheses may be viewed as complementary rather than competing explanations of innovation, with the former being more of a long- run theory and the latter, a short~rurt theory "I In a 1979 article, Mowery and Rosenberg criticized a number of empirical studies of technical change at the firm level, all of which had purported to establish the proposition that market demand is the dominant influence upon the innovation process. flowery and Rosenberg argued that To establish the primacy of demand-~ide factors one has deco show that demand conditions changed in ways more significant or decisive than changes in supply conditions--e. g., in cost', None of these studies deals witch this identification problem. They maintained the the methodology used in the studies they surveyed--open-ended interviews with businessmen--made it "virtually impossible to distinguish the demand-pull situation from that of technology-push-" Mowery and Rosenberg argued that, by failing to distinguish between demand- and supply-side determinants of innovation 9 The studies they reviewed erroneously attributed too much influence to demand. The situation is reversed in the case of econometric studies of the relationship between federal industrial R&D and private R&~ : Estimates of the correlation between federal and private R&D obtained in econometric studies are generally (albeit not explicitly) interpreted as reflecting effects of federal R&D on the supply of - 134 _
gr face R&D output, rather than as a potential link between federal R&D and ache demand for private ly produced innovations . The remainder of this section attempts to document the existence of such ~ link and to show how its existence affects the Interpretation of ache observed relationship bet:ween. private and federal R&D. Estimates of this relationship arguably "control for" the hypothesized link between federal R&D and the demand for private R&D output, and thus provide an at least partial resolution of the identification problem cited by Mowery and Rosenberg. Me estimates reveal that the relationship between federal and priorate R&D changes substantially when an indicator of federal demand for privately sponsored R&D output is controlled. The fact that a maj or fraction of privately financed R&D in the United S cares is directed primarily at the federal market for defense and space goods demonstrates the importance of taking federal demand for priorate innovations into account when analyzing the relationship between federal and pri~rat~;R&D. Some evidence on this point has been developed by Scherer, who, on the basis of ~ comprehensive analysis of patents issued to the 443 largest R&D-perfo~=ing companies, classified their privately funded it&l) expenditures by Ache principal ''target, n beneficiary, or ~user. ~ Scherer estimated the value of R&D expenditures classified by "users under the assumption that the benefits of R&D flowed to ~ single (the principal) user atone (which he termed the ''private goods" assumption), and under the alternative assumption that the benefits flowed to several users s imul~caneous ly ~ the "public goods " assumption) . As indicated in Table 8, Scherer's estimate of the value of 1974 privately £inan`:ed R&3) expenditure "used" by the government defense and space sector under the private goods assumption was $1,206.3 million, which represents about 8.3 percent of the topical estimated $14; 6 billion priorate industrial R&D expenditure during that year. Me amount of company- sponsored R&D expenditure "used" in final (personal ) consumption was estimated (also using the "priorate goods n assumption) to be S4, 111.0 million. In 1974, the two corresponding components of gross national product (GNP), federal national de fens e purchases and personal consumption expenditures, were, respectively? $77 billion and $888 billion. Hence, the ratio of priorate R&D expenditure "used" to total purchases (or average R&D intensity) was 1. 57 percent for defense purchases arid O .46 percent for consumption expenditures . Thus, not only is a s ignificant fraction of private R&D defense-oriented, but the average privately financed R&D intensity of defense purchases is 3 . 39 times that of consumption expenditures. This evidence casts serious doubt on the assumption underlying the National Science Board's interna~cional comparisons of civilian R&D investment Scent all privately funded U.S. R&D is "civilian" in orientation. A comparison of the company- sponsored R&1) expenditures of companies ' "government-oriented" business segments with the R&D expenditures of other business segments provides evidence quite consistent with Scheres's. Table 9 presents estimates for 1983 of - 13S -
TABLE 8 'value of 197~ Privately Financed R&D Expenditure "Used" and Value of 1974 Final-Demand Purchases, by Selected Final-Demand Component Federal Personal Defense Consumption Line Expendi Sure Expendi ture (1) Value of priorate $1, 206.3 $49 Ill R&1) "used" (millions ~ (2) Value of $77, too $888, 000 expenditure (millions ~ (3) (~/~2) 1.57 0.46 (percent) Sources : Line (1~: F..~. Scherer. "Using Linked Patent and R6cD Data co .~easure Interindustry Technology Flows'. In R&D, Parents ard Productivi By. Educed by Z . Griliches . Chicago: Univers i .~ o Chicago Press, 1~84. Line (2~: Economic Repore of the President. Washington, DC US S . Government Printing Office, February 1984, Table B- ~ . . - 136
TABLE 9 1983 Values of Total Company-Sponsored R&D Expenditure, Sales, arid Government Sales of Business Segments Classified by Extent of Government Intensity (dollars in billions) Government - Intensive Other Line Segments* Segmen (1) Sates $106 ~ 8 $3 ~ 503 e 6 (2) Company-funded 1.835 18.869 R&D expenditure (3) Government sales 57.136 1.591 (4) (29/~1) 1.72% 0~54% * Segments such that the ratio of government sales to total sales > Source: Compustat Industry Segment data file r ~ 137 .10.
