|Results? Why, man, I have gotten a lot of results. I know several thousand things that won't work.|
A strategic alliance (Chan et al., 1997, pp. 199-200) "enables a firm to focus resources on its core skills and competencies while acquiring other components or capabilities it lacks from the marketplace." Such alliances extend beyond research, the focus of this paper, and take a variety of forms.
It is sometimes said that theoretical research in economics is a largely self-contained activity, firmly insulated from the vagaries of evidence about the subjects it analyzes. The extent to which this might be true in general is beyond the scope of this paper, but it does seem to be an accurate description of a good deal of the recent theoretical literature on R&D cooperation.
What is even more discouraging, however, is that much of the policy literature on R&D cooperation seems disconnected from both mainstream theoretical and empirical work on the phenomenon.
Much of the theoretical work on R&D cooperation in the 1990s took off from d'Aspremont and Jacquemin, AJ, (1988). The impact of this linear demand, quadratic cost of own-cost reduction, deterministic R&D model may well have surprised even the authors, who referred to it as "an example." In its basic form, it considers a duopoly market. Since there are only two firms, R&D cooperation, if it occurs at all, includes all firms in the industry. This is in sharp contrast to the kind of R&D cooperation described by, for example, Hagedoorn and Schakenraad (1993), who document that major players in innovation-intensive industries simultaneously undertake a great many cooperative R&D projects, often with narrowly defined targets.
In the AJ model, innovation is deterministic: firms select a certain cost reduction that they will pay for, and they obtain that cost reduction with certainty. Since innovation is deterministic, if a firm undertakes R&D at all, it performs only one R&D project. A deterministic formulation abstracts from the uncertainty that is inherent in innovation. It is not obvious why the deterministic formulation is preferred in the literature, since racing models allow for uncertainty and are not technically more difficult than deterministic models.
In the AJ model, if a firm performs R&D, it is R&D outputa cost reductionthat spills over to the other firm. This spillover just happens: if firm 1 pays for (and therefore obtains, given the deterministic nature of the model) a unit cost reduction x1, firm 2's unit cost falls by x1, where the spillover parameter lies between zero and one.
The d'Aspremont and Jacquemin model is unstable for large spillover levels (exactly when the impact of cooperative R&D on dynamic market performance is of greatest interest). Instability does not arise in the model of Kamien, Mueller and Zang (1992), who consider R&D input spillovers rather than R&D output spillovers. The two models are compared by Amir (2000), who shows that
Amir argues that the KMZ model is better suited for the analysis of independent and cooperative R&D. Amir (2000) distinguishes 7 R&D cooperation scenarios from the theoretical literature:
Specifications in which cooperating firms set the spillover parameter equal to 1 are considered by KMZ. While it may be that firms can increase their knowledge spillover rate, it is by no means certain that they choose to do so, and there is case study evidence that firms do not behave in this way. At least, it would seem preferable to model the choice of spillover rate, rather than simply assume that it becomes 1 for some cases of R&D cooperation.
Much of the theoretical (and policy) literature takes off from the idea that imperfect appropriability is a source of innovation failure in market economies, often appealing to Arrow's (1962, p. 615) frequently quoted observation that:
Even on a theoretical level, this view has been challenged. Cohen and Levinthal (1989, pp. 569-70) emphasize that information often does not flow freely from innovator to other users:
On an empirical level, Levin et al. (1988) present survey evidence suggesting that innovating firms have many strategies available to exploit their innovations, if not uniquely over all time, at least in advance of follows, allowing realization of first-mover advantages.
A frequent justification for promoting R&D cooperation is that it eliminates "wasteful duplication." This justification should by now be thoroughly discredited. It fails both theoretically (Dasgupta and Maskin, 1987) and empirically (Nelson, 1982b, pp. 455, reviewing case studies).
From a social point of view, effective pursuit of technological advance seems to call for the exploration of a wide variety of alternatives and the selective screening of these after their characteristics have been better revealeda process that seems wasteful with the wonderful vision of hindsight.
When the outcome of R&D projects is uncertain, as it always is, it is socially beneficial and frequently privately beneficial as well, to pursue multiple research paths toward a common target.
