Partha Dasgupta [a] and Paul A. David [b]
Toward a new economics of science
1. Introduction and motivation: Science, economics and politics
2. The old economics of basic research, and the emergence of a new economics of science
3. Knowledge: Codified or tacit? Public or private?
3.1. Knowledge, information, and the endogeneity of tacitness
3.2. Science and Technology: Public and private knowledge
4. Priority and adherence to the norm of disclosure in the reward
system of science
4.1. Priority and the science reward system
4.2. Priority and secrecy in Science: Public virtue and private vices
4.3. Culture: The enforcement of cooperative rivalry and collective regulation in Science
5. Resource allocation within scientific fields and programs
6. The timing of research programs within Science
7. Policy challenges: Maintaining Science and Technology in dynamic
7.1. Capturing the training and screening externalities generated by open science
7.2. Managing competition for scientists between complementary research activities
7.3. Promoting greater ‘industrial transferrability’ of university research findings
It has long been recognized in the sociology of science that priority of discovery or development is the basis for legitimate reputation-building claims, and that an individual’s reputation for ‘contributions’ acknowledged within his or her collegiate reference groups is the fundamental ‘currency’ in the reward structure that governs the community of academic scientists.  That scientists take intense interest in disputes over ‘priority’, and spend much effort collectively in determining for what and to whom this ‘coin’ shall be distributed, suggested to a functionalist like Merton (1973) the central role that competition for priority was playing in the organization of
28. To the extent that information-commodities are exchanged against money, or other goods, the system governing the production, dissemination and use of knowledge within the realms of Technology, resembles the market mechanism, most especially markets in which commodities are auctioned. The analytical similarity between competition among producers of (private) technological information and competition among bidders at auctions has been explored by Dasgupta (1986).
29. This retains the emphasis placed upon differences in the norms regulating information disclosure, and the functional rationalization provides for them in our earlier papers in this vein (David, 1984; Dasgupta and David. 1987, 1988). An expression of some reservations concerning the explanatory value of functionalist interpretations for economic institutions that have evolved historically, and a proposed historical explanation for the differentiation that arose between the two communities in the West from the late sixteenth century onwards are provided by David (1991).
30. We shall examine the arrangements within Science more closely in the following sections (5, 6 and 7) in order to support this assertion. The efficiency of resource allocation for R&D in a regime based on intellectual property rights is discussed in many places in the modern economics literature, including David (1993a, 1993b, pp. 225-229).
31. See, for example, Merton (1973). See also Blume (1974), and Whitley (1984). Gaston (1970), Cole and Cole (1973), and Cole (1978) provide quantitative evidence supporting the view that science is an unusual institution in the sense that it comes close to achieving a reward system that is ‘universalistic’, in Merton’s sense, rather than particularistic, i.e. achievement-oriented rather than ascription-oriented. In a study of university physicists, Cole and Cole (1967) found that quality of published research was the most important determinant of recognition that came in the forms of honorific awards, appointments to professorships at prestigious departments, and wide citation. While the relative weighing of quality and quantity of ‘contributions’ have not been established across different fields of science, there is a presumption that where a dominant paradigm, or research program reigns, there will be greater consensus concerning ‘quality’. As fields of inquiry go through paradigm shifts, the quantity-quality weighing is likely to be disrupted, and it should not be supposed that stability in the relative weights will be maintained across fields.
the Science community.  In the context of the reward system in science, the rule of priority serves two purposes at once: hastening discoveries, and hastening their disclosure. How it does that is seen readily enough. 
First, tieing rewards to priority sets up a contest, a race, for scientific discoveries. Since a scientist’s effort cannot in general be observed by outside monitors, payment cannot be based upon it. If funds were to be allocated for ‘effort’, scientists like anyone else would be given an incentive to slack off while declaring that they were working hard. Nor can intention be the basis of payment, for intention cannot be observed publicly either. By contrast, performance, if disclosed, can be observed and vetted publicly. So rewards can be based upon it; the greater the achievement, the larger the rewards - which may come, eventually if not immediately, in the form of salary increases, subsequent research grants, scientific prizes, eponymy and, most generally, peer-group esteem.
A method of payment alternative to one based on priority would be a fixed fee for entering science, but this would dull the individual’s incentive to work hard, since scientists could collect the fee irrespective of whether they produced anything of interest. So the reward has to be based in some way on achievement. However, it is often difficult to determine how far behind the winner the losers of a scientific race are when the winner announces his discovery. (Those who were left behind can merely copy the winner’s results and claim that they were very nearly there). For this reason, it is not possible in general to award prizes on rank. Thus, unlike tennis tournaments, science does not pay big rewards to the runners-up. This suggests a system of payment which is compatible with individual incentives. It is one where, roughly speaking, the winner is awarded all that is to be dispersed by the community for the discovery. The rule of priority mimics this.
We have offered a rationale for the rule of priority among scientists involved in parallel research on the basis of what is publicly verifiable. Fortunately for society, there is congruence between this requirement and the relative social values of the outputs of parallel research teams. For note that among the discoveries (or inventions) made by rivals involved in parallel research only the first is worthwhile to society; there is no social value-added when the same discovery is made a second, third or fourth time. 