.o Cal company- funded R&D, total sales, and sales to the government for over 9, 000 business segments, classified by extent of government orientation. Government-or~ented segments are defined as Chose for which the ratio of reported sales deco the government to the to Cal segment sales was at least 10 percent. This definition was adopted because a segment is required to disclose government sales only if the ratio of ache latter deco total sales is at least 10 percent. 1 The ratio of topical government sales reported by govers~mes`t - oriented segments to their total reported sales is S3 .5 percent; the corresponding ratio for ocher segments is less than one half of one percent . The average privately financed R&D inters icy of go~rernment-oriented segments is lo 72 percent, compared deco a value of O . 54 percent for other segments . lathe R&D intensity of go~rernment- oriented segments is 3.19 times ~cha~c of other segments, reflecting the fact that the total sales of ocher segments were 320 7 times those of government-oriented segments, Anile the total R&D expenditure was 10.3 times as high. The estimated difference between the R&D intensities of gover~en~c-oriented and other segments is almost identical to the estimated difference between the "used" R&D intens ities of defense and consumption expenditures implied by Scherer's deco. The substantial difference between the R&D intensity of federal expenditures and that of other components of the GNP suggests dacha. it is useful to represent the average private R&D intensity of the U. S . economy as a weighted average of the two R&D in~censi~cies, with shares in GNP as we fights: CEt.D ~ C=FED GNPFEI:) ~ C9DNON GNP GAPPED GNP GNPNON crd ~ crdf ~ f + crud . ( 1 where CAD GNP CRDF£0 - G=FED , C8DNoN GNENON crd crdf crd-, GNPNON ~- GNP f) aggregate U. S . company- funded Red) expenditure gross national product company-funded R&D oriented toward federal purchases of goods and services federal purchases of goods and ser~r~ces CAD - C3DrED GNP ~ GAP FED CRD / GNP CRDFED / GAFFED GNPFE~ / GNP CADNoN / G~PNON . - 138 -
Suppose Marc crdf and crdn are cons cants . Then, the above equation imp lies that the variation over tinge in average R&D intensity (crd) can be interpreted as attributable in part to variance on in the federal share in GNP (f): crd ~ Sac ~ ~ crdn + ~ crdf - crdn) . f ~ ~ ~ The regression of crd on f has been estimated using annual data for the period 1956-1983. Because both series exhibit trend (positive in the case of crd, 2:ega~cive in the case of f), a time trend is included in the equation. In the presence of ache time trend, the coefficient on f captures the effect of a deviation of f from its trend value on the deviation of crd from its trend value. Loosely speaking, the coefficient measures the short-run (transitory) . rather than the long- run (permanent) influence of f on cord. Estimates of the model are presented in Table 10. The estimated coefficient on f is positive and its difference from zero is highly statistically significant, a finding consistent witch the hypothesis that changes in ache composition of final demand account for a significant fraction of the short~run ~raria~cion in crd.2 lithe point estimate of this coefficient, .036 9 is much larger- -about three times larger- - than the estimated difference ~ crap - crdn) implied by both the Scherer and the business segment data. This discrepancy may reflect ache distinction between short-run and long~run effects on R&D intensity of changing composition of demand. In any extent, all the evidence presented points to the conclusion that the amount of R&D expenditure firms choose taco sponsor, conditional on their total Parolee of sales, depends upon the distribution of their sales between government and other purchasers. Because ache underlying priorate R&D irrtens icy of goods purchased by the federal government is substantially higher than that of goods purchased by other final demand sectors (households, stance and local governments, for example), the equilibrium quan~ci~cy of private R&O is likely to be influenced by the composition (federal versus other) as well an the level of final demand. In econometric studies of the determinants of private Red) expenditure, it is conventional to control for total sales (at the firm or industry level) or GNP (~t ache aggregate level) but not for the composition of sales or GNP. Thus, these models embody the a priori restriction that a given increase in federal sales or demand has ache same impact on priorate R&D expenditure as a corresponding increase in nonfederal sales, a res Friction that is unlike ly deco be true . lathe fo flowing evidence suggests that inappropriate imposition of this restriction results in seriously (upwardly) biased estimates of the partial effect of federal R&D expenditure on priorate R&D expenditure. Estimates (based on annual time series data for the period 1956 -1983 ~ of the regression of aggregate company- sponsored R&D expenditure (CAD) an aggregate federal industrial R&D Expenditure (=D), GNP, and a time trend can serve as a benchmark. The CRD, - 139 -