In the policy literature Teece (for example, 1996) and others have emphasized the tacit nature of some kinds of knowledgein sectors where technology transfer is difficult without the transfer of particular individualsas a justification for R&D cooperation. It may be so in some sectors: but if knowledge does not flow freely because of its tacit nature, then it cannot also be that firms cannot realize the commercial benefits that flow from their innovations because the underlying knowledge spills freely over outside the firm.
One of the motivations often cited for the U.S. National Cooperative Research Act of 1984 is that it served to reduce business-sector anxiety about possible antitrust liabilities incurred because of participation in R&D joint ventures. It is difficult to know upon what such anxiety might have been based. The European Union has always had a positive attitude toward R&D cooperation. U.S. antitrust, to the best of my knowledge, records one antitrust case involving an R&D joint venture. The government's theory in that case was that automakers had used an R&D joint venture to delay the development of environmentally motivated emission control equipment. The case was settled by a consent decree.
Keeping these characteristics of the theoretical and policy literatures in mind, I have preferred to examine selected portions of the empirical literature for regularities that might appear and offer some insight into possibilities for measuring the impact of public support on technological performance. I have focused in particular on studies of the pharmaceutical sector and of government support for or collaboration with private sector innovation.
The pharmaceutical industry was early on and remains (now along with the more broadly defined biotechnology sector) a proving ground for the study of innovation. In part, the pharmaceutical sector attracts attention simply because of its high policy profile. In part, it attracts attention because it seems apparent that static and dynamic performance in the industry is directly affected by government policy (although opinions differ about the nature of the effects). More recently, the industry has attracted academic attention because it seems to offer a close real-world counterpart to widely used racing models of innovation.
The evolution of the U.S. pharmaceutical industry was shaped by two early policy developments. The 1939 U.S. Food, Drug, and Cosmetic Act separated the purchase decision from consumption for prescription drugs (Temin, 1979, p. 34). Consumption could take place only by prescription; the physicians who issued prescriptions did not pay for the drugs they prescribed. The result was to make the demand for prescription drugs price inelastic. Further, pharmaceutical industry researchers in the 1940s developed systematic techniques for identifying antibiotics in soil samples. U.S. patent law was interpreted so that such antibiotics were patentable. This interpretation was by no means inevitable; antibiotics might have been found to be natural substances and not patentable (Nelson, 1982b, p. 456).
As a consequence economic profits were diverted into rent-seeking activities (Temin, 1979, pp. 443):
It has long been recognized that the pharmaceutical sector is one where patents are relatively effective in securing property rights in innovations. Temin's analysis suggests that effective patent protection did not result in good overall market performance in the U.S. pharmaceutical industry of the 1950s and early 1960s. Drug companies had the option of licensing products in which they had effective property rights, but opted instead to restrict output and dissipate the accompanying economic profits on marketing efforts aimed at physician-prescribers.
Temin finds no evidence of economies of scale in production: leading pharmaceutical firms grew by increasing the number of establishments they managed, but the size of the typical establishment did not increase. On the same point, Comanor (1986, pp. 1191-3) finds empirical studies inconclusive, some suggesting the presence of economies of scale, others not. One explanation suggested by Comanor is (1986, p. 1193) "that larger firms are relatively more important when all new drugs are included but not so in regard to the most important innovations."
Graves and Langowitz analyze the probability of introducing a new chemical entity for a sample of 16 large pharmaceutical firms over the period 1969-87. They find the elasticity of the expected number of patents with respect to R&D spending to be less than one and to fall as R&D spending rises, evidence for diseconomies of scale in innovation.
Caves et al. (1991) document that patent protection is not the only device that allows innovating drug firms to collect economic rents. They study 30 drugs that lost patent protection in the decade 1976-87 and find that after patent protection expires generic substitutes sell at substantially lower prices than the formerly protected version, but that the first variety suffers only modest reductions in market share. Advertising to prescribing physiciansan activity that is cut back in advance of the expiration of patent protectionapparently creates product differentiation that survives well after the introduction of generic substitutes.
Scherer (1993, p. 99) discusses the U.S. Orphan Drug Act of 1983, which gives exclusive marketing rights and tax benefits for firms developing drugs targeting diseases that affect relatively small numbers of people, which Scherer describes as having had "a marked impact." This is noteworthy because it suggests the ability of narrowly targeted measures to promote particular policy goals.
