Research for this paper was aided by funding from the National Science Foundation (PRA 7721852), the Division of Research of the Harvard Business School, and an Andrew Mellon Foundation grant in the Program in Technology and Society at Carnegie-Mellon University. Nathan Rosenberg, Stanley Engerman, Peter Temin, Richard Nelson, Wesley Cohen, Alexander Field, Therese Flaherty, Zvi Griliches, Timothy Bresnahan, Mark Kamlet, Steven Klepper, Louis Cain, Albert Link, Edward Steinmueller, Jeff Oxley, and the participants in seminars at Harvard, Yale, and the 1983 Cliometrics Conference provided useful comments on earlier versions of the paper. Alfred Chandler and Takashi Hikino provided valuable advice and support in the analysis of the data in the paper, and Arvind Sathi provided programming assistance.

¹ Schumpeter, J. A., Capitalism, Socialism, and Democracy, third edition (New York, 1954), p. 132.Google Scholar

² Schmookler's analyses of patent data (which measure only inventive, rather innovative, activity) concluded that “During the [1900–1950] period invention changed form an activity overwhelmingly dominated by independent individuals to one less overwhelmingly dominated by business enterprise.” Schmookler, J., “Inventors Past and Present,” Review of Economics and Statistics, 39 (08 1957), 321–333. p. 333.CrossRefGoogle Scholar

³ See Hamberg, D., “Size of Firm, Market Structure, and Innovation,” Canadian Journal of Economics and Political Science, 30 (02 1964), 62–75;CrossRefGoogle ScholarScherer, F. M., “Firm Size, Market Structure, Opportunity, and the Output of Patented Inventions,” American Economic Review, 55 (12 1965), 1097–1125;Google Scholaridem., “Market Structure and the Employment of Scientists and Engineers,” American Economic Review, 57 (June 1967), 524–31; Mansfield, E. F., “Size of firm, Market Structure, and Innovation,” Journal of Political Economy, 71 (12 1963), 556–76; andCrossRefGoogle ScholarComanor, W. S., “Market Structure, Product Differentiation, and Industrial Research,” Quarterly Journal of Economics, 81 (11 1967), 639–57.CrossRefGoogle Scholar

⁴ Perazich, G. and Field, P. M., Industrial Research and Changing Technology, Works Progress Administration, National Research Project Report M-4 (Philadelphia, 1940).Google Scholar

⁵ “…13 companies with the largest research staffs, representing less than I percent of all companies reporeting in the National Research Council sruvery, employed in 1938 one-third of all research workers …” (Industrial Research and Changing Technology, pp. 9–10), and “of the 45 largest research laboratories employing half the total laboratory personnel, all but 9 are owned or controlled by companies which are among the 200 largest leading non-financial corporatios.” (Industrial Rsearch and Changing Technology, p. 11).Google Scholar

⁶ Temporary National Economic committee, Technology in Oure Economy, TNEC Monograph #22 (Washington, D.C., 1941), p. 212.Google Scholar

⁷ Mueller, W. F., “Du Pont: A Study in Firm Growth,” unpublished Ph.D. dissertation, Vanderbilt University, 1955;Google Scholaridem., “The Origins of hte Basic Inventions Underlying Du Pont's Major Product and Process inventions, 1920 to 1950,” in The Rate and Direction of inventive Activity (Princeton, 1962).

⁸ Chandler, A. D. Jr and Salsbury, S., Pierre S. Du Pont and the Making of the Modern Corporation (New York: Harper and Row, 1971); andGoogle ScholarChandler, A. D. JrStrategy and Structure (Cambridge, Massachusetts, 1962).Google Scholar

⁹ Jenkins, R. V., Images and Enterprise (Baltimore, 1975).Google Scholar

¹⁰ Reich, L. S., “Research, Patents, and the Struggle to Control Radio: A Study of Big Business and the Uses of Industrial Research,” Business History Review, 51 (Summer 1977), 208–35;CrossRefGoogle Scholaridem., “Industrial Research and the Pursuit of Corporate Security: The Early Years of Bell Labs,” Business History Review, 54 (Winter 1980), 504–29; Brock, G. W., The Telecommunications Idustry (Cambridge, Massachusetts, 1981).Google Scholar