There is, however, one immediately apparent difficulty about the rule of priority. If the losers of a scientific race were to receive absolutely nothing, the rule would place all the risks involved in the production of knowledge firmly on the shoulders of scientists. This cannot be an efficient system if scientists, like other mortals, are averse to taking risks which involve their survival and comforts. (One would expect individuals without private means to be particularly averse to absorbing all the risks. ) We conclude, then, that those who regularly engage in basic science research need to be paid something regardless of the extent of their success in the scientific races they choose to enter. Qtherwise many, if not most, scientists today would enter other professions.
All this suggests the desirability of a payment schedule which consists of something like a flat salary for entering science, supplemented by rewards to winners of scientific competitions, with the proviso that the better is the performance, the higher will be the reward. The flat-salary
32. See Lamb and Easton (1984, Chap. 10) for a historical survey of priority disputes and races for priority. Scientists may be motivated to establish claims of ‘priority’ because they seek fame through the attachment of their name to a discovery or hypothesis; because, as creative individuals, they need to secure the validation of their creation - in this case from an expert audience the feeling of having produced something new to the world and not just to the self (see Storer 1966), or because material rewards like salary and access to research facilities are linked to their reputational standing among their scientific colleagues. For the purposes of the immediate argument, the precise nature of the underlying motivation does not really matter.
33. Material in this section draws upon our earlier papers (Dasgupta and David, 1987, 1988).
34. By this we do not, of course, mean independent confirmations of a scientific discovery, which is a different matter altogether.
35. The analysis here is thus most appropriately interpreted to apply to arrangements for the patronage (or employment) of professional scientists, from either private or public sources, and does not bear on the pursuit of scientific knowledge by ‘amateurs’.
component of the public payment schedule acts as a drag on incentives to do research (for this reason it must not be all that high), but as we have seen, it is a socially necessary drag if there is to be Science.  Fortunately for the evolution of ‘academic science’, it has been found possible to tailor the flat salary to a complementary, productive activity - teaching - and thereby reduce the wastage occasioned by the drag. Roughly speaking, a modern scientist is paid in the form of a fixed salary (e.g. for teaching, should he be in academia) and bonuses (e.g. promotions, scientific awards, and general recognition) for priority in discoveries and inventions. 
The second purpose the rule of priority serves is in eliciting public disclosure of new findings. Priority creates a privately-owned asset from the very act of relinquishing exclusive possession of the new knowledge. To put it dramatically, priority in science is the prize. Now, the public disclosure of new findings provides two additional-social benefits. First, it widens the span of application in the search for new knowledge. It raises the social value of knowledge by lowering the chance that it will reside with persons and groups who lack the resources and ability to exploit it. Second, disclosure enables peer groups to screen and evaluate the new finding. The result is a new finding containing a smaller margin of error. The social value of ‘reliability’ established by disclosure to the community of scientists is that users of new discoveries can thereby tolerate a higher degree of risk arising from other sources of incomplete knowledge and information.
There is a third beneficial consequence, stemming from the fact that for priority to matter the race must be run towards a goal that is widely recognized, either at the outset, or subsequently, as one worth achieving. The autonomous governance system that has characterized academic (and, indeed, much non-academic) science in the West means that communities of scientific peers define what contributions to knowledge it is worth bothering to have arrived at before others. What effect does this have? It creates a cumulative, chain-linked impetus to the advance of knowledge, because what turns out generally to be appreciated is the disclosure of knowledge that aids (or is expected to aid) colleagues in the field in generating findings on the basis of which they can establish priority claims of their own. 
Of course, the reward system sets up an immediate tension between cooperative compliance with the norm of full disclosure (to assist oneself and colleagues in the communal search for knowledge), and the individualistic competitive urge to win priority races. This can engender neurotic anxieties on the part of researchers and ‘deviant’ patterns of secretive behavior.  Conceptually one may want to distinguish between departures from the norm of disclosure that take the form of remaining taciturn until ‘a result’ has been obtained and can be publicly announced, and incompleteness in disclosure - i.e. not revealing all that has been learned. The cleanest basis for such a distinction presents itself when results are put into codified form, as in the draft of a
36. In addition, most academic institutions limit the potential inefficiency of paying flat-fees, unrelated to research productivity, by awarding tenure only after some extended period of trial during which research capabilities and motivation may be assessed (albeit with some error). Advancement to tenure, and the security represented by entitlement to future flat-fee payments, is thus a part of the performance-based-bonus feature contained in the young researcher’s ‘contract’.
37. Employment contracts in industry and government research establishments also exhibit the generic two-part structure of compensation described as characteristic in academic research institutions, although in the former cases the fixed portion has to be viewed as a payment corresponding to the option value of the proprietary rights to the knowledge about future inventions and discoveries that are relinquished as a condition of employment. It may be noted that these arrangements are not uniquely modern; the post-Renaissance system of noble patronage of mathematician-scientists also generated a two-part payment structure, as David (1991) points out.
38. The channelling of individual research efforts by means of the emergence of collegiate consensus as to which priority races are worth entering does certainly impart momentum to particular research programs, but, as will be seen in Section 5 below, it also has some undesirable effects on resource allocation within scientific disciplines.
39. The sociologist W.O. Hagstrom, in a 1967 study, found that 30-40% of American (university) scientists in some fields expressed concern over being ‘scooped’ in their current work, to the extent that it could inhibit their willingness to discuss it freely with colleagues. See unpublished paper cited by Blume (1974, p. 38).
publishable paper, but this is not circulated for some time. Such delays are risky for priority-seekers, so typically, even when the hazard is borne for some advantageous purpose such as to permit a patent application to be filed, the delay periods are kept quite short. 