Debackere and Clarysse (1997) study a sample of 118 U.S. biotechnology firms over the period 1982 to 1994. They find that patent probability rises with the number of years a firm has been involved in collaborative research, suggesting that collaborative R&D enhances R&D productivity.
Cockburn and Henderson (1994) study research programs of 10 pharmaceutical firms in 38 research areas. They find substantial evidence of knowledge flows across firms (1994, pp. 507-8):
Henderson and Cockburn (1996, p. 33) identify three possible sources of large size in research and development, that:
A fourth benefit of large size, widely noted in the literature, might be termed the serendipity effect, and is associated with diversification as much as with large size alone: a large, diversified firm is more likely to be able to be able to exploit an unexpected discovery.
Henderson and Cockburn find some evidence that economies of scale in pharmaceutical research increased in importance in the period after that covered by Temin (1979), but also that such economies of scale may have disappeared after 1978. They do find that larger firms enjoy greater R&D productivity than smaller firms, and attribute this to economies of scopeknowledge spillovers across research programs within a firm and the accompanying ability to make productive use of knowledge spillovers across firms (1996, p. 55):
Cockburn and Henderson (1998, p. 159) highlight even more strongly the importance of absorptive capacity for dynamic market performance:
To the extent that strategic alliances add to or maintain absorptive capacity, they have a positive social benefit that is unlikely to be recognized by conventional evaluation methods.
Pisano (1989) studies the organizational form of 195 biotechnology sector collaborations involving private firms. His analysis suggests that firms favored equity holdings over contracts as an organizational framework for R&D collaboration. He interprets these findings from a transaction cost perspective: equity holdings raise the cost of opportunistic behavior (which would reduce the value of the equity holding), and representation on the board of directors of the collaborative entity is a vehicle for continuous monitoring of performance.
Pisano (1989, p. 124) suggests that antitrust treatment of R&D collaboration organized by means of equity holdings should balance these efficiency advantages against the possibility that an R&D joint venture might facilitate tacit collusion and worsen static market performance. Pisano (1991) notes that much (at least, much early) collaborative biotechnology R&D has been vertical in nature, involving on the one hand small specialized firms in a position to offer specific expertise and skills and on the other large established firms able to offer financial backing and access to distribution channels.
Taking these two contributions together, to the extent that R&D collaboration is vertical in nature, the potential for worsening static market performance is lower than would be the case for R&D (horizontal) R&D collaboration among firms operating in the same product market.
Pisano (1991) also notes a more recent tendency for established firms to integrate backward into R&D activity and for specialized biotechnology firms to integrate forward into production and distribution. Such integration, increasing the number of actual and potential competitors, improves static market performance. On the one hand, this finding suggests that public support for biotechnology sector innovation should be structured in a way that does not raise the cost of entering the market. More generally, it suggests that the way public support for biotechnology innovation is structured should not take for granted that observed private-sector relationships are fated to continue: the kinds of contracts and market structures that are privately optimal in one phase of an industry's life cycle may change over time.
Lerner and Merges (1998) analyze the allocation of control in vertical biotechnology alliances. They find that the greater the financial resources of the specialized biotechnology firm, the less the degree of control allocated to the larger (typically a pharmaceutical) firm.
Public policy affects equilibrium market structure and therefore equilibrium market performance. Any program of public support will alter the balance of bargaining power between biotechnology firms and larger partners. Programs of public support that increase the financial resources of specialized biotechnology firms are likely to increase the bargaining power of those specialized firms with respect to established pharmaceutical firms, reduce entry costs, and improve static as well as dynamic market performance.
Tapon and Cadsby (1996) analyze private sector-university pharmaceutical collaboration. Their discussion, like the work of Cockburn and Henderson, suggests the positive impact of knowledge spillovers on innovation. Tapon and Cadsby argue that private pharmaceutical firms link up with university laboratories to promote basic research and as a way of taping into specialized stores of knowledge in areas that developments in the field reveal to be important.
Quoting a biotechnology researcher, they document the inherently uncertain nature of biotechnology innovation (1996, pp. 389-90):
Tapon and Cadsby also find that physical proximity of research facilities has a positive impact on research productivity.