T^~L~- ~ O Regression of Aggregate Company-Funded R&I) Intensity on Share of .-ederal Defense Purchases in GNP, Annual Data, 1956-1983 Es ~ ima ted S tandard NTar'~ble Coe~¢ic;-en~c Error t- "at 0 f O.03611 0.01016 3. 5~3 ~ ime trend 0.00021 0.00002 10.054 interc ep sac - O . 415 0 . 042 6 - 9 . 7 3, p ~ . .377 sad. error (p) ~ . 189 R2 ~ . 96 10 Notice: f is defined as ache ratio of federal purchases of goods and se~~.ces to GNP. 1'.0 -
FRY, and SAP are all deflated us dung the GNP implicit price deflator.' Because of the uncertainty about the timing of the relationship between COD on the one hand, and E.RD arid GNP on the o ther, a future and a past value, as well as the con~cemporaneous value, of both FAD and GNP are included in the equation. Me estimated equation was of the form: So + 51E=~t + 1) + B2F~D(t) + B3=D(t ~ L) 1 ~ t2G~P(t) + <3GNP(t - 1) + /- t Estimates of the effects of ERD and of GNP on COD are based on the es timated sums of the corresponding coeff indents, that is ~ B B2 ~ B3) and (~: + Y2 +~3) . respec~ci~rely. Estima~ces of these suns, their standard errors, t- ratios, and associated probability; values are shown in the left-hand Coleman of Table 11. In this model, the point estimate of the son of the ED coefficients is 033, a value similar to (slightly larger than) th'8corresponding parameter estimate obtained by Levy and Terleckyj using a similar data- set and specification The estima~ced sum is highly significantly different from zero, a finding also consisten~c with Leery and Terleckyj ' s results . The right-hand Calvin of Table 11 displays statistics about the suns of coefficients when GNP is disaggregated into its components, FEI)GNP (federal purchases of goods and services) and NONGNP (GNP - FEDGNP). The res~criction Chat FEDGNP and NONGNP have identical coefficients, imposed a priory in the lef~c-hand column estimates, is no longer maintained in the right-hand column estima~ces. As expected, the estimated sea of the coefficients on ache FEDGNP terms is much larger- - about ten times larger - than the estimated son of coefficients on the NONGNP terms. lathe hypothesis that the sues of ache FEDGNP and NONGNP coefficients are identical may be rejected at the . 05 significance level. It is the reduction (relative to the restricted model) in the size and significance of ache sun of the FRD coefficients that is of primary concern. The estimated effect of ERD on CAD is only one shirt as large when FEDGNP and NONGNP are allowed to have different coefficients as it is when these coefficients are constrained to be equal. Moreover, the estimated see is insignificantly different from zero. Even though FEDGNP appears, in a highly restricted fashion, on the right-hand side of the restricted model, the fact that FEDGNP is a relatively small fraction of total GNP (in the neighborhood of 10 percent) implies that the restricted model does not adequately control for variation in federal demand. Relaxing the restriction in effect controls for the influence of federal demand and makes it possible to obtain a truer estimate of the partial effect of federal R&D expenditure on private R&D. Because FRD is a component of FEDGNP (~hat is, federal purchases of R&D services are included in total federal purchases of goods and services), the statistical insignificance of FRD signifies not that
TABLE 41 S to tistics from Aggregate Real Comply- Funded R&D - Expends Cure Regressions ~ ~ values in parentheses ~ Res tric~ced~ Made ~ Unres Eric ted "ode t Sum of FAD .330 .log coefficients ~ ~ (2045) (0. 60) ~ ~ O 013 a ~ 280 Sum of GNP O 006 coe ~ficients ~ ~ c 19 ~ ~ ~ ~ 125 Sin of FrDGNi . 052 coefficients (2 . 09 ~ ~ ~ . 028 She of NONGNP . 005 coefficients ~ 0 . 9 2 ~ a ~ .187 YEAR . 301.7 327.0 ( 1 e 54 ) ( ~ ~ 5 8 ) p ~ 404 ~ 173 (1~. 82) (O ~ 66) Wrote: ~ denotes probability value for testing null hypothesis that sum of corresponding coefficients equals zero. - 142 -