¹¹ Chandler, A. D. JrThe Visible Hand (Cambridge, Massachusetts, 1977).Google Scholar

¹² National Research Council, Bulletin #16, “Research Laboratories in Industrial Establishments of the United States, Including Consulting Laboratories” (Washington, D.C., 1927);Google Scholaridem., Bulletin #60, “Industrial Research Laboratories of the United States” (Washington, D.C., 1933); idem., Bulletin #91, “Industrial Research Laboratories of the United States,” (Washington, D.C., 1940); idem., Bulletin #104, “Industrial Research Laboratories of the United States” (Washington, D.C., 1946); idem., Bulletin #113, “Industrial Research Laboratories of the United States” (Washington, D.C., 1946).

¹³ The listings of firms were assembled by Mr. Takashin Hikino under the direction of Professor Alfred Chandler of the Harvard business School. I am grateful to Professor Chandler for making these tabulations available.Google Scholar

¹⁴ Use of the procedure developed by Heckman to correct for sample bias within this small-firm population was not possible due to the lack of data on other characteristics of firms inthe smaple. See Heckman, J., “Sample Selection Bias and Specification Error” Econometrica, 47 (01 1979), 153–62.CrossRefGoogle Scholar

¹⁵ National Research Council, Bulletin #2, “Research Laboiratories in Industrial Establishments of the United States of America” (Washington, D.C., 1920), p. 47.Google Scholar

¹⁶ This argument draws upon Mowery, D. C., “The Relationship Between Intrafirm and Contractual Forms of Industrial Research in American Manufacturing, 1900–1940,” forthcoming, Explorations in Economic History.Google Scholar

¹⁷ As Table I indicates, the share of total research employment within manufacturing accounted for by the 200 largest firms ranged from nearly 44 percent to 50 percent during 1921–1946. The share of total manufacturing activity accounted for by the 200 largest firms after 1929 may be measured by simply summing the book value of assets across firms and employing data from the relevant volumes of the Statistics of Income. This technique was employed by Collins, N. R. and Preston, L. E., “The Size Structure of the Largest Industrial Firms, 1909–1958,” American Economic Review, 51 (12 1961), 986–1011, where the authors note that the resulting estimate is likely to be low.Google ScholarIn 1933 the firms in Data Set B accounted for at least 38 percent of the total assets of manufacturing firms filing corporate income tax returns. In 1946, the firms in Data Set C accounted for 50.7 percent of total manufacturing assets. Comparable data on manufacturing assets are not available in the Statistics of Income prior to 1929.Google Scholar

¹⁸ Research intensity, defined as scientific research professionals per million dollars of assets, also grows substantially during 1921–33. This apparently dramatic growth trend is somewhat spurious, however, because of the impact of the Depression on the book value of firm assets.Google Scholar

¹⁹ “See Hamberg, “Size of Firm, Market Structure, and Innovation,” and Comanor, “Market Structure, Product Differentiation, and Industrial Research.”Google Scholar

²⁰ The use of ordinary least squares in this situation reduces the slope of the “true” relationship between research employment and firm size; large values of the independent variables will underpredict the dependent variable. The logarithmic transformation of zeroes in these data also creates obvious difficulties. Ordinary least squares regression is employed to analyze the small-firm data on firm size and research employment below, since all observations for research employment in that data set are nonzero.Google Scholar

²¹ See Tobin, J., “Estimation of Relationships for Limited Dependent Variables,” Econometrica, 26 (01 1958), 24–36, as well asGoogle ScholarGoldberger, A. S., Econometric Theory (New York, 1964), pp. 251–55.Google Scholar

²² Slope and intercept dummy variables originally were inserted for all industries with more than eight observations. As none but those for chemicals and primary metals proved to be consistently significantly different form zero, or contributed significantly to the explainmed variance, they were omitted from the specification whose results are reported in Table 2.Google Scholar

²³ See Scherer, “Firm Size, Market Structure, and the Output of Patented Inventions,” Hamberg, “Size, Market Structure, and the Output of Patented Inventions,” Hamberg, “Size of Firm, Market Structure, and Innovation,” and Comanor, “Market Structure, Product Differentiation, and Industrial Research,” among others.Google Scholar