The importance of knowledge spillovers in the pharmaceutical sector means that the benefits of obtaining innovative results extend beyond the particular program that produces those results. Even a very precise measurement of the output of a particular pharmaceutical research program will provide only a lower bound measurement of the welfare impact of that output from a social point of view.
More generally, these results suggest thatcertainly in the pharmaceutical sector, and perhaps elsewhere in the economy as wellthat public support for private innovation should be carried out in a way the promotes the free flow of knowledge among R&D-active firms. A performance-enhancing quid pro quo for public support of private R&D is that patents obtained with such support should be openly licensed on reasonable terms by the private-sector firm that holds the patent., 
The inherent uncertainty of R&D outcomes in this area also signals difficulties for evaluation. Results must be assessed ex post, but any cost-benefit analysis must include the cost of programs that were reasonable ex ante but happened, by the luck of the draw, not to mature as rapidly or in the directions expected. Evaluation should not be carried out at too disaggregated a level.
Joly and Mangematin (1996) study 20 French public laboratories associated with the National Institute for Agronomic Research (INRA). They confine their analysis to two research departments, but even working with this limited sample they distinguish three types of public laboratories:
While some of their conclusions may be specific to agronomics and to the French institutional setting, the point that public laboratories are heterogeneous in terms of assets, expertise, and activities seems likely to be quite general. Further, the finding that some public laboratories apply specialized expertise to some specific, long-term problems for industry may well be a leading indicator of the role that will be played by (former) US defense laboratories.
This conclusion is consistent with Nelson (1982b, p. 453), who emphasizes the importance of the presence or absence of a government procurement interest in designing a program of public support for private innovation.
In the same vein, Ham and Mowery (1998), who present five cases studies of Cooperative Research and Development Contracts (CRADAs) between private firms and the Lawrence Livermore National Laboratory, conclude that the most successful joint efforts are those that "draw on the historic missions and capabilities of the laboratories" and that "defense laboratories are poorly suited to that task of civilian technology development in areas not directly linked to their historic missions." They also highlight the importance of the "generic" benefits derived by private firms from Cooperative R&D Agreements with Lawrence Livermore National Laboratory (Ham and Mowery, 1998, p. 663): "design principles, engineering techniques, testing methods" (surely, a kind of spillover contributing to absorptive capacity). What private firms pay under such contracts, and the value of the inputs they commit, may be one measure of what they expect the output to be worth. But to the extent that the benefits are generic, it will be a lower bound of the social benefit.
Another noteworthy result of the Joly and Mangematin study appears in their account (1996, pp. 917-8) of interviews with the director of research at INRA, who expressed disappointment with his experience that the information flows in public-private contracts were all one way, from the public to the private sector. They comment that this "shows that co-operative research is not synonymous with a process of combined learning." It is common in theoretical models to assume the innovation spillovers are complete within R&D joint ventures, and this interview evidence suggests that other specifications may be appropriate.
Jaffe et al. (1998) examine patenting practices of US Federal laboratories, and patenting and citation practices of the NASA-Lewis Research Center. Although they interpret their findings as confirming that patent citations are a valid index of the importance of the innovation covered by the cited patent, they also find that (1998, p. 196) "approximately one-fifth...of citations are cases where neither the technology nor the application is clearly related to the cited patent...apparently spurious citations" and that (p. 198) "citations are clearly a noisy indicator of spillovers." The conclusion that counts of patent citations are a valid but noisy indicatorwhether of importance of the innovation cited or of spilloversis not necessarily comforting from the point of view of using patent citations to evaluate the impact of an innovation or of a program to support innovation.
Leyden and Link (1999) discuss the empirical regularity that cooperative R&D projects that include public laboratories tend to be larger, all else equal, than those that do not. They point out that public laboratory participation in a joint R&D project most likely reduces the ability of private participants to appropriate profits flowing from successful innovation but may also reduce the cost (to private participants) of monitoring the R&D efforts of participating firms. If a joint R&D effort includes a large number of partners, any incremental reduction in appropriability is likely to be small, while the reduction monitoring costs remains as a private benefit to the participating firms. To the extent that such a reduction in monitoring cost enables joint R&D that would not otherwise take place, or makes such joint R&D as does take place more effective, there is a public benefit as well. The social benefit due to this type of reduction in transaction cost is unlikely to be caught by traditional measures of innovative output.