Secretiveness that takes the form of partial disclosure is rather more difficult to identify, but as it is a more pervasive and persistent pattern of behavior among researchers, it seems potentially the more serious source of wastage of social resources. Incomplete disclosure at the publication stage expresses itself in two forms, one of which is more readily detectable, i.e. .omission of information required for replication of experimental results. Since non-replicability will be reported, this reduces to a stratagem for ‘buying time’ and raising rivals’ costs while attempting to establish a claim to priority.  Self-evident incompleteness, such as the practice of not divulging at the time of journal publication the coordinates of the large protein molecule whose structure has been determined by use of X-ray crystallography and synchrotron radiation, appears to have increased in frequency and emerged as a point of controversy among investigators in the field in recent years.  Whatever else may be said of such practices, it should be observed that they represent a less socially wasteful mode of ‘post-publication delay’, inasmuch as resources will not be expended in the process of detecting the omissions.
More problematic is secretiveness about the special technical apparatus that has been created along with the published ‘results’. If, for example, researchers develop a more efficient computer algorithm for rapid computation or computerized database searches (as, for example, in mass spectrometry), without access to which replication of the findings reported by others will be infeasible, do the norms of Science call for them to share it with colleagues as soon as it becomes available? Whether or not they might do so in some centrally administered organization, the community of decentralized university-based researchers, each with control over access to their laboratories, cannot readily detect the suppression of ‘intermediate research products’ of this kind; it would be infeasible to enforce so strict a standard of disclosure. So, the conflict point typically arises among individual scientists or rival research groups when ‘a result’ has been codified and made public.
The value of intertemporal spillovers of tacit knowledge (between one project and the next), and the costs to the original possessor of transferring (‘sharing’) tacit knowledge, lie at the heart of these conflicts. The techniques that are created as a by-product of research leading to the first set of results often become the basis for the creators’ hopes of winning the race to the next set of results, and their claims for continuing research support. Preserving these as ‘craft mysteries’ is valuable, and taking the time and resources to calibrate instruments, or to adequately document computer code and datasets for use by other investigators, detracts from the private pursuit of new results. However, these private incentives may result in the new, more powerful techniques
40. Even so, the clash between the interest of researchers who fear being ‘scooped’ by rivals and those of (corporate) research sponsors who wish to be able to defer disclosure for periods of 30-90 days is well known to occasion difficulties in setting up university-industry cooperative research programs. See Peters and Fusfeld (1983 pp. 39-40). Of 23 universities for which there is information about prepublication review policy (from an N.Y.U. field survey in the early 1980s), only six insisted on no delays whatsoever or none exceeding 30 days. Pre-publication review requests were found to be considerably less of a sticking point for university officials than is industry interest in proprietary control. Other conflicts in establishing university-industry cooperative agreements to facilitate ‘transfer’ of academic research for commercial exploitation are noted below in Section 7.3.
41. When the practice is regularized, readers of publications from that source should eventually learn not to bother attempting replication experiments, and the publication ceases to communicate anything save the message that the researcher has obtained a new result. We would say that such a practice puts the research organization into the Technology camp, even though it may make a pretext of belonging to the Republic of Science. It has been alleged, although to date it remains undocumented, that chemical abstracts published by some research institutes in Germany during the 1920s were widely known to systematically misreport some important specification (such as the temperature applied) for the new synthesis that was being announced.
42. See discussion of the decline of the ethic of data-sharing in Science, 1990
remaining, at least for, a time, exclusively in the hands of their developers rather than being placed at the disposal of others who might have the complementary talents, techniques and resources to put them to more productive use. This wastage must be viewed as a regrettable necessity only if the reward system at each stage cannot sufficiently compensate researchers to induce them to develop research tools that would be useable (by anyone) in subsequent inquiries.
These same economic forces work to determine the location, in each field, of the customary boundary line between what gets codified and what gets left in tacit form. More generally, it can be seen that the boundary line between tacit and codified knowledge is not simply a question of epistemology; it is a matter, also, of economics, for it is determined endogenously by the costs and benefits of secrecy in relation to those of codification. One can see that accelerations in the progress of instrumentation and research techniques, made possible in part by synergisms and feedbacks between developments in the realms of Science and Technology, can raise the private marginal benefits of (a greater degree of) ‘tacitness’ for researchers in Science. If the marginal costs of transferring tacit information to others, being largely the time of the researchers and their support staff, is constant or rising, there would be an understandable tendency for the boundary between private tacit knowledge and shared tacit knowledge to shift towards the former, which in turn would weaken the motivation to bear the marginal costs of codification for the purposes of public disclosure. At the same time, however, falling costs of information transmittal, deriving in large part from computer and telecommunications advances, have lately been encouraging a general social presumption favoring more circulation of timely information and a reduced degree of tacitness. One of the resultants of these conflicting forces would seem to be the emergence of more active sharing of intermediate results, via computer networks, among quasi-private alliances of researchers. The advantages of pooling knowledge and swapping complementary techniques, being no less than formerly, and the costs of communication required for selective cooperation having fallen, this phenomenon is explicable by reference to the workings of self-interest. In other words, it is possible that cooperative behavior within a limited sphere can emerge and be sustained without requiring the prior socialization researchers to conform, altruistically, to the norm of communalism.