~ has no effect on priorate it&l) expenditure, but rancher Chat one cannot rej ect ache hypothesis that FRD has the same impact on private R&D expenditure as do federal purchases of non-R6`D services and products. In other words, the insignificance of AD implies that federal R&D expenditures do not have an impact on prince R&D above and beyond that of federal non-~&l) expenditures, although ache point estimate of the impact is about three times as large. If a similar model were es timated on a larger data set ~ for example, cross - sectional or longitudinal data for a large number of firms ), the estimated difference between the effects of federal R&D and non-A&D purchases Light be significantly different from zero. Butch as in the case here, controlling for ache composition of sales or demand. is likely deco reduce the estimated effect of FRD on CRD substantially. SUMMARY ANI) TENTATIVE CONCLUS IONS It has been argued in this paper that there are two principal mechanisms or paths by which federal industrial R&D may influence the cost to private R&D sponsors of producing given improvements in products and processes. Increases in federal R&D expenditure may, in principle, increase the average and marginal cost of private R&D performance by driving up the prices of R&D inputs of which the supply is less than perfectly elastic. On the other hand, federal R&D may lower private R&D costs by generating knowledge that is useful in the priorate production of innovations, hence raising the producti~ri~cy of privately employed R&D inputs. There is reasonably strong econometric evidence supporting the hypothesis that increases in federal R&13 result in significan~cly higher ~ starting) salaries of scientists, engineers, and technical personnel, at least in the short run. Because the supply of technical manpower is much more elastic in the tong run than it is in the short run, the effect on wages and salaries of a federal R&D increase is probably much smaller in the long run than it is in ache short run, and it may even be zero. Evidence regarding ache incidence of cost-reducing (from the perspective of private R&D sponsors ~ spillovers from federal R&D is extremely limited. Data on the distribution of contract R&D by restages of R&D, on the distribution of property rights to government-financed contractor inventions' and on the extent of licensing of government-owned patents all suggest that most federal industrial R&D expenditure produces little or no spillover. It is probate ly the case, however, that a small fraction of federally supported R&D generates very large spillo~rers (some of which may be negative ~ . I.c does not appear pass ible Deco estimate directly, from data actually or even po~centially available, the magr~i~cude of private R&D cost reduction yielded by the "average" dollar of federal R&D expend) ture . - 143 ~
Because it is not feasible to observe or directly estimate the net effect of federal R&D on the supply schedule of private R&D output, a number of inves tigators have attempted to draw ind~rec t inferences about this effect from estimates of the reduced- form relationship between privately funded and federally funded R&D expenditures, controlling only for focal demand. If the demand for privately produced innovations were stable, or if shifts in demand were uncorrelatad Erich changes in federal R&D, then the finding of a significant positive relationship between federal and company Rap) expenditure would provide rather strong evidence of spillovers. It has been demonstrated here, however, that changes in real federal R&D spending are correlated with changes in the composition of final demand, which shift the aggregate demand for private R&D Output. When statistical controls for the composition as well as for the level of demand are used, the hypothesis that federal R&D expenditure has ache same effect on priorate R&D expenditure as federal expenditure on non-A&D services and products (which stimulates considerably more private R&D than other components of final demand) cannot be rej ected. Because federal R&D expenditures are a relatively small fraction of total federal. purchases of goods and services, and do no appear to have ~ significan~cly larger effect on pri~ra~ce R&D than federal nonskid purchases, one must consider. and take account of government procurement behavior as a whole, and not just R&D contracting, deco understand the extent and nature of federal influence on private decisions to invest in R&D. 144
