In Chapter 7 we argued that one would expect decision rules that have stood the test of time to be plausibly responsive to variables to which firms should attend if they are to achieve their purposes . This will be our hypothesis about decision rules guiding R&D behavior. Many different kinds of organizations do R&D. The objectives of a university laboratory certainly differ from those of an applied labora tory of a government agency, and the objectives of an R&D labora tory in a profit-seeking business firm are likely different from either of these. Here we will focus on R&D done by business firms and pri vate inventors and will assume that profit is a dominant or at least an important goal- and the more profit the better. Our hypothesis about decision rules, then, is that their “form” can be explained in terms of how different variables impinge on the profitability of various levels and allocations of R&D. One would expect, therefore, that decision rules would be linked both to factors relating to the de mand for or payoff of R&D and to factors relating to the supply or cost of R&D.
Numerous studies have documented quite strikingly to what ex tent the amount of inventive effort is sensitive to the level of demand for or sales of the product in question. At the sectoral or industry level, Schmookler ( 1966) powerfully argued the proposition that the anticipated size of the market for a product is a consideration that influences the amount of R&D effort directed toward improving that product or reducing its cost. At least within manufacturing industry, R&D expenditure, direct and indirect (done by input suppliers), is correlated with the sales of the industry. And shifts in the pattern of sales tend to pull the allocation of R&D inputs in the same direction. There are certain analytical subtleties in the connection between the size of the market for a product and the amount of R&D that it pays to perform; these subtleties have tended to be overlooked by some of the researchers who have argued the important role of p roduct de mand. In particular, a whole set of problems that is associated with “externalities” from R&D needs to be analyzed in assessing the strength of the connection. We will pay special attention to the exter nalities problem in Chapter 15. It is very plausible, however, that the size of the market is positively related to the amount of research and development that it pays to do.
Th is plausible relationship seems to have been built into the deci sion rules that individual firms use to guide the level of their research and development spending. Several studies have documented that many firms have as a decision rule that R&D expenditures should be a roughly constant fraction of sales . Decision rules of this type at the firm level will generate the observed empirical relationships at th e industry or sectoral level if systematic interindustry differences in target R&D/sales ratios are not too large.
In contrast with demand-side factors whose influence on the level of R&D spending is reasonably well understood, the effect of vari ables that influence the ease or difficulty of inventing in particular product fields upon the amount of R&D effort directed to those fi elds is quite uncertain. Part of the reason for the uncertain ty is that it has proved difficult to get a firm conceptual grip on how one would mea sure differences in the ease of inventing across fields or on the v ari ables that would influence the ease of inventing. A number of con jectures have been put forth. Several scholars have proposed that liknowledge” relating to certain technologies is stronger than that re lating to others and that a strong knowledge base facili tates technical invention. However, it is difficult to state precisely just what IJ’stronger knowledge” means in this context. Some writers have as sociated knowledge with formal science and have attempted to clas sify certain industries (for example, electronics and chemicals) as more closely based upon science than other industries (for example, textiles) . But even when there has been an agreement regarding clas sification, there has been dispute regarding the effect of a s tronger scientific base upon the research and development inputs and out comes. Some economistsl notably Schmookler, have argued that the relationship between the strength of the scientific base in an in dustry and the amount of research and development that goes on is much weaker than the connection between the level of product · de mand and the amount of research and development spending. O thers , such as Rosenberg (1974)1 have argued that the pace of tech-nical advance is much higher in industries that are close to science than in those that are not. Notice that these positions are not neces sarily inconsistent, the former relating to inputs to research and development and the latter to outputs from research and d evelop ment.
Let us employ a variant of the schematic search model outlined above to try to bring a little order to this chaos . Assume, for ‘ the present, that product attributes are constant over the set of all pos sible p roduction methods and that all these technologies have con stant returns to scale and fixed input coeffici ents . The R&D decision maker does not know the economic characteristics of as yet unin vented or undiscovered technologies, but he knows certain tech nological attributes. These enable him to divide up the set of possi bilities into subclasses -a set of “blue” technologies, a set of “yellow” ones, and so on . At the existing set of factor prices he may know, for example, that the blue technologies are more promising ones for exploration than any of the others in the sense of stochastic dominance of the distribution of unit cost reductions.
The research and development process can be stylized as follows. The decision maker can “sample” from any of the subpopulations and “study” or “test” the elements of his sample. A study will ex actly identify the economic attributes, and hence the cost saving (if any) associated with the use of that technology if developed. Devel opment consists of making a known technology usable in practice. Suppose that the cost of the study is the same for all technologies and independent of the number of technologies tested. Development cost is the same for all technologies.3
For the present, let us consider a single period and assume that at most one technology will be developed . The decision m aker will draw a sample of given size determined by his level-of- effort deCi sion rule. He will direct his sampling to a particular subclass; in this case a sensible allocational decision rule will obviously indicate “blue.” If the “best” of the sample, when compared to the prevailing technology, has a cost reduction that more than offsets development costs, that technology is developed.
Assume that the level-of-effort rule is plausibly responsive to the relevant variables in the sense discussed above. Consider the effect of an increase in the “size of the market.” An increase in the volume of production that would be undertaken with a new technology in creases the total cost saving associated with any reduction in input coefficients. In this model of search, there are decreasing but positive returns to increasing the number sampled before commitment to development. The magnification of econom{c advantage from a “better technology” shifts upward the marginal returns schedule. Thus, a “plausibly responsive” decision rule would tie the amount of search monotonically to the level of expected production or sales.
Suppose there is an increase in the strength of the knowledge base, in the following sense. It suddenly is learned that the set of all blue technologies can be partitioned into a set of blue- striped tech nologies and a set of blue technologies without stripes, with the former a better set than the latter in the sense of stochastic domi nance. Drawing a given number of elements from the striped blue set will lead to a probability distribution of the expected unit cost saving of the “best” of the elements sampled that stochastically dominates the expected value of a cost saving of the best element of a sample of comparable size drawn from the whole set of blue elements. Stronger knowledge (in this sense) leads to a lower expected cost (smaller number sampled) of achieving an advance of given magnitude or to a larger expected advance from a given search expenditure (given number sampled). However, if one could compute an optimal strat egy, there is not necessarily any more sampling (R&D input) in the case where knowledge is stronger than in the original case.s Just as in more traditional cases, a decrease in the cost of achieving something increases the amount it pays to achieve, but not necessarily the inputs applied to achieving. The connection between the strength of knowledge and the amount of research and development that one II ought to do” is more complicated and difficult to see through than the connection between the level of demand for a product and the amount of research and development it pays to do. There is no reason to expect a plausibly responsive decision rule to link R&D spending closely to “the state of knowledge.”
This model, then, does provide some theoretical support for Schmookler’s conclusions about the loose connection between the strength of the scientific base and research and development spending. But it also provides support for Rosenberg’S argument. The “effectiveness” of R&D input is directly related to the strength of knowledge in the sense modeled above. Even if R&D input is no greater in industries where the scientific base is strong than in those where it is weak, one might expect the pace of technical advance to be faster.
Source: Nelson Richard R., Winter Sidney G. (1985), An Evolutionary Theory of Economic Change, Belknap Press: An Imprint of Harvard University Press.