We have here a rather straightforward instance in which insights from the theory of repeated games are applicable to explaining cooperative behavior among potentially rivalrous researchers. To give precision to the essential ideas, we may start with the simplified case of two researchers (or compact research teams) working towards the same scientific goal, which entails solving two subproblems. Suppose that each ‘team’ has solved one of the problems. Once each gets the other solution, it will be a matter of writing up the result and sending it off for publication, the first to do so being awarded priority. We may further assume that the write-up time is, determined by a random process, and that if both get both halves of the problem at the same moment, each will have the same (one half) probability of being the first of the pair to submit for publication. Whether the winner will be awarded priority will depend, however, on whether or not some other researchers also obtain the full solution and succeed in preempting its publication. The question is: should the first research team to learn any part of the solution follow the strategy (S) of sharing that information with the other one, or should they adopt the strategy (W) of withholding? If, without prior communication, they play the strategy pair (S, S) they can proceed immediately to the write-up stage; if they play (S, W) the second member of the pair will be able to proceed to the write-up, and the opposite will be true if they play (W, S). Should they both withhold (W, W), they must both spend further time working on the other problem. It is evident that if they are only going to be in this situation once, the rule of priority alone will induce each of them to withhold, and they will end up (collectively, if not individually) at a relative disadvantage vis-a-vis other researchers who are hurrying to publish. If nobody else has the full solution yet, society also will have been forced to wait needlessly, because each member of the pair has a dominant (private) strategy of withholding what it has discovered.
The game just described will be recognized to have the structure of a classic two-person ‘Prisoners’ Dilemma’, from which bad consequences
can be anticipated.  It is well known, however, that an escape from the pessimal outcome (W, W) is possible in certain circumstances - namely, were this a game that was part of an open-ended sequence of such encounters (i.e. an infinitely repeated game), were the future not discounted too heavily, and were the players to expect the other team members to remember, and punish on future occasions their present refusal to cooperate.  However, the value in the future of developing and maintaining a good reputation for sharing has to be large to discipline the self-interested researcher into adhering to the sharing mode of behavior in the current period. If repetitive play comes to an end, or if the future is valued only slightly, cooperation will unravel from the distant terminal point in the game right back to its inception.
Yet, that is not the whole of the story. As there are other researchers in the picture, we should really be considering an n-agent game (again, where the agents may be individuals or small teams), involving the solution of an m-part problem, given n > m. Now the question of sharing information becomes one of sharing not only what you have learned yourself, but also what you have been told by others. It is obviously advantageous to belong to a coalition among whom information will be pooled, because that will give the coalition members a better chance of quickly acquiring all m parts of the puzzle and being the first to send it in for publication. On the other hand, if there are individuals who behave opportunistically by exchanging what they have learned from one group for information from people outside that group, but do not share everything they know within their group, they can expect to do still better in their current race for priority of publication. However, because others would see that such ‘double-dealing’ will be a tempting strategy, cooperation will be unlikely to emerge unless ‘double-dealers’ (who disclose what you tell them to third parties, but don’t share their full knowledge with you) can be detected and punished. What is the form that retribution can take? Most straightforward will be punishment by exclusion from the circle of cooperators in the future; and even more severely, not only from the circle that had been ‘betrayed’ but also from any other such circle. This may be accomplished readily enough by publicizing ‘deviance’ from the sharing norms of the group, thereby spoiling the deviators’ reputation and destroying their acceptability among other groups. 
43. It is not only in the fantasies of game theorists that bad things happen. Consider the following recent British newspaper account, appearing in The Independent on Sunday (31 October 1993):
Several teams of scientists closing in on the discovery of a gene that causes breast cancer have abandoned collaboration in their intense rivalry to win the race. Secrecy - spiced with misinformation - has replaced the co-operation that once aided the efforts of geneticists in Britain, the United States, Canada and France; such are the rewards of coming first... Three years ago, when Mary-Claire King at the University of California at Berkeley placed the breast-cancer gene somewhere on chromosome 17, scientific teams around the world formed a consortium to pool their resources in an effort to isolate it. They exchanged information regularly to identify regions of the chromosome that could be eliminated. But as the groups edged closer to identifying the gene they began to split apart, said Simon Smith, head of a Cambridge University research team funded by the Cancer Research Campaign. ‘Things have now gone quiet because none of us wants to give information to the others’, he said. ‘In an ideal world we’d be talking to each other and not holding back information. But our work is judged on what is published. If we are always second, it’s no good.
It is interesting to observe that the original consortium, or coalitional agreement, did not involve pre-commitments to joint authorship, presumably, because the number of participants was so large that internal monitoring of effort would be difficult, and because to do so would vitiate the point of a race for priority. Our two-person PD game, above, abstracts from the possibility of forming sub-coalitions against the rest of the field, a consideration that will be developed in the text below.
44. Indeed, there is a so-called ‘folk theorem’ to that effect. For a non-technical introduction to the literature on the repeated Prisoners’ Dilemma and its broader implications, see Axelrod (1984). The ‘folk-theorem’ of game theory holds that (if future payoffs are discounted by each player at a low rate) in the ‘super game’ obtained by repeating a finite, two-person game indefinitely, any outcome that is individually rational can be implemented by a suitable choice among the multiplicity of Nash equilibria that exist. See Rubinstein (1979, 1980) and Fudenberg and Maskin (1984).
