On Concepts and Measures of Changes in Productivity Author(s): Luigi L.

Pasinetti Reviewed work(s): Source: The Review of Economics and Statistics, Vol. 41, No. 3 (Aug., 1959), pp. 270-286 Published by: The MIT Press Stable URL: http://www.jstor.org/stable/1927453 . Accessed: 23/03/2012 14:06
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to The Review of Economics and Statistics.

http://www.jstor.org

ON CONCEPTS AND MEASURES OF CHANGES IN PRODUCTIVITY

Luigi L. Pasinetti *
T-flECHNICAL progress has received unprecedented attention by economists in the last few decades, but more because statistical evidence has imposed the subject on them than as a spontaneous development of economic thought. Indeed, the whole neo-classical movement and the increasing modern application of mathematics, which have contributed so much to improving the tools of economic analysis and to conferring rigor and definiteness on economic thought, have preferred to leave technical progress aside. The trouble is that technical changes are hard variables to deal with in analytical terms. The statistical data which have appeared in the last few decades have come as quite a surprise to economists, because their theories, from the Ricardian to the neo-classical, did not lead them to expect such results. I refer in particularto some statistical evidence for the United States and for other "capitalistic" countries which has shown that, on the average and in the long run, shares of labor and capital in the national product have not changed very much, that wages have not remainedat the subsistence level but have risen in proportion to national income, that capital per man has indeed increased but output per man has also increased in proportion, that the rate of remuneration of capital has remained almost constant. In the absence of an economic theory giving a straightforward interpretation of all these outcomes the attention of economists has been called back more and more to changes in technology; and economic statisticians, in trying to evaluate these changes, have followed the easiest way: they have normally taken ratios of production to man-hours (labor productivities) and computedtheir changes through time. The procedure is very useful for many pur* The author is grateful to Professors Duesenberry, Kaysen, Conrad, and Orcutt, who read the paper and made comments and criticisms. He is especially grateful to Professors Samuelson and Modigliani, whose sharp criticisms enabled him to introduce many improvements. Of course, the author alone is responsible for what he has written.
[270]

poses but, among other limitations for example, the impossibility of taking into account qualitative improvements- it has the major defect of referring only to labor, while the production process involves as well other factors of production whose productivity might change in a different way. This leads to different conclusions as to the productivity of the system as a whole. There have been some attempts by economists to complete these evaluations and to introduce capital into the picture, by making use of theoretical notions like the production function, but these attempts - in the writer's opinion - have neglected an important characteristic of capital - that it is reproducible and that its process of productionis also subject to technical change. It is my purpose in this paper to go into these problems. I shall try to give a short theoretical economic interpretation of technical change and suggest a procedure for evaluating it, with respect to all factors of production. When particularstudies about qualitative improvements are available they may be incorporatedin the same framework. Traditional to TechnicalChange Approach Let us try to give an economic interpretation of technological change along traditional lines. We have been accustomed to think of output as the outcome of a combinationof factors of production (traditionally reduced to three: land, labor, and capital). Let us consider for the time being two factors, labor and capital. If, at a certain point of time, we observe an actual combination of them and a resulting output, we consider that particular combination one of many which are technically possible for that particular output and which can be graphically represented by a region delimited by an isoquant of a production function. If the observed combination is the optimum one, it must lie on the isoquant at the point where the marginal rates of substitution among factors equal the