²⁴ To test for a possibly nonlinear relationship between research employment and firm size in the tobit, a squared assets term was inserted into the specification. The coefficient for this variable was negative and significant (at the .10 level) only in 1946, indicating that the firm size-research employment relationship was roughly linear in 1921 and 1933. By 1946, however, the relative research intensity of the largest firms in the top 200 had declined substantially.Google Scholar

²⁵ The likelihood ratio test utilizes the difference in the logarithm of the likelihood function obtained with only a constant term from that obtained with the insertion of the explanatory variables. Minus twice this difference is distributed chi-square.Google Scholar

²⁶ The threshold firm size is computed by solving the estimated tobit equation for zero.Google Scholar

²⁷ This threshold is substantially above the threshold for the existence of an in-house research facility reported in Mowery, “The Relationship Between Intrafirm and Contractual Forms of Research.… “The difference reflects the fact that the sample of client firms of independent research organizations analyzed in that paper is comprised exclusively of firms actively engaged in the performance of research.Google Scholar

²⁸ See A. D. Chandler Jr., Strategy and Structure, as well as idem., The Visible Hand.

²⁹ The elasticities are computed by use of the following formula:

where Ŷ_t, is the predicted value of the dependent variable, Î_t is the index estimated by the tobit, σ^_t is the standard error of the equation, F(Î_t/σ^_t) is the value of the standard normal cumulative distribution at (Î_t/σ^_t), and f(Î_t/σ^_t) is the value of the standard normal density function at (Î_t/σ^_t). See Goldberger, Econometric Theory, p. 252.Google Scholar

³⁰ See Mansfield, E., “Industrial Research and Development Expenditures: Determinants, Prospects, and Relation to Size of Firm and Inventive Output,” Journal of Political Economy, 72 (08 1964), 319–40, andCrossRefGoogle ScholarHamberg, D., “Size of Firm, Oligopoly, and Research: The Evidence,” Canadian Journal of Economics and Political Science, 30 (02 1964), 62–75.CrossRefGoogle Scholar

³¹ Firms in the interwar American chemicals industry participated in a number of international cartel agreements, which could have facilitated collusion on price among domestic competitors. See Haber, L. F., The Chemical Industry 1900–1930 (Oxford, 1971), andGoogle ScholarReader, W. J., Imperial Chemical Industries: A History, vol. 2 (Oxford, 1975), for discussions of the interwar cartel agreements.Google Scholar

³² See Rosenberg, N., “Technological Change in the Machine Tool Industry, 1840–1910,” this JOURNAL, 23 (12 1963), 414–46, for this description of the machine tool industry. Haynes's rather florid description of the impact of the new chemicals industry processes and products on other industries is worth noting: “Manufacturers of textiles, rubber goods, coatings, fertilizers, insecticides, metaiwares, medicines, and a host of other consumer goods were severely jolted by the new chemical materials and chemical methods, but nobody then [in the 1920s] appreciated that a profoundly disturbing, powerfully stimulating element had been thrust into the American industrial system. Within the next twenty years the remotest corner of American industry was to feel the impact of these chemical repercussions.”Google ScholarHaynes, W., American Chemical Industry, vol. 4, The Merger Era (New York, 1948), p. 430.Google Scholar

³³ This conclusion is based on a comparison of the results of a tobit specification similar to that in Table 2 for survivor firms (those remaining in the top 200) and new entrants during 1921–1933.Google Scholar

³⁴ The mean size of new entrants into the 1933 top 200 was $55.7 million in assets, with mean research employment of 15 persons. Firms remaining in the top 200 during 1921–1933 with research laboratories, on the other hand, had a mean size of $155 million in assets and a mean research employment level of 71 persons. The research intensity of new entrants thus was roughly one-half that of survivor firms with research laboratories during 1921–1933.Google Scholar

³⁵ New entrants to the 1946 top 200 display levels of research intensity greater than those of firms remaining in the top 200 during 1933–1946. New entrants had a mean firm size of $118 million in assets, and mean research employment of 101 persons. This research intensity of .86 compares with a mean research intensity of 55 among firms with research facilities remaining in the top 200 during 1933–1946, based on a mean firm size of $250 million and mean research employment of 135 professionals.Google Scholar