45. See Greif (1989) and Milgrom et al. (1990) for analysis of repeated games of incomplete information that have this structure.
What, then, is the
likelihood that this form of effective deterrence will be perceived and
therefore induce cooperative behavior among self-interested individuals? If a coalition, i.e. ‘a research network’
numbering g players (g
The foregoing suggests that small cooperative ‘networks’ of information-sharing can be supported among researchers because cooperative behavior furthers their self-interest in the race for priority, and denial of access to pools of shared information would place them at a severe disadvantage vis-a-vis competitors.  Does this imply that the normative content of Merton’s communalistic norm of disclosure is really redundant, and plays no essential role in fostering conditions of cooperation among citizens of the Republic of Science? Not at all! For it can be shown that networks of cooperative information sharing will be more likely to form spontaneously if the potential participants start by expecting others to cooperate than if they expect ‘trust’ to be betrayed, and cooperative patterns of behavior will be sustained longer if participants have reason to expect refusals to cooperate will be encountered only in retaliation for transgressions on their part. Furthermore, detection of deviant behavior warranting punishment and implementation of the retribution of ostracism from a particular network will have more broadly damaging reputational consequences when the norms of behavior involved (i.e. the ‘custom’ within the network in question) are common knowledge, and part of the shared socialization among all the potential members of networks. It is evident from this that even if the process of socialization among academic scientists were weak and imperfect, the common ‘culture of Science’ makes it much more possible for the rule of priority to engage the self-interest of researchers in reinforcing adherence to the norm of disclosure, at least among a restricted circle of colleagues. 
The more general burden of the analysis presented in this section is that the workings of the rule of priority and its interaction with the norm(s) of disclosure are not just a matter of parochial
46. However, when there are inhomogeneities in communications that would tend to divide the coalition into tighter ‘sub-cliques’, a grand coalition will be vulnerable to defections by some among its members. This seems to have been the situation of the breast cancer gene research consortium, which was formed from a number of pre-existing national research teams (as described above, in note 43).
47. These ‘circles’ or ‘networks’, which informally facilitate the pooling of knowledge among distinct research entities on a restricted basis, can exist as exceptions to both the dominant mode of ‘public knowledge’ characterizing Science, or the dominant mode of ‘private knowledge’ characterizing Technology. Thus, von Hippel (1990) and others have described how firms in fact tacitly sanction covert exchanges of information (otherwise treated as proprietary and protected under the law of trade secrets) among their engineer-employees. Participants in these ‘information networks’ who accepted money or remuneration other than in kind would most probably be dismissed and prosecuted for theft of trade secrets.
48. The formal structure of the argument made here parallels a point about the role of ‘culture’ in defining and transmitting mutually held expectations about the consequences of ‘off-diagonal play’ in coordination games, which has been elaborated in quite another context by Greif (1992).
concern to egotistical scientists; even though the public typically remains unaware of its centrality in the reward system that controls academic science, priority matters greatly to society at large. The collegiate role of the scientific community extends into other forms of social service. We have so far concentrated on the inherent uncertainties involved in scientific research, but these uncertainties are conditioned by the quality of the researchers themselves. For the public at large are incapable of screening scientists by their innate abilities, and they are equally incapable of evaluating the relative importance of scientific discoveries; not only does one scientist look much like another, one publication looks pretty much like another as well! So scientists are themselves commodities of uncertain quality to the public, as are their past publications. Here, too, the community of scientists plays a crucial role, functioning collectively as an “agent’ for the society at large. It produces new scientists, and provides a check on their quality. It fails some, bestows a stamp of superior approval on others, and so forth. It constantly vets their research outputs, ranks their quality, and so on. There is, of course, an underlying danger of professional bodies abusing the public trust upon which rests the autonomy permitted them in their performance of the functions of an agent in these specialized matters. Professional bodies are often tempted to use their control of screening and evaluation mechanisms to make entry qualifications unduly stiff, and the costs of certification needlessly high; it is not unknown for some to succumb to the temptation, especially where society has delegated regulatory jurisdiction, by allowing the profession to set the terms on which its members will be ‘licensed’ to practice. There is less of a danger where the professional body is only a loose-knit one, so that there is some competition among sub-groups and sub-disciplines. This would appear to be the case with Science. 
In fact, the cohesiveness of the scientific community plays another role. It reinforces the political claims of scientists to ‘autonomy’ during periods when the public, or their putative political representatives, or the bureaucracy of the funding agencies try to impose closer direction and control.  Now, in addition to the benefits that individual scientists may enjoy in being left freer from the vexations of strict supervision, especially from attempts at strict control by inexpert authorities, the exercise of autonomy in the sense of the scientific community’s self-governance and control over the research agenda carries some obvious benefits for a society that values the growth of knowledge. Self-governance enables those who know to decide where research priorities lie (by combining societal evaluations of the importance of various research problems with expert assessments of the prospects of their being solvable within some relevant time-frame). It also leads to a better matching of scientific talent with the problems such talents are encouraged to attack. Furthermore, it encourages a better matching between scientific talent and the methods they pursue for solving these problems.