CONCEPTS AND MEASURES OF CHANGES IN PRODUCTIVITY ratios of their costs. The production function, however, refers to a given state of technical knowledge, and since technical knowledge changes through time, the consequence is that the production function also shifts through time. If, therefore, at a later point of time we observe another combinationof labor and capital, we are accustomed to interpret it in the same theoretical framework as before, but we consider it as coming from a different production function. The problem we are facing here is to compare the two production functions and to give an evaluation of changes which have intervened. This is certainly a formidabletask. For meaningful comparisons, it is necessary first of all to compare homogeneous quantities; and this, throughtime, is practically an insoluble problem. However, economists have followed conventions and accepted approximations in this field, and I shall not dwell on them here, but only mention their major limitations. The proper units of measurementought to be physical quantities. For labor, this means considering all workers as alike (and similarly for land, considering all acres as alike). This represents the sort of assumption that, after proper definitions of unit equivalents, economists have always worked with. Much more complicated is the problem of measuring output and capital. Here physical items are extremely heterogeneous and their comparison impossible. The only commonunit of measurewe can find on the market for all outputs and capital is the monetary unit. The procedurethat has thereforebeen followed is to use current values and to reduce them into real quantities by a price index correction. The procedure carries with it the well known "index number problem," connected with the arbitrariness, in making aggregations, of using prices of the beginning or of the end period for weighing the quantities and of using whatever criterion may be assumed for weighing new products as against old ones. These difficulities are more serious for large aggregates and become less important as we deal with more and more disaggregated quantities. There is another more subtle limitation involved in these comparisons which is connected with the fact that the unit of measurement with which capital is measured is itself not inde-

27I

pendent of the rate of profit.' There are however many ways in which this complication vanishes, and one of them is when the rate of profit does not vary over time. With these qualifications let us reckon labor (L) in man-years, output (Q) in number of pieces (one piece being equal to that quantity which was sold in the base year at $i million), and capital (K) in, say, special tons of "'steel" (one ton being equal in this case to that quantity which was bought in the base year at $i million). Now, given at time t a production function Q = f (L,K) and at time t + 0 another production function Q = O(L,K) and supposing that we have a way of establishing a one-to-one correspondence between each point of the first and each point of the second production function, we have an infinite number of couples of points, and therefore the possibility of making comparisons within each couple. Suppose for example that, as a way of making a one-to-one correspondence, we have chosen to relate the quantities produced on the two production functions at the same combinationsof labor and capital. We can say, for each pair of points, that, between t and t+O, there has been a "neutral" improvement if marginal productivities of labor and capital have increased in the same proportions, that there has been a "labor-saving" improvement if the marginal productivity of capital has increased more than the marginalproductivity of labor, and that there has been a "capital-saving" improvement if the marginal productivity of labor has increased more than the marginal productivity of capital. The reader will recognize that these are nothing but Hicks's notions of "neutral," "labor-saving,"and "capital-saving" inventions.2 When we come to actual observations, comparisons become somewhat more restricted, because what we can observe are not entire production functions but only actual combinations of factors. In the traditional framework this means we have no unique way of distinguishing between a change of the production function
' On this point, Mrs. Robinson has recently particularly insisted. Joan Robinson, The Accumulation of Capital (London, I956). 2'J R. Hicks, The Theory of Wages (reprinted, New York, I948) I2I-22.

272

THE REVIEW OF ECONOMICS AND STATISTICS the conclusion that capital intensity in that period increased4). So much for the traditional interpretationof technical change. I shall add only that the analysis has been carried on so far in terms of labor and capital, but it could be extended along the same lines to land. A MoreComplete Analysis It is my contention that the foregoing analysis, although satisfactory for land and labor, is still incomplete with respect to capital, because it does not take into account the characteristic that capital is not something which comes from heaven, or from outside, but from the productionprocess itself: it is reproducible. In the background of the production function we have dealt with in the previous section, and which we come now to identify as a production function of a consumptiongoods industry, there is another production function - the production function of the capital goods industry. We have, therefore, to push our analysis further because technical change may equally well take place in the capital goods industry. In this case the same quantity of physical capital the same quantity of tons of steel of capital, in our picturesqueexpression,has a differentmeaning at time t and at time t+O if at t+0 the same quantity of steel can be obtained from smaller quantities of labor and capital. We could start our analysis again for the capital goods industry in the same way that we did for the consumption goods industry. Howeverand this is the important point - whatever form technical change may take in the capital goods industry - whetherneutral, labor-saving, or capital-saving- its effect is always capital saving for the consumption goods industry. Let us start with a simplified case. Suppose that capital goods require labor but no capital to be produced. Moreover, suppose that capital, once made, lasts forever. From the previous analysis, we could get the remuneration of capital in absolute terms but not its rate of re4'Owing to his assumptions about the production function, Solow splits the ioO per cent increase in output in two parts: 87.5 per cent is attributed to a shift of the production function - neutral by assumption in Hicks's terms - and I2.5 per cent to an increase in capital intensity (a movement along the same production function).