Lichtenberg (1987) criticizes econometric studies of the impact of direct Federal funding of R&D on private R&D spending that ignore differences in the composition of demand across industries. Since much private-sector R&D spending is aimed at satisfying government demand, industries that benefit from substantial government demand will conduct substantial R&D to satisfy that demand and also tend to receive greater-than-average government support for R&D. Ignoring demand variations (and simultaneous causality) tends to bias upward the estimated impact of private on public spending.
This point is correct in principle. It is not clear how important it is in practice. Using a sample of Federal Trade Commission line-of-business data, Lunn and Martin (1986) find that a greater share of industry sales to the Federal government lowers privately financed R&D spending per dollar of sales, while a greater share of industry sales to state and local governments increases it. Both effects are especially pronounced for a subsample of high-technology industries.
Cohen (1994, p. 162) makes a point about the impact of government demand on government-supported R&D that is perhaps more relevant to the question of program evaluation. In sectors where the public sector is a significant consumer, it can virtually guarantee the commercial success of sponsored projects by its purchasing decisions. In such sectors, commercial success is a weak indicator of program effectiveness.
Wallstein (2000) looks at another manifestation of simultaneous relationships in this area. He examines the impact of R&D grants made under the U.S. Small Business Innovation Research and emphasizes the importance of taking into account the incentives of funding agencies (2000, p. 83):
In the event, his results suggest that SBIR grants to publicly owned recipients crowd out private R&D spending on a one-for-one basis, so that public grants replace private R&D spending but do not increase total R&D spending. The implied risk for evaluation programs that adopt commercial indices of success is that they would create just such an incentive to fund R&D activity that would have been funded in any event.
Fouml;lster (1995) analyzes a sample of 540 R&D projects of Swedish firms and their research competitors. His results indicate that R&D subsidy programs that allow firms to choose the form of cooperation do not increase the probability of cooperation, but increase the incentive to invest in R&D. Subsidies that require firms to cooperate and to share results increase the probability that firms will cooperate, but decrease the incentive to invest in R&D. 
Rosenfeld (1996) reports two case studies of evaluations of U.S. state programs to promote network cooperation among small- and medium-sized enterprises. The evaluation methodology included surveys of participants, interviews, and some analysis of data describing the activities of the firms involved. One of the evaluations included an assessment of cooperation on the local economy. Evaluations of this kind have an unavoidable subjective element.
Luria and Wiarda (1996) report on objective evaluation of programs of the Midwest Manufacturing Technology Center. Their description will evoke admiration and give pause to those contemplating similar efforts. It is clear that objective evaluation is time consuming, costly, requires considerable effort, and is likely to be sector-specific, in that indicators of success developed for one industry often will not carry over to another. (Examples that they mention include manufacturing lead-time, inventory-sales ratios, and the on-time delivery percentage). They used the offer of benchmarking reports to entice firms that were not recipients of MMTC funding to contribute comparative data to the evaluation process.
Westerback (2000) studies the impact of Strategic Information Technology Management practices imposed on Federal agencies by the Clinger-Cohen Act of 1996. As one conclusion of the study, she finds that (2000, p. 38) "Use performance measures as a proxy for return on investment" is a useful information technology management practice for Federal agencies. But she also writes:
If this is true for measuring the return on government practices that are reasonably close parallels to functions performed and evaluated in the private sector, how much more serious will the problem of evaluation be for the federal contribution to strategic alliances, when the assets the federal agency brings to bear are fundamentally different from the kinds of assets found in the private sector, this very difference is what makes the alliance worthwhile for the private partner, and in any case the private return to investment in innovation can be measured in only the most approximate way?
Boulding and Staelin (1995) use the PIMS database to examine the impact of private R&D spending on the private rate return. There are many studies of this kind, and they generally find that the impact of private R&D spending on the private rate of return is positive, as do Boulding and Staelin. Such techniques might be applied to study the impact of public funding for or cooperative R&D on private returns (seeking to avoid the critique of Lichtenberg (1987)). The result would be a lower-bound indication of the social return.
Zahra and Bogner (2000) examine the impact of technology strategy on the performance of new firms in the computer software industry. Their measures of performance are the rate of return on stockholders' equity and the growth rates of sales and of market share. All three variables seem to have been afflicted by measurement difficulties.