³⁶ Among firms within the 1921 top 200 that are in existence throughout 1921–1946, those that remain in the top 200 in 1933 and 1946 display patterns of research employment and (not surprisingly) firm growth that contrast sharply with the behavior of these attributes among firms that do not remain among the ranks of the 200 largest firms. During both 1921–1933 and 1933–1946, the mean levels of research employment and overall research intensity among firms leaving the top 200 are lower than is true for firms remaining within the large-firm group. Results are available from the author.Google Scholar

³⁷ This treatment of mergers among the largest firms, which was done for analytic convenience, does not affect the results of the logit. Inclusion of merged firms as survivors within the data yields coefficient estimates of similar size and statistical significance. In the chemicals industry during 1921–1933, the presence of the Allied Chemicals merger results in the chemicals industry dummy variable acquiring a somewhat larger standard error when mergers are included. The sign of the variable's coefficient is unchanged, however, and it remains significant at the .05 level.Google Scholar

³⁸ In 1921, 5 of the 17 chemicals firms in the top 200 had no research facilities. By 1933, of 14 chemicals firms in the top 200, none were without research facilities.Google Scholar

³⁹ See Edwards, R. C., “Stages in Corporate Stability and Risks of Corporate Failure,” this JOURNAL, 35 (06 1975), 428–57;Google ScholarCollins, and Preston, , “The Size Structure of the Largest Industrial Firms, 1909–1958”; and Kaplan, A. D. H., Big Business in a Competitive System (Washington, D.C., 1964).Google Scholar

⁴⁰ Because smaller-firm observations are restricted to firms with research laboratories, it is infeasible to merge the large- and smaller-firm data sets. The results of regressions for merged datasets consisting of the large and small firms with in-house laboratories for 1921, 1933, and 1946 display coefficients for assets that are moderately larger in size, but still significantly less than one. Changes in the assets coefficients during 1921–1946 for this merged data set are identical in direction and magnitude to those observed in Table 4.Google Scholar

⁴¹ The omission of firms without research laboratories from the dataset analyzed in Table 2 affects the size of the coefficients, as was hypothesized, but does not affect the direction or significance of changes over time in these coefficients. Results are available from the author.Google Scholar

⁴² In view of the marginal statistical significance of the coefficient for the chemicals slope dummy variable, this conclusion for 1921 is subject to strong qualification.Google Scholar

⁴³ Again, give the limited statistical significance of the coefficients for the primary metals slope dummy variable, this conclusion must be qualified.Google Scholar

⁴⁴ The firms establishing new laboratories tended on average to be significantly smaller than survivor firms in this population.Google Scholar

⁴⁵ The elasticity research employment with respect to firm size exhibits no significant change within the nonchemicals survivor group during 1921–1933.

⁴⁶ Empirical support for this “Law” is mixed.Google ScholarMansfield, E., “Entry, Gibrat's Law, Innovation, and the Growth of Firms,” American Economic Review, 52 (12 1962), 1022–51, found no support for the law.Google ScholarScherer, F. M., Industrial Market Structure and Economic Performance, 2d ed. (Chicago, 1980), pp. 146–49, reviews a number of studies on the size distribution of firms that are consistent with the operations of Gibrat's Law.Google Scholar

⁴⁷ Clearly, such a measure of growth isa highly imperfect one, responding to shifts in investment and accounting policies that may produce spurious growth trends. But there exists no strong a priori basis for the belief that such interfirm differences are distributed among this population in a nonrandorn fashion.Google Scholar

⁴⁸ In these specifications, research employment is assumed to be exogenous; i.e., possible simultaneity between growth in firm size and growth in research employment is ignored. Data limitations preclude the estimation of a simultaneous-equation model.Google Scholar

⁴⁹ Regressions for firm growth during 1933–1946 also were run with variables measuring 1921 levels of research intensity and firm size, in order to examine possibly longer lags in the returns to research employment. The additional variables failed to acquire statistically significant coefficients, and did not contribute significantly to the explained variance. The implication is that most of the growth-based returns to research investment were captured within a 13-year period. This finding is consistent with the results in Leonard, W. N., “Research and Development in Industrial Growth,” Journal of Political Economy, 79 (03/04 1971), 232–56.CrossRefGoogle Scholar