We thus see that uncertainty in the outcome of research and the inevitable privacy of much relevant information, taken together, provide the basis for offering a rationale, or functionalist explanation of much that is observed in the social organization and salient institutions of modern science. Sticking to this very gross level of observation, our discussion suggests that the distinctive institutional features and the reward system of Science does rather well in satisfying the requirement of social efficiency in the allocation of resources, but when one looks more closely at the detailed workings of this system, its many inherent inefficiencies begin to come into view. Taking these ‘fine-grain’ inefficiencies together, the resulting mal-allocation of valuable resources may be far from negligible. The following two sections, therefore, will be devoted to examining some of their main manifestations and underlying causes.
Because the outcomes of research projects are uncertain, it is generally in society’s interest to
49. The practice of medicine, as distinguished from medical research, clearly lies within the realm of Technology as we have defined it.
50. On distinctions between individual and group autonomy in regard to science, and the relationship between autonomy and power, see Cozzens (1990) and Turner (1990, pp. 198-204).
hold a portfolio of active projects which are run ‘in parallel’ within a particular field, or under the auspices of any specific scientific program that is determined currently to be worth pursuing. Therefore, in and of itself, parallelism or a multiplicity of projects aiming at essentially the same result - isolation of a virus, or development of a vaccine, or development of superconducting ceramic filaments - does not imply waste.  Society should thus be prepared to tolerate multiple discoveries in the sense of Merton (1973).  Nevertheless, a legitimate question arises as to whether the rule of priority and the reward structure in academic science encourage a more than desireable degree of duplication of research efforts, leading both to too many projects being discontinued by those who perceive that they have lost a race for priority, and to an excessive probability that researchers will unknowingly ‘multiply’ the findings of others. There are, in fact, a number of reasons why the incentive structure built around the rule of priority in Science is prone to cause wastage of resources in the form of excessive numbers of projects being launched in the same area, and an excessive correlation of research strategies among them. To identify these may suggest at least the broad lines along which remedial institutional adjustments and public policy interventions might usefully proceed.
We can begin here by calling attention to one generic cause of inefficiencies arising from the reward system in science, a cause that resembles the root of the conventional market mechanism’s allocative ‘fai1ures’ in many situations: the non-congruence of the way in which society at large benefits from the activities of scientists and the benefits being held out as inducements to individual researchers in this institutional setting - which is to say, under the operation of the rule of priority. The fundamental point is that society does not care who is successful in solving a given scientific problem, it cares that the problem is solved; and, in all save for the most exceptional of circumstances, society does not care whether the solution is obtained an hour, a day, or a month sooner or later. Yet for the individual scientists (or the scientific team), the identity of the problem-solver and the precise time at which his/her solution can be announced are matters of great concern; the priority-based reward system imparts great significance to differences in timing that are inconsequential from a societal standpoint. This sort of non-congruence between private and social rankings of final outcomes creates fundamental grounds for suspecting that the research portfolio that would be, in effect, selected, for society by the self-governing community of scientists will be an inefficient one. Now, a misallocation of research resources can manifest itself here in at least three ways. First, competition among researchers may encourage rival teams to undertake what in the aggregate turns out to be an unduly risky set of research projects (strategies) within a given program. Second, competition may encourage them to choose overly similar (i.e. positively correlated) projects within the program. In the third form of inefficiency, the system of rewards attracts too many research teams to a given race, to the possible neglect of other areas in which the entry of even a few competitors might be socially beneficial. 
It turns out that, provided private gain from a research success is made commensurate with the
51. The exception to this rule is, of course, the set of circumstances where experimental facilities are indivisible and the fixed costs entailed are so large as to rule out the benefits of diversification. Under such conditions, which more or less fit the case of the superconducting supercollider project (see Office of Technology Assessment, 1989), it is desirable to pursue only one project within the program, if it is desirable to embark on the program at all. However, within such large and complex projects, typically, there will be many sub-projects that present opportunities to pursue several solutions in parallel.
52. The term ‘multiples’ is. ambiguous but, following Merton (1973, p. 364 ff), its use among sociologists of science is not. Multiplicity connotes the occurrence of more than one research entity expressing essentially the same theory, or making what is essentially the same discovery or invention (including inventions of apparatus), and not that of a given research unit making more than one discovery. See Lamb and Easton (1984) for a recent treatment of the subject, which argues that the phenomenon of multiple discovery is inherent in the collective, evolutionary process through which scientific knowledge grows. In the present discussion, however, we are less concerned to account for what might be thought of as a ‘normal’, or ‘background’, level of multiplicity, and more with ‘excess multiplicity’ created by certain features of the science-resource allocation mechanism.