NOTES AND REFERENCES For a critical review of some of this literature, see F. Lichtenberg;. 'the Relat' onship Between Federal Contract R&D and Company R&D. ~ In American Economic Association Papers and Proceedings. Nashville, TN: May 1984. See, for example, Z. Griliches and F. I-ichtenberg. "R&D and Productivity- at ache Industrial Level: Is There Still a Relationship? " In ~R&D, Parents, and Productive By. Evinced by Z. Griliches. Chicago: University of Chicago Press, 1984. 3 . There also are problems associated with measuring ache quant' ty of ~ real ~ R&D input emp toyed for a particular purpose . Three such problems are worth noting. First, there may be problems of measuring total R&D expenditure in current dollars. A recent National Science Foundation report ~ see A Comparative Analysis of Information on National Industrial R&D Expenditures. NSF 85 - 311. t~ashing~con, DC: National Science Foundation, 198S retreats discrepancies between es timatPs of R&D expenditure reported by firms in their responses to the NSF/Census RD- ~ survey and Chose reported in 10-K reports to ache Securities and Exchange Commission; evidently, these discrepancies are due large ty to different treatment o f expenditures on " engineering and routine technical ser~rices.'' Second, there are problems associated with the deflation of nominal R&D expenditure required to obtain an index of real R&D input . ~ Industry- specific R&D deflators, however, have been developed recently by Mansfield; see E. Mansfield. "Price Indexes for R and I) Inputs, 1969-83. " video. ~ Third, the segregation of tomcat. R&D expenditures into company- and go~rernmen~c- funded components, particularly at the firm or industry Petrel, is problematic. The difficulty arises, in part, because of the way in which Independent Research and De~relopmen~c (IR&D) expenditures are reported in the official (NSF) R&D statistics. (For a detailed discussion of this issue, see J. Reppy. The IRKS Pro gram of the Departmen~c of Defense. ~ Cornell University Peace Studies Program Occasional Paper No. 6, March 1976; and J. Reppy. "Defense Department Payments for 'Company-Financed' R6iD, ~ Research Policy, Vot. 6(1977), pp. 396-410.) Serious as these problems of R&D input measurement are, they pale in comparison deco the conceptual and practical difficulties of measuring R6:D output. 4. See, for example, M. Nadiri and I. Prucha. "Comparison and Analysis of Productivity Growth and R&D Investment in ache Electrical Machinery Industries of the United States and Japan. " Pacer presented at ache National Bureau of Economic Research Conference on Productivity Growth in Japan and the United States, Cambridge, MA, August 25 - 28, 1985 . ~ 145
~ . E. Mans fly eld. "Commentary: Capital Formation, Technology, and Economic Policy. n In "Industrial Change and Public Policy, " from ~ shakos in sponsored by the Federal Reserve Bank of Kansas City, Missouri, 1983, p. 261. The smaller be is, the greater do /da~ is, the wage increase resulting from a given federal demand shifts but ache smaller the private employment response to a gives wage increase . S ince the latter is a first ~ order effect and the former is a second~order effect, for gibes bS and bF, the absolute value of P is increasing in be,: P _ - (1 + S + F ) i daF bp bp 7 . R. B. Freeman. "A Cobweb Model of ache Supply and Starting Salary of New Engineers," Industrial and Labor Reasons Review, Vo: . 29, No ~ 2 ( 1976 ), p ~ 244. 8 . R. B Freeman. "Supply and Salary Adjustments to the Changing Science Manpower Market: Physics, 1948-1973, n American: Economi c Revi en, To 1 . 6 i, No . 1 (March 19 7 5 ), p . 3 7 . 9. Federal obligations Go industrial firms for R&D, in current dollars9 have increased more rapidly°°about 40 percent from $18 0 6 billion deco an estimated $270 7 billion. 10. See 'For Military Suppliers, Growth Uncertainties, n New Fork Times, Section 12 (October 13, 1985 ), p . 10 . Research and development, or the management or ad~m~nis~cration of R&D, was the "primary work activityn of 62 percent of the 186,000 engineers reported deco be working in the area of national defense in 1982 (see 1982 Postcensal Sunder of Scientists and Engineers. NSF 84- 330. Washington, DC: Na~ciona1 Science Foundation, 1984, Table Be 13) . Hence, most of the changes in engineering salaries associated with changes in overall defense spending probably are at~cribu~cable to changes is defense R&D expenditure. 11. M. Shankerman and A. Palces. "Estimates of the false of Percent Rights in European Countries during the Post-1950 Period. ~ National Bureau of Economic Research Forking Paper No. 16SO, June 1985, pp. 19-21. . Product and Service Codes. Fairfax, VA: Federal Procurement Da~ca Center, 1982, p. 8. 3. Annual Report on Gotrernmenc Parent Policy. Washington, DC: Federal Council on Science and Technology, December 1971 and December 1972 combined. 1' 6