and a movement along the same production function. If we want to make this distinction, we have to introduce some assumptions-for example to assume particular types of production functions and, as before, a particular way of relating them to each other. I prefer, however, not to introduce such assumptions and rather to give up the above distinction altogether, for reasons which will appear in the course of our analysis. If therefore a combination (Q,K,L) is observed at time t and a combination (Q', K', L') at time t + 0, an evaluation of the intervening technical change can be given by comparing the two ratios Q/L and Q/K. If both of them have changed in the same proportion, we can say that the change has been neutral. In this case either one of the two ratios taken alone (for example Q/L) can be used to evaluate the intervening increase in productivity. As a matter of fact, this is the way by measuring Q/L - in which changes in productivity are usually measured. But this is a complete measure only if Q/K changes in the same proportion. If Q/K does not change in the same proportion, a more complete evaluation requires a distinction of the change in at least two parts: a neutral effect given by the proportional change in that one of the two ratios which has changed the least, and a laborsaving effect - or alternatively a capital-saving effect - given by the excess of the proportional change of Q/L over Q/K - or alternatively of Q/K over Q/L. It is roughly along these lines - in this second more complete way-that R. M. Solow3 has recently made a study of technical change in the United States economy between I909 and I949, except that he has preferred to make the distinction I have given up (he has assumed a production function which is linear and homogeneous and which shifts by a multiplicative term which is a function of time). His findings- interpretedin the light of the above analysis-show increases of about ioo per cent in the output-labor ratio and of about 6o per cent in the output-capital ratio (in physical terms), between Igog and I949; which lead to
8 R. M. Solow, "Technical Change and the Aggregate Production Function," this REVIEW, xxXIX (August I957).

CONCEPTS AND MEASURES OF CHANGES IN PRODUCTIVITY muneration. The introduction now of a production process for capital brings into the analysis a cost of production of capital and therefore determines a rate of profit. This is on the supply side of the market. On the demandside, capital is demanded not as such but for the results of its possible combinations with labor, and the market will determine the value of capital through capitalization of its expected remuneration. In conditions of equilibrium, wage rates and capital remuneration will be such as to make the cost of capital equal to its market evaluation.5 Going back now to the analysis of the previous section and wishing to compare two equilibrium situations in time we have to consider in addition to Q/L and Q/K some other quantities reflecting changes in productivity in the capital goods industry. In our simple case, the quantity to be considered can only be K/N, where N is the quantity of labor which would be necessary for reproducingthe stock of capital. Here some complicationsarise because cap5In analytical terms we can write, at a certain point of time:
()

273

Q=f(K,L)

(2)

W = a-

(3) R-=7

where W = wage rate and R = remuneration of capital, both in terms of final goods. This is the system we get from the previous section. By introducing now a process for the production of capital goods, and by calling a the capital goods (in physical terms) produced by one unit of labor, we have

(4) K = aN
where N = units of labor (man-years) necessary for reproducing the existent stock of capital. The cost of reproduction (II) of the existing capital stock will be II = NW. On the other hand, given on the market a rate of interest r which expresses tastes and intertemporal preferences, the market evaluation of capital will be V = RK/r. Condition of equilibrium (stationary equilibrium) is RK (5) NW= r r Equations (I)-(5) make up a system of five equations in which we can consider as given: a (by the state of technical knowledge), L (by the existent labor force), and r (by intertemporal preferences of people). Taking r as given here is a shortcut. We could of course explain it by introducing another equation embodying some particular theory of demand-for example a Modigliani-Brumberg type of consumption function. Now, the system determines Q,N, R, W, and the equilibrium value of capital (K). If the existing stock of capital is less than its equilibrium quantity, its market value will be higher than its cost. Entrepreneurs will shift some labor from the consumption goods industry to capital goods production until equilibrium is re-established. In real terms this means that there will be some savings in the system until the capital stock is brought up to its equilibrium amount.