53. We need hardly add that there would be a resource misallocation if the reverse of what we have described were to occur under each of the three categories we have just listed.
benefits accruing to the collectivity of researchers from the success of any individual investigator or team, there is a tendency for the rule of priority to give rise to the first two of these three forms of misallocation.  To see why the first two kinds of distortion result from the rule of priority, it is sufficient to expose the tendency favoring excessive positive correlation between projects within a field or program. Consider a portfolio of possible research projects within a given field or program, each pursuing a particular strategy or experimental design, all of which offer the same expected social payoff.  The risks associated with some strategy pairs are highly correlated, those with other pairs less so. Assume that one of the research teams has chosen one of these research designs (projects). It -remains for the other to choose its project design. As between any two available projects, were the second team to choose the one which is less correlated with the one chosen by the first team, it would bestow a positive benefit to its rival. Specifically, the likelihood that team I is successful when team II is not would be higher. As we have already seen, this kind of portfolio diversification, or ‘insurance’, is socially desirable, but it is not necessarily considered in the second team’s private calculations concerning the course it should commit itself to follow. If the team were altruistic and more other-regarding, they would recognize that by picking a project identical to that pursued by another group, they would be lowering their (other) colleagues’ chances of achieving priority, and they would choose to be less duplicative in their research design. However, the ‘all-or-nothing’ aspect of priority-based reward structures encourages self-regarding egoistic choices among the community of scientists, rather than altruism, and so reinforces the tendency for researchers to be drawn into duplicative ‘races’. This works against the emergence of a diversified societal research portfolio. 
The fact that the reward structure it faces is hitched to the rule of priority pulls each research entity toward entering some well-defined ‘race’ in which the contestants are lined up along essentially the same track. Each may believe that some particular feature of their research design, say some special instrumentation or data analysis technique that has not been mastered by others, will give it a competitive edge, and all observe that winning a bigger race, in which there are a larger number of entrants, will do more for one’s collegiate status. The positive correlation among projects will not be perfect, of course. Having some feature of one’s research design differentiated from that adopted by competitors, even when the entire design is made common knowledge (say, by the process of peer review of proposals), may remain an attractive strategy for a risk-averse researcher who finds him or herself in head-to-head competition with a small number of identifiable rivals; creating some possibly inessential dimension of non-comparability in the outcomes may make it more difficult to pronounce the competition to have had a unique winner, and so allow those who arrive at a successful result later to share in the award of prizes. 
54. For the third phenomenon to occur we must assume in addition that the program involves Little Science, or in other words, projects that do not involve large fixed costs. In the next section, we will assume this to be the case.
55. How to conceptualize and measure the societal ‘payoffs’ from basic research is an immensely complicated question. Even the narrower question of what determines the magnitudes of the distribution of purely economic ‘payoffs’, and how these can be assessed, cannot be entered into here. For a critique of the methodology attempting to apply the techniques of cost-benefit analysis by tracing the commercial sequels of basic scientific discoveries and inventions, and the proposal of an alternative framework of analysis, see David et al. (1992).
56. The argument in the text does not, of course, constitute a proof. It offers a hint about how the proof goes. Since we wish to avoid technicalities here, we will not go into the reasons why under a wide range of circumstances the rule of priority encourages what from the point of society is excessive risk-taking on the part of rival teams (for this, see Dasgupta and Maskin, 1987).
57. An instance of inessential differentiation, undertaken largely for strategic rather than scientific reasons, is documented by Nicholas Wade’s (1978, esp. p. 279) account of the famous 21-year race between Andrew Schally and Roger Guillemin, who eventually shared the 1977 Nobel Prize for their work on the endocrinology of the brain. It seems that after 1962, when Schally ended 5 years of collaborative work with Guillemin aimed at discovering the hypothalamic hormone in the brains of sheep and formed a competing research group, he changed his research material and sought to obtain the hypothalami of cattle. This switch was justified by Schally on the ground that if Guillemin discovered the hypothalamic hormone first, his own work might be considered worthless were he also using the hypothalami of sheep. Other aspects of this complex case are discussed also by Lamb and Easton (1984, pp. 152-155).
There are, then, important respects in which academic scientists, while wanting to differentiate their work, come under strong systemic inducements not to be ‘lone wolves’ ranging far from the rest of the pack in their selection of research problems and approaches. To be sure, in any community there will be some people who seek to avoid the conflicts engendered when community members have goals strongly imposed upon them (seek priority) without being allowed recourse to effective means (maintain non-cooperative secrecy) whereby the goals can be obtained.  To the extent that they can do so by retreating to areas where there is little competition for priority, this imperfection in the collective response to the private reward structure will occasion some countervailing measure of diversification of the social research portfolio, but this functions simply as a random dispersal mechanism. There is nothing systematically operating, as far as we can see, that would match the talents of those scientists whose personalities and individual values dispose them to avoid competitive situations to the research requirements of those fields where few competitors are to be found. Moreover, this is the behavior of ‘deviants’; it is the more pervasive tendency towards the positive correlation of strategy choices in the race for priority among the more typical members of the research community (as much as the open communication of scientific theories and techniques) that, at certain moments in time, makes particular discoveries and inventions imminent and, so to speak, ‘in the air’, thereby promoting the occurrence of too many of Merton’s multiples.
In recent years the choice of the current mix of investment projects has been much discussed in the literature on social cost-benefit analysis; less so the timing of investments.  In almost all spheres of economic activity, most projects that are offered for appraisal are rejected. This is inevitable, but the fact that a project is not worth undertaking now does not mean that it will never be worth undertaking. Often enough, the right thing to do is to accept as a package a set of projects that are better when sequenced than when run simultaneously. Success in the first project might, for example, mean a reduction in the cost of running the second project, and so forth.
