How important to economic growth in advanced countries is the accumulation of physical capital? My short answer is that increased capital is one of several important sources of output growth. This appraisal, which is amply supported by research results, should surprise no one-but doubtless it will. Capital is not the source of growth despite the contrary view common in financial circles and on Capitol Hill. Output, and therefore growth, are governed by many determinants. Moreover, there is no substance to the recurring notion that the United States' growth rate would have matched those of countries like Germany or Japan if only we invested as much as they did. Finally, it is quite wrong to blame investment for the recent sharp reduction in the growth of American output and productivity, or to suppose that merely raising investment would go far toward restoring the old growth rate of productivity. Why many people share a vision of growth that assigns exclusive attention to capital I do not know. Growth models that feature capital, while assigning other output determinants to ceteris paribus, may be partly responsible, but it should have been apparent to all that such models were meant only to illuminate a relationship, not to describe a whole economic system. Analyses based on incremental capital-output ratios may have contributed to the illusion. If so, as Robert Solow says of these ratios, Economists have a responsibility to do better. To deny that capital is everything is not to imply that it is nothing. I do not share the other extreme view, sometimes encountered, that capital can be ignored because its significance is hard to establish if one fits a production function by correlation analysis. I stress again: capital is an important growth source. It has sometimes contributed importantly to differences in growth rates between periods and places. More capital formation would raise the growth rate. The contribution of capital to growth is evaluated best in the context of a complete analysis of the sources of growth. The following summary, drawn from my Accounting for Slower Economic Growth, refers to total potential national income originating in nonresidential business. During the period from 1948 to 1973 the growth rate of this series was 3.8 percent per year. Of that amount, 15 percent resulted from more capital, that is, more nonresidential structures and equipment and more inventories. Another 15 percent is ascribed to changes in employment and working hours, with account also taken of the age-sex composition of workers. Fourteen percent was due to increased capabilities of workers resulting from more education. Ten percent resulted from improved resource allocation, taking the form of a reduction in the amount of labor overallocated to farming and to self-employment and unpaid family labor in nonfarm establishments too small for efficiency. Thirty-seven percent was contributed by advances in technological, managerial, and organizational knowledge as to how to produce at low cost, together with miscellaneous output determinants not separately estimated. This is the residual in the calculation. In the 1948-73 period it probably provides a tolerable approximation to the contribution of advances in knowledge alone. If so, advances in knowledge were much the largest single source of growth. Economies of scale made possible by the growth of markets contributed an estimated 11 percent of the growth rate. Finally, certain changes in the legal and human environment, together with irregular factors, subtracted 2 percent. *Associate Director for National Economic Accounts, Bureau of Economic Analysis.

David Schwartzman argues in a recent issue of this Review [5] that the estimate g,iven in my Sources of Economic Growth [1] for the rise in the quality of labor due to additional education was far too big. His comment appeared nine months after, but was written before, publication of my Why Growth Rates Differ [3]; I shall also refer to that book in order to bring the discussion up to date.' I must first correct Schwartzman's summary of my substantive results. He reports [5, p. 508] that in Sources ... . Denison ascribes over three-fifths of the growth in ouitput per man-hour to increased My estimates actually implied 29 percent: .67 percentage points [1, Table 33] out of a growth rate of outpuit per man-hour in the whole economy of 2.34 percent.2 In Why Growth Rates Dier, my procedures were altered to reduce the effect of changes in days of school attended per school year and, with more data available, to incorporate some refinements; for the 1950-62 period they yielded .49 percentage points as the contribution of education [3, Table 21-1]. If applied in the 1929-57 period, the change in my allowance for school days would cut my estimate of .67 percentage points to about .5, or 21-22 percent of the growth of output per man-hour. In discussing Schwartzman's specific points, I shall usually refer to growth rates of educational quality indexes for civilian labor. These exceed contributions of increased education of the labor force to growth rates of national income because 1), civilian labor represents only a fraction of total factor input, and 2), in some activities the change in output is measured by employment or manhours so a change in labor quality cannot affect measured output. My educational quality index for civilians grew at a rate of .75 in 1950-62 [3, Table 8-5, column 4] while the contribution was .49 percentage points. Thus, any change in the civilian quality index would alter the contribution by about one-third less. My quality index is constructed in two stages. First, I compute an index based only on changes in the distribution of the labor force by years of education. Second, I allow for increased school attendance per school year. Schwartzman does the same. Growth rates of civilian quality indexes compare as follows:

An article by Michael D. Intriligator entitled Embodied Technical Change and Productivity in the United States 1929-1958 appeared in the February 1965 issue of this REVIEW. To judge the importance of embodied and disembodied technical progress, Intriligator computed 40 correlations of gross national product, excluding GNP originating in government and housing, with capital and labor input. All 40 regressions resulted from combining four alternative measures of labor input, three of them mine, with five alternative capital series based on different assumptions about embodiment, and two methods of handling trend. To match Intriligator's GNP measure, labor input must exclude all general government employees. The data Intriligator actually used, however, include all government employment: civilian, military, and work relief. The period covered, 19291958, spanned the Great Depression, World War II, and the Korean War, periods in which the share of government employment was fluctuating violently, as well as a postwar period in which most of the employment increase was devoted to government. Intriligator's results consequently seem to have no apparent meaning. An interesting feature of the article is that all 40 regressions Intriligator computed have high correlation coefficients despite his use of irrelevant data; only one is below 0.95 and most are above 0.98. This may give further reason for skepticism about the utility of inferring causation -in this case, degree of embodiment-from minor differences in the goodness of fit of time series.

M OST descriptions of the behavior of private saving have dealt separately with net personal saving, net corporate saving, or capital consumption allowances. However, for many analytical purposes, and particularly the appraisal of inflationary or deflationary pressures, it is gross private saving as a whole the sum of these three components -that is crucial, and analysis of the details is of interest mainly in arriving at the implications for this total. Should it appear that total gross private saving bears a more stable relationship to key income or output measures than do its components, analysis could be both simplified and improved by dealing with this total directly. This is, in fact, what the data for past periods show. I recognize that this conclusion is contrary to the common a priori expectation, that it may not hold in the future, and that in retrospect it may appear as merely a statistical oddity. But the record of past experience seems to me clear enough to demand exploration. In the Survey of Current Business for January I955 1 I pointed out that the ratio of gross private saving to gross national product was about the same in I929 and each of the years from I948 through I953. This ratio continued to hold in I954, I955, and, on the basis of preliminary estimates, in I956. In these ten relatively prosperous peacetime years-all those for which estimates are available except for I946 and I947, when the saving rate clearly was sizably distorted by war-time influences gross private saving averaged I4.63 per cent of GNP, and never deviated more than onehalf percentage point from the average.2 The article cited also pointed out that in the 1930-4I period of substantial underutilization of resources, when saving fluctuated much more than GNP, a formula taking account of the difference between current-year GNP and GNP in the most recent year of substantially full employment would satisfactorily describe the saving pattern.3 As a result, gross private saving for the entire I929-56 period, excluding I942-47, is quite satisfactorily described by the following simple formula: Gross private saving = I463 Po .27 (POP.) where PO is the gross national product in the most recent high-employment year (including the current year) and P1 is gross national product in the current year. For high employment years, PO and P1 are the same, and we have simply gross private saving equals I4.63% of gross national product. The gross private saving estimate calculated from this formula for each of the 22 years of the period is compared with reported gross private saving in Table I.4 The mean deviation is $0.7 billion, or 3.I per cent of the average level of reported gross private saving. This is probably within the margin of error of the reported data. Indeed, reported gross private saving is more closely approximated by this formula than by the alternative estimate of gross private saving that can be derived from the national accounts statistics by deducting the government surplus on income and product transactions from gross investment. The difference between the two estimates, which is equal to the statistical discrepancy in the national accounts, averaged $I.o billion. For the ten high-employment years included in the period with which this note is primarily concerned, saving calculated as I4.63 per cent of GNP differs from reported saving by an average of $o.8 billion, or I.9 per cent, and the statistical discrepancy averages $I.3 billion.

Edward F. Denison, Capital Formation and the Recent Productivity Slowdown: Discussion, The Journal of Finance, Vol. 33, No. 3, Papers and Proceedings of the Thirty-Sixth Annual Meeting American Finance Association, New York City December 28-30, 1977 (Jun., 1978), pp. 1006-1010