ital might have been made in the past under very different conditions, but we can dispose of the difficulty in a very useful way. A "real" measure of capital has lost its importance at this point of our analysis, and it is more useful to re-define capital in terms of capacity (C). In this way the ratio Q/K becomes Q/C, which is always equal to one,6 while N can be interpreted as the quantity of labor which would be needed for reproducing the existent capacity, with the technique available at the time observations are made. Therefore we are left with only two ratios, Q/L and C/N, the first of which represents the labor productivity - in the consumption goods industry - and the second the over-all capital productivity - over-all, because it embraces both the consumption and the capital goods industries. At this point, the definitions of section i can be modifiedin a more complete sense. A change through time of Q/L can be assumed by itself alone to be an indication of change in productivity only if C/N changes in the same proportion. If C/N does not change in the same proportion at least two parts of the change have to be distinguished-a neutral effect equal to the proportional change of that ratio which has changed the least, and a labor-saving effect - or alternatively a capital saving effect - given by the excess of the proportional change of Q/L over C/N - or alternatively of C/N over Q/L. A simple index of directionwhich in time will decrease, increase, or remain constant according to whether technical change is capital saving, labor saving, or neutral-of technical change, without a quantification, is obviously given by Q/L C/N AnotherInterpretation The conclusions of the previous section can be looked at from a different,more illuminating point of view. Let us use output as the num&raire in our system and, therefore, put its price (Pq) equal to one. Suppose now that at time t we are in equilibrium. The market value of capital is equal to its cost
8 Remember that we are dealing here in a traditional, neo-classical framework, where capacity is always fully utilized.

274

THE REVIEW OF ECONOMICS AND STATISTICS

PC = N

where P0 = price of one unit of capacity; W = wage-rate. By calling N/C = ak (input of labor in one unit of capacity), we can write
P,C
= akWC.

(8)

The market value of output can be expressed, in terms of costs, as the remunerationof capital plus the remunerationof labor7
QDQKDQL eLL ZL D) pCC DQ K + aQ Q= L DaK P0C

Q= aK K+

(9)

L-.
=

Q'

Let us now call

r (rate of

profit); DQ/DL = W (wage rate) and L/Q = aq

(input of labor in one unit of output). We have Q = r PC + aqWQ and remembering that P0
= akW from (8) and C = Q by definition we

obtain

Q = (rak + aq) WQ. Let us now take the ratio between capital and output
PqQ
P0C ak_WQ (rak+ aq) WQ
ak

Harrod's notions are broader and more comprehensive than those of Hicks. Of course, changes in the capital-outputratio can give only the direction of technological change but no quantificationof it. In order to get a quantification, we have to computeseparately the variations which intervenein its componentsak and a,. Turning back to Prof. Solow's study, we can easily see to what cumbersomeresults the neoclassical approach to the problem of technical change may lead. His conclusions are that in the United States economy between I909 and I949 there was an increase in capital intensity, but he has not explicitly taken into account the changes which occurred in the production of physical capital itself, i.e. the changes in productivity in capital goods industries. If we take from Goldsmith'sresearch9the same categories of capital goods Solow has used and compute the capital-output ratio in current values, we can find that such a ratio decreased from 2.75 in I909 to 2.20 in I949. I would say, therefore, that capital intensity in that period decreased.