⁵⁰ See Chandler, The Visible Hand.Google Scholar

⁵¹ Federal research contracts with private and nonprofit organizations grew from $38 million in 1938 to $706 million in 1944. The Senate Military Affairs Committee's Subcommittee on War Mobilization noted in 1945 that “…of nearly 2,000 industrial organizations receiving a total of almost $1,000,000,000 in research contracts for the Government during 1940–1942, 100 firms accounted for more than half of the total.” The Government's Wartime Research and Development (Washington, D.C., 1945), p. 22.Google Scholar

⁵² During 1921–1927, employment of research professionals in industrial laboratories grew at an average annual rate of 12.1 percent. During 1927–1933, employment grew at an average annual rate of 9.5 percent. During 1933–1940, this growth rate stood at roughly 14.3 percent. See Mowery, D. C., “The Emergence and Growth of Industrial Research in American Manufacturing, 1899–1946,” unpublished Ph.D. dissertation, Stanford University, Ch. 2.Google Scholar

⁵³ This conclusion differs with that of a 1946 report by the Smaller War Plants Corporation to the Senate Special Committee on the Problems of Small Business: “… only 68 top corporations received two-thirds of the value of Federal research and development contracts and the top 10 corporations received nearly two-fifths of these contracts. …Obviously the companies in whose laboratories this research work has been carried on will be its chief beneficiaries not only because of their direct acquaintanceship and knowledge of the research but also because of patents. The investigations of the Subcommittee on War Mobilization of the Senate Military Affairs Committee show that over 90 percent of the contracts made between Government agencies and private industrial laboratories for scientific research and development placed the ownership of patents with the contractors, the Government receiving a royalty-free license for its own use.” See Economic Concentration and World War II, Report of the Smaller War Plants Corporation to the Senate Special Committee on the Problems of Small Business (Washington, D.C., 1946), p. 53. The contrasting assessment of the impact on industrial research of World War II in this paper reflects several differences in focus. The Smaller War Plants Corporation's analysis was concerned primarily with the future impact of Federal wartime R&D funding, a topic not covered in this paper. It is also the case that the government report once again makes no attempt to analyze the relationship between firm size and research activity in concluding that the largest firms are destined to dominate research activity.Google Scholar

⁵⁴ N. Rosenberg, “How Exogenous is Science?”, in idem., Inside the Black Box: Technology and Economics (Cambridge, 1982).

⁵⁵ Gort, M. and Klepper, S., “Time Paths in the Diffusion of Innovations,“ Economic Journal, 92 (09 1982), 630–53.CrossRefGoogle ScholarGort and Klepper note that important influences on the probability of entry into an industry include the “ways in which returns can be maximized on a component of organisation capital, namely information on new product technology. We distinguish organisation capital from human capital in that the returns to the latter can be appropriated by the individual employees who possess such capital. In contrast, organisation capital belongs to the firm either because it has legal title to it, as in the case of a patent, or because it depends upon the interdependent actions or information of more than one employee” (p. 632).Google ScholarSimilar definitions of organization capital can be found in Prescott, E. and Visscher, M., “Organization Capital,” Journal of Political Economy, 88 (06 1980), 446–61, and inCrossRefGoogle ScholarNelson, R. R. and Winter, S., An Evolutionary Theory of Economic Change (Cambridge, 1982).Google Scholar

⁵⁶ This hypothesized impact of Federally supported research on market structure reflects the fact that the fruits of such research often may be less appropriable by an individual firm than are the results of privately funded R&D, due to regulations affecting the assignment to the government and mandatory licensing of patents and other results of publicly funded R&D. This analysis in turn suggests that Federal and private funds are complements, rather than substitutes, in the research expenditures of the firm. Recent empirical work supports such a conclusion. See Link, A. N., “An Analysis of the Composition of R&D Spending,” Southern Economic Journal, 49 (10 1982), 342–49;CrossRefGoogle ScholarScott, J. T., “Firm Versus Industry Variability in R&D Intensity” R. C. Levin and P. Reiss, “Tests of a Schumpeterian Model of R&D and Market Structure” E. Mansfield, “R&D and Innovation: Some Empirical Findings”; and D. Levy and N. Terleckyj, “Government-Financed R&D and Productivity Growth: Macroeconomic Evidence,”Google Scholar all in Griliches, Z., ed., R&D, Patents, and Productivity (National Bureau of Economic Research: forthcoming).Google Scholar