What is alluded to here is that there may be important positive spillovers across projects in the form of ‘learning effects’. Quite aside from the conceptual contributions that the codified findings of a research program in one field may make to accelerate research progress in another field, there is the question of spillovers affecting scientific equipment and skills, which often remain in the region of tacit knowledge but are nonetheless transferrable. Productivity gains in the performance of specified experimental tasks are likely to emerge as a by-product of the conduct of research through the development of superior instrumentation techniques (see, e.g. Moulton et al., 1990) including the development of generic computer software for performing data processing, storage, retrieval and network transmission. The training of specialized technical staff and post-doctoral researchers, whose skills eventually become available to other projects, represents another important source of spillovers.
The underlying idea here is that of optimal ‘waiting’. If society were indifferent about the order in which scientific advances occurred, but cared about the costs of the program as a whole, it would obviously want to sequence research projects so as to take into consideration the expected magnitude of the productivity spillovers from one project to the next. However, allocative decisions about science as a rule are not so centralized, and individuals and groups within the
58. Merton (1957) makes the point that norm inconsistencies of this sort can stimulate a range of ‘deviant’ responses, which include not only ‘innovations’ - such as new and ingenious forms of partial disclosure (secretiveness) among scientists, which we have examined in Section 4, but also ‘retreatism’. Gaston (1971) found that among those high energy physicists in his sample whose research findings had not been anticipated by other scientists, 44% said this was because they preferred working in the ‘less fashionable’ areas, where competition presumably was less intense.
59. On social portfolio choice and cost-benefit analysis, see e.g. Dasgupta et al. (1972), Little and Mirrlees (1974), Squire and Van der Taak (1975) and Lind (1982). David et al. (1992) seriously question the usefulness of applying the cost-benefit approach to public project evaluation in the case of basic science research projects.
scientific community, as well as in society at large, certainly take an interest in the timing of results. To make some further headway in the analysis of this, we can introduce a convenient simplification: imagine that the productivity spillovers take the form of reductions in the time (and resource inputs) required to reach a specified research goal, so that delaying the initiation of a project (until better techniques have been developed elsewhere) will not delay the expected date at which its findings can be announced. In this way we can put to one side the general motivation that all project leaders would have to start racing for priority as soon as possible. Consider, then, the implication that when rival funding agencies are involved the resulting competition could assume the form of a waiting game rather than a race (see Dasgupta, 1988).
Thus, it may be that it would be individually (privately) optimal for researcher A to delay initiating his project should researcher B initiate his project now. For example, if a major effort were being undertaken in the field of superconductivity, which promised reductions in the costs of building powerful electromagnets, it would be advantageous (under the conditions assumed) for high energy particle researchers to wait until those advances were available for incorporation into the designs for their ‘superconducting supercollider’. Moreover, it may be that it is privately optimal for B to initiate his project now should A choose to delay. In this case the sequence ‘A following B’ is consistent with individual incentives, but there are also may be situations in which it is privately optimal for A to initiate his project now should B choose to delay, and where it is privately optimal for B to delay should A choose to initiate his project now. In this case the sequence ‘B following A’ is also consistent with incentives. However, it could be that the sequence ‘A following B’ is, from the scientific community’s point of view, the superior alternative. Only a coordinated plan from the funding agency can ensure that this will come about.
In all this there is a key underlying assumption that the funding agency (or agencies) can make credible commitments about future funding. A would be willing (possibly even eager) to wait until B’s project is completed only if (s)he is assured now that (s)he will obtain the funds at the right time as part of an optimally packaged sequence of research endeavors. (S)he will, however, not be so willing to delay if there is substantial uncertainty surrounding the funding agency’s promises concerning projects to be supported at future dates. In this case, what should ideally be a measured plan of sequenced investment activity becomes merely a scramble for early support. If the investigators are equally well-credentialed, they would each expect to receive some funding, and this could be the worst outcome from a societal standpoint, with both projects now having to forward more slowly on their own.
The problems of attempting to sequence basic science research projects are exacerbated by the uncertainties surrounding the estimates of completion times. While some part of this difficult is inherent in the very nature of the activity, that should not be exaggerated; ‘the nature of the activity’ characteristic of Science reflects the influence of the reward system under which the researchers are motivated and organized. The winner-takes-all (of ‘almost all’) structure of the payoffs makes individuals and teams of scientists more inclined to select programs of research that are characterized by a high variance in the distribution of completion times for the constituent sub-projects. The difficulties of predicting completion dates in basic research are remarked upon frequently, and sometimes are viewed as an inherent characteristic of inquires of this nature, as has been remarked above. Yet, it is obvious that the researchers’ interests in early success is tantamount to making them concerned not with the mean and variance of the distribution of completion times, but rather with the extreme value distribution derived from it; they care about the expected minimum completion times, and the dispersion around that statistic. Now, it is known that in a sample of a given size, drawn from a continuous unimodal probability distribution, the expected minimum (maximum) value will be smaller (bigger) where the variance of the underlying distribution is larger.  The search domain characterized by the bigger dispersion of outcomes (here, project completion times) will be
60 See David et al. (1992) for citations to the statistical literature on extreme value distributions and discussion of other respects in which researchers may be hypothesized to be concerned with the shape of the extreme value distributions of payoffs and costs, and their implications.
the more attractive on that score, other things being equal. So, even if fundamental science was not such a hard activity to predict, and distinctive steps in some lines of inquiry were easier to plan in sequence, the prevailing reward system (while it encourages scientists to work quickly towards the completion of any step that is expected to bear a publishable result) draws them to work mostly on the less accurately predictable, and hence less dynamically ‘sequenceable’, among the available classes of problems!