FurtherComment Recalling that ak = I/C/N and aq=I /Q/L, we disThis section will be devoted to further comcover that the capital-outputratio, with the only ments, analytical refinements, and to dropping difference of the term rak, is nothing but the some assumptions. The analysis has been deratio between the productivity of labor and the veloped in the traditional framework of comproductivity of capital as above defined, that is parative statics; in other words comparisons the ratio ('7). were made between positions of stationary equiIf r remains constant through time, the term librium, which is what determines uniquely the rak modifies the absolute value of the ratio but quantities we have been using. (Otherwise outnot the direction of change. What we said of put and capacity at any time would depend the ratio Q/L/CN can thereforebe extendedto on which point of the production function we ,B, i.e. we can say that between two points of refer to). This means that the variables we time technical change has been neutral, capital have been using are not simple definitions but saving, or labor saving according to whether 3 result from a theory. We all know, however, has remained constant, has decreased, or has what kind of assumptions static positions of increased. These are nothing but Harrod's no- equilibriumimply and it would be naive to extions of technical change.8 We have shown that pect that such positions are ever observed in reality, except by a fluke. Observed variaII have expressed the remuneration of capital and of labor in terms of marginal productivities. This does not tions of Q/L and C/N embody not only techentail any particular assumption about the production funcnical change but also all deviation from optition. All that it implies is that at the present state of demum positions caused by any kind of stickimand, all economies of scale have been exploited (analytirak+ aq
cally, that there exist constant returns to scale in the neighborhoods of equilibrium). 8Quoting from Harrod's Towards a Dynamic Economics "Technological advances may be 22-23: (London, I948), labour saving or capital saving. The correct definition of a neutral advance has been a matter of disagreement....... I define a 'neutral' advance as one which, at a constant rate of interest, does not disturb the value of the capital coefficient." 'R. W. Goldsmith, A Study of Saving in the United

States, Vol. iII (Princeton,I955).

CONCEPTS AND MEASURES OF CHANGES IN PRODUCTIVITY ness, indivisibility, market imperfection, and so on. Therefore, satisfactory measures of technical change should be expected only over reasonably long periods of time and between positions which, from a cyclical point of view, do not differ greatly. However, it is worth noticing that, although the whole analysis started in the neo-classical way, the neo-classical framework is not necessary to it. As a matter of fact, it has been the purpose of the last section to frame the analysis in such a way that it can be interpreted also, perhaps much better, in other theoretical frameworks, such as the Leontief models or the dynamic growth models which pay more attention to fixed coefficientsand to idle capacity. In this way, many of the assumptions which underlie the traditional theory (continuities, divisibilities, transferabilities) become unnecessary.This is a useful property especially for comparisons for single industries, where the occurrence of fixed coefficients and the notion of capacity acquire much more precise content than for the economy as a whole and can be explicitly accounted for. In this respect, we have not specified whether the foregoinganalytical frameworkrefers to the economy as a whole or to single sectors. In principle it can be applied to both. In the case of the whole economy, the units of measurement we have been using are subject to the usual limitations, particularly to the index number problem, but with an important difference. The measure of output is subject to those limitations but not the measure of capital. A "real" measure of capital has been in fact discarded from our analysis in order also to take into account changes in productivity in the capital goods industries. In the case of particular industries there is a further adaptation to be made. Industries are usually not integrated- they buy raw materials and semiproducts from other industries. For the purpose of computing changes in productivities total production has to be replaced by "real"quantities of values added, while capital-ouput ratios have to be interpreted as capital necessary for producing one dollar's worth of value added. The index number problem here becomes less and less serious the more the industries represent disaggregatedentities (in the same industry

275

prices move roughly in a proportionalway)it would become irrelevant altogether in the limiting case of an industry producing only one product subject to no qualitative change. The time has now come to turn to the assumptions which were introduced in the second section. The assumption that capital, once made, lasts forever is easily dropped. All quantities could be interpreted in net terms but, since depreciation allowances always contain elements of arbitrariness,it is better to work with gross quantities. In this case prices of final goods contain (besides wage cost and profit) an element representing depreciation, which can be expressed in terms of the input of capital (ak) by multiplying it by a constant term (r) which, in the case of straight-line depreciation,represents the reciprocal of the length of time capital
goods last. Hence 10
Pq = [+(r+)ak+aq] W.