14 . Government and Technical Progress: A Cross - Indusery Analysis. Edited by R. Nelson. New York: Person Press, 1982. 15. Jacob Schmookler. Invenriora and Economic Growth. Cambridge, but\ Harvard Unifiers ity Press, 1966 . 16. M. KAmien and N. Schwartz. market Structure and Innovation. Cambridge: Cambridge University Press, 1982, p. 36. 17. D. Mowery and N. Rosenberg. "The Influence of Market Demand Upon Innovation: A Critical Review of Some Recent Empirical S tudies, " Research Policy, sol . 8 ( 1979 ), pp . 102 -15 3 . 18. Ibid. 19. F.~.. Scherer. "Using Linked Patent arid R&D Data to Measure Interindustry Technology Flows." In R&D, Parents, and Productivity. Edited by Z. Griliches. Chicago: IJni~rersi~cy of Chicago Press, 1984. 20. National Science Board. Science Indicators, 1982. Washington, DC: U. S . Government Printing Office, 1983, pp . 7-8 . 21. A relatively small number of segments for which the ratio is be low 10 percent also report government sales . 22. In view of the rule for reporting government sales, this is a lower bound estimate of the true ratio for other segments. I t is cus tomary, and appropriate, when analyz ing re la~cionship s among time series, to de trend or otherwise render stationary the series. Inclusion of the time trend reduced substan~cially, but did not eliminate, serial correlation of ache residuals, so an adjustment for first-order serial correlation was made in es timating the mode l . 24. The partial r2~ ~c2/(t2 + resid. do f. ~ ~ on f is . 34. 25. An equation is which CRD is expressed as a function of FRD and GNP (or itch components ~ may be interpreted as the reduced form of a system of supply and demand (of private RED output) equations in which "D appears in (shifts) the supply equation and GNP appears in the demand equation. Such a system may be represented as fo flows: supply) Ps ~ aO + alC8D + a2FRD (demand) PD ho + b~CRD + b2GNp where PS denotes the " suppl3r price " (marginal cost), and PD the " demand price n (marginal returns ), of private R&D output . equilibrium, PS ~ P:> Equating the right-hand sides of the supply and demand equations, and soldering for CRD, _ ~ 4 7
COD (50 ~ Do) 2 ~ + 2~ GL;? (~1 - b:) (~1 ~ bl) (a1 - b: ) GNP mav be expressed as the soul of two co~npones~ts, FEDGNP ~ federal purchases of goods and services ~ and NONGNP (GNP F_DGNP). The evidence cited above suggests that a billion-dollcar increase in FEDGNP strifes the demand for private R&D output substantially more than a corresponding increase is NONGNP does. Hence, the restriction that F£DGNP and NONGNP have identical coefficients in the demand equation should be relaxed ~ and tested). Since FIND is correlated fairly highly with FEDGNP, relaxing this restriction is likely to have a nontrivial effect on the size and significance of the reduced- foot coefficient on FEDRD . 26. Unfortunately, Mansfield' s it&D-deflator time series begins one y in 1969. See E. Mansfield. "Price Indexes for R and I) Inputs, 1969 - 83 . " Video . 27. lathe author experimented witch a large number of alternative specifications and found that estimates of stems of coefficients were much less sensitive to minor changes in equation specification than were estima~ces of individual coefficients. 28 . D. L eve and N. Teriecicyj . "Effects of Government R&D on Or ~ date R&D Investment and Productivity: A Macroeconomic Analysis," Bead Journal of Economics,, Vol. 14~1983~, pp. S51-561. Because values corresponding to periods ~ ~ + 1 ), ~ t), and ~ t - 1 ) of FRD, FEDGNP, and NONGNP are included as regressors, and each of these series is fairly smooth 5 individual coefficients Thor example, the coefficient on FEDGNP(~c) ~ are estimated imprecisely. Thus, tt is nor possible to rej ect the hypothesis tha. (using variable names to represen~c the names of their respective coefficients) FEDGNP (~c + j ~ - NONGNP (t ~ j ), j - -1, O. 1-- the hypothesis that one would test by comparing ache residual sums of squares of the restricted and unrestricted models. The hypothesis ~; (FEDGNP(t + j ~ ° NONGNP(t + j ~ ~ ~ O. however, can jest, be rej ected at the 5-percent level of significance . 30. Because, however, as is observed below, FED is a component of FEDGlJP, Me point estimate of the effect of Fed on CRD is . 161 ~ ~ .109 + . 052), about half of ache estimated effect in the restricted model. 31. Disaggregating GNP into its components also has the (desirable) effect of reducing the extent of serial correlation of =he res iduals substantially . 148