It might be possible to measure the impact of strategic alliance participation on the rate of return on equity for relatively undiversified firms that participate in at most one strategic alliance at a time. If all private-sector firms allying with a government agency fell in this category, a summary of the effects might serve as an indicator of the value of such collaboration to the private business sector.
Externalities limit the use of the rate of growth of sales or of market share as an indication of the social return to innovation: against the benefit received by a cooperating firm the market share of which grows more rapidly than would otherwise have been the case must be offset the losses of rival firms the market shares of which grow less rapidly than would otherwise have been the case.
Chan et al. (1997) examine the impact of the formation of strategic alliances on movements in share prices of 460 firms involved in 345 alliances. They do not limit their attention to innovation alliances. They find that strategic alliances increase the combined market value of the firms involved, and that for horizontal alliances the increase is larger, all else equal, if it involves knowledge transfers.
While the event study methodology might be applied to evaluate the private returns to specific companies and for specific innovation alliances, it seems unlikely that it could be used to evaluate results of a support program aimed at a wide range of firms, not all of which would be listed on financial markets. The diversified nature of many firms and the large number of strategic alliances in which some are involved might also mute the impact of a particular alliance on firm value.
Yang et al. (1999) analyze factors determining the performance of NCRA-registered joint ventures. They find that performance is enhanced by alliance stability and if the alliance combines complementary assets. Their performance measure, however, is based on subjective evaluation of the extent to which a joint venture achieved its objectives. They specifically suggest (1999, p. 116) that "[a]chievement of objectives is an appropriate measure of intermediate performance for R&D strategic alliances in cross-sectional studies." But there will be many circumstances in which such a measure could not be constructed even with respect to the private rate of return.
Industrial economics as a field overwhelmingly employs partial equilibrium analysis. Strategic research alliances surely have some general equilibrium consequences, although it may well be that their primary impacts are confined to particular segments of the economy.
Keeping the existence of such general equilibrium effects in mind, it seems nonetheless to be the case that just as Nelson (1982a, p. 2) wrote "if they are to be successful, public policies to stimulate technical progress need to be nicely tuned to the particulars of the different economic sectors," so today we can write that if the evaluation of strategic research alliances is to be effective, so evaluation methods need to be tuned to the particulars of different sectors and of the types of alliances.
This observation is consistent with the evolution of empirical research on static market performance in industrial economics, which has passed from reliance on industry cross-section data in the 1960s and 1970s to analysis of time series and panel data at and below the firm level.
The data requirements to carry out such a study would be severe. Results would, of course, depend on the specifications used for estimation. That is true for all empirical work. Analysis taking product varieties as given would for the most part apply techniques that have appeared in the literature. Analysis that allowed for new product developmentan essential aspect of strategic alliance output in some sectorswould probably require use of techniques only recently developed and not yet widely applied (Bresnahan and Gordon, 1997). The results of such a study would give some indication of consumer and net producer benefits from strategic alliances in the sector under investigation. The results would not give an indication of spillovers outside the target sector.
There are valid questions about any evaluation scheme.
One relates to interpretation of whatever "grade" is generated. Low values of a particular performance index may simply reflect the highly skewed distribution of "big ticket" innovations. Scherer and Harhoff (2000, p. 563):
If major innovations come along only once in a great while, failure to achieve stunning results is not failure.
In such a world, the question the evaluator should seek to answer is not "Were the results good?" but rather "Ex ante, was it reasonable to think that there was a high enough probability that the results would be good to devote resources to the project?" For basic research, at least, ex post peer review might answer that question. Link and Scott (2000) present estimates of just such expected rates of return, based on survey and interview evidence, for a sample of projects subsidized under the U.S. Advanced Technology Program.
There is also the "spillover problem" of Klette et al. (2000, pp. 482):
Such spillovers might not be such a problem in sectors for which knowledge has a high tacit component. Firms in such sectors might, however, invest more-or-less adequate amounts in innovation without strategic alliances or other support mechanisms, since the tacit nature of knowledge would offer them the prospect of appropriating most of the returns from an innovation.
Nor should one lose sight of the impact of evaluation schemes on incentives. It should not be necessary to belabor this point to an audience the members of which have had to answer the question "What do we need to know for the exam?" If students need good SAT scores to get into college, and if high schools are evaluated based on how many of their students get into college, then it should not surprise if high school courses end up being organized not so much to educate, but rather to educate along the lines examined in SAT tests.
Keeping these caveats in mind, table 1 lists targets that might be associated with particular strategic research alliances. Innovative activity proceeds along many dimensions, strategic research alliances have many targets (and, most likely, any one strategic research alliance will have multiple targets). How one measures depends on what it is that one wishes to measure, and for each target, table 1 lists possible evaluation methods (column 2) and possible shortcomings of the suggested evaluation method (column 3). This litany of shortcomings is not a plea to abandon evaluation. Rather, the point I wish to make is that any and all evaluation will be highly imperfect, and that the results of project evaluation should be treated with appropriate caution.
|Increase knowledge||Peer review; publications, citations||Subjective; Edison problem: knowing that one line of research does not work is a result.|
|Transfer knowledge in public laboratories to the private sector||Count number of cooperative agreements signed; survey private sector partners||Not all strategic alliances are created equal.|
|Increase diffusion of (commercially applicable?) knowledge||Count patent licenses; count commercial applications; survey users; (event studies?)||Either requires subjective evaluation of patent, citation quality, or treats as equal things that are not|
|Augment absorptive capacity of commercial partners||Ex postsurvey, interview||Subjective; natural tendency to view one's own part of the world through rose-colored lenses|
|Increase level, effectiveness of innovation||Econometric studies of R&D inputs or outputs||Inputs are not the same as outputs; output studies: not all patents are created equal; superfluous citations?; may be an index of benefit to private partners; does not take impact on rivals, consumers, into account; does not give indication whether benefits would have been obtained without strategic alliance|
|Correct insufficient innovation in a market system||Full-fledged structural estimation||Stringent information requirements|
 My own preference is for the "operating entity JV" and "secretariat JV" classification of Ouchi (1989) and Vonortas (1994), which at least has the merit of being based on functional differences.
 Leontief's (1982) comments are well known: "Page after page of professional economic journals are filled with mathematical formulas leading the reader from sets of more or less plausible but entirely arbitrary assumptions to precisely stated but irrelevant theoretical conclusions."
 In some cases, at least, deliberately so (Teece, 1996, p. 194).
 Most of the small literature that uses racing models assumes that if a firm undertakes R&D at all, it undertakes one R&D project (my own work falls in this category); much evidence is to the contrary. Scott (1993, Chapter 8) is an exception.
 The possibility that firm 1 might license full use of the technology that allows the cost reduction x1 to firm 2 for a royalty payment (1 - )x1 per unit of output must surely have been considered in one of the many generalizations of the basic model. Such licensing does occur in the real world, and must be an element affecting the decision to carry out stand-alone or cooperative R&D.
 The literature abounds with taxonomies of R&D cooperation, with definitions depending on the number of R&D operations, on whether or not formation of an R&D joint venture means in increase in the spillover parameter, on whether or not firms cooperate in production as well as R&D. For alternative classifications, see d'Aspremont and Jacquemin (1988), Hagedoorn (1990), Kamien et al. (1992), and Hagedoorn et al. (2000, p. 569).
 Much of the policy literature seems to take it for granted that firms cannot reduce the spillover rate, at least, not to zero. This is why appropriability of the revenue that flows from successful innovation is thought to be incomplete. There is also the possibility that if firms could reduce spillover rates to zero, they would not find it value-maximizing to do so; Martin (2000).
 See Sigurdson's (1986) account of Japan's VLSI project. See also the discussion, below, of Joly and Mangematin (1996).
 I resist use of the common term "innovation market failure." My own view is that if there are such things as markets for innovation, they tend to be narrowly defined: the efforts of pharmaceutical firms seeking to develop an aids vaccine have not much to do with efforts to develop commercially applicable materials that will act as superconductors at room temperature.
 See also Kamien and Zang (2000) and Martin (2000).
 See also the discussion, below, of Tapon and Cadsby (1996).
 Scott (1989, p. 68) notes that in its policy proposals "the Reagan administrationsurely at least in part because of its concern with declining competitiveness of U.S. firms in global markets and in part because of its desire to deregulate marketsjustified these policies by extraordinarily selective reference to theory and facts."
 In the EU, this policy stance is constitutional in its foundation: Article 81(3) of the EU Treaty makes the promotion of technological advance one basis upon which the European Commission may permit cooperation that would otherwise be prohibited under Article 81(3) (which sets out EU policy on cooperation among firms).
 United States v. Automobile Manufacturers Association 1969 Trade Cases (CCH) Para 72,907 (C.D. Cal. 1969) (consent decree), modified 1982-3 Trade Cases (CCH) Para 65,088 (C.D. Cal. 1982).
 Scherer (1993, p. 101) suggests that the prices of first varieties may actually rise after the expiration of patent protection, with much of the retail market supplied by the first variety at a high price and institutional demand supplied by generic substitutes at a lower price.
 Similarly, Scott (1996) finds that the U.S. Clean Air Act Amendments of 1990 were able to promote private investment to control specific targeted pollutants.
 See also Mowery (1982, p. 352):
 See Martin (1996) for a formal model.
 The potential to worsen static market performance is present: if established firms systematically seal relations with small knowledge-intensive firms, costs of entry to the biotechnology sector might increase.
 See also Pollak (2000).
 See Jaffe (1989) and Adams and Jaffe (1996) for similar findings.
 Despite the fact, as noted by David et al. (2000, p. 506), that to the extent that public policy that promotes information dissemination, it may lower the expected profitability of later innovators discouraging follow-on innovation.
 R&D cooperation agreements that restrict the access of one party to the agreement to the results of the cooperation do not normally qualify for the EC block exemption permitting R&D cooperation "because they do not, as a general rule, promote technical and economic progress by increasing the dissemination of technical knowledge between the parties" (EC Commission, 2000, Para. 64).
 See Cicotello and Hornyak (2000) for an analysis of the terms of CRDA contracts. They do not assess the impact of contract form on CRDA performance.
 This reduced appropriability may be a private bad, but it is likely to be a social good. Furthermore, to the extent that the knowledge embodied in the innovation is tacit, public laboratory participation may not reduce appropriability to any significant extent.
 Robson (1993) finds that Federal support increases private spending on basic R&D one-for one. However, he works with aggregate data.
 Martin and Scott (2000, pp. 440-2) suggest a scheme of indirect public support for private innovation, with funding going in the first instance to venture capitalists, in effect reducing their cost of capital and allowing them to identify and fund R&D projects that would not otherwise receive funding.
 Fouml;lster (1995) distinguishes between information trading (of intermediate research results) among firms carrying out their own R&D projects and result-sharing within the context of a cooperative agreement. "Result sharing" seems to be a secretariat R&D joint venture with complete information spillovers among cooperating firms.
 Martin (1983) finds that greater line-of-business spending on R&D lowers line-of-business profitability, all else equal, while greater firm-level R&D spending increases line-of-business profitability.
 On this point, see Luria and Wiarda (1996).
 For examples, see Feenstra and Levinsohn (1995) or Roberts and Supina (1996, 1997). The purist approach to measuring the impact of strategic alliances on sectorial performance would be to specify and estimate a complete (demand-side and supply-side) structural model, allowing for firm-specific rates of technical progress and allowing those rates to depend on explanatory variables measuring both the firm's own participation in strategic alliances and on the sector-average frequency of strategic alliances.
 Not obviously more severe, however, than that confronted by Luria and Wiarda (1996).
 It should also be kept in mind that it may take some time before the value of an innovation is evident. Cournot did not even receive his first review, a harshly critical one, until after his death. The full import of his work did not begin to be appreciated until more than 100 years after it appear. The fundamental innovation embodied in the ubiquitous post-it sticker was developed in 1968, the product first introduced in 1980 (http://www.3m.com/about3M/pioneers/fry.html).
 Of course, if one is going to conduct peer review, one might as well do it ex ante, when it might have some effect on the allocation of resources.
 Their estimates are a lower bound, as expected returns to consumers are not taken into account.
 One might call this the Heisenberg evaluation principle.
 If public universities emphasize both teaching and research in evaluating faculty performance, but the availability of external funding is related to research performance only, then it should not surprise if greater weight is given to research performance in evaluating...but I digress.
 And recalling once again the motto of the Christopher Society: "It is better to light one candle than curse the darkness."