After substituting this term into (iO) we have

'0This result can be shown in many ways. If we want to stick to the comparative statics framework, we can resume the system of note 5 and develop an analogous system with different assumptions as to the durability of capital goods. Again we have BQ ZQ (2) R = (3) W = (I) Q =f(K,L) Suppose now that capital goods have a limited life-n years -and then they are scrapped. Call M the number of manyears employed in the capital goods production process and L the number of man-years employed in the consumption goods sector. where P = total labor force. (6) P = M + L As before, a being the quantity of capital goods made by one unit of labor, the annual production of capital goods will be aM and the existing capital stock will be composed of n groups of capital goods produced in n different years. (4a) K = anM (Now nM takes the place of N in the previous note.) The value of each one of these groups will be given by the present value of its expected remuneration from now to the time of scrapping. Using for analytical purposes a continuous law of capitalization, each group, when it is new, is worth e-rn I = RaM RaMS ertdt 0

~~~~~r

Therefore, if the whole capital is balanced, its value will be rn - I + e7, RaM = RaM fn[i-e-")t)idt r r ? 1 which in equilibrium must be equal to its cost, i.e.: e7T rn-I i = MnW. (5a) RaM H Now, P,n,r being given, the 6 equations determine Q,M,L,R,W and the equilibrium value of K. The consequence for the analysis of the third section is that the annual total remuneration of capital (RK) includes now two parts-a replacement charge (MW) and a net remuneration; (D) becomes

276

THE REVIEW OF ECONOMICS AND STATISTICS posite commodity which is made up of physical units from different industries and in proportions which change in time. Now, a "real" measure of capital can also be used for the computation of that gain in capital productivity which is attributable to substitution of some capital goods for others. This can be done by comparing, at time t + 0, the composition of an observed unit of capacity with a hypothetical unit of capacity in which proportions among capital goods have remained those of time t. Examples of these computations will be given. Finally, let us consider the assumption that capital goods requireno capital to be produced. Dropping this assumption means that the capital input ak represents not only labor but also capital (employed in the capital goods industry), i.e., ak must be split in the same elements which were specified for the consumptiongoods industry:
ak =
aki

to modify the interpretation of the capital-output ratio, but our previous conclusions about the direction of the changes remain unaffected. I shall add that in the formula arrived at above there appear explicitly a rate of profit and a life of capital goods. Their estimation is not
needed, however, because
ak

and aq can be

obtained directly. Possible variations of (T + r) through time only would make effective, if we want to interpret our analysis in the neo-classical framework,the limitation of giving up the distinction between a change and a movement on the production function. The evidence mentioned at the beginningshows, however, that the rate of profit does not change very much over time. Some interesting refinements can now be introducedby making use also of a "real"measure of capital, not however for the purpose of measuring total changes in capital productivity but in order to distinguish changes in productivity of capital which originate in the consumption goods industry from those which originate in the capital goods industry. Of course, this distinction is somewhat arbitrary, but arbitrariness refers only to attributions of origin, not to the total evaluation of the change. The procedure is simple. After expressing capital in physical units at two points of time we can compute, for the capital goods industry, the input of labor in one unit of physical capital and its change in time. By subtracting from the over-all change in capital productivity, that part can be obtained which is attributable to improvementsin the consumption goods industry. This computation suggests a further step. Capital, even for a single industry, is not a homogeneousgood. A unit of capacity is a comK-MW -?+MW----LaK aL Q / P0c P0C By making use now of the right-hand side (5a) for expressing the value of capital and by recalling that the rate I aQ - we have of profit is now - K-MW 3K Pc,C

(T + r) akk

where akl = input of labor in one unit of capital goods; akk = input of capital in one unit of capital goods. Empirically, what we can get easily
is not
ak

(the one which is needed) but

aki.

Therefore, measuring changes in the latter, instead of changes in the former, akk is implicitly assumed to vary in the same proportion as ak1. Normally this is not a serious drawbackbecause capital-producingindustries are usually highly labor intensive and [ (T+ r) akk] is, therefore, not the major part of ak, so that in whatever way it may change, variations of akl will not be very different from variations of ak. We can, however, consider under what conditions these changes coincide and, in order not to carry the
analysis ad infinitum, let us suppose that the

PqQ=

(aQ

P0C

P0C

aQ

capital goods industry makes capital goods both for the consumption goods industry and for itself. In this case ak = yakk (the ak's refer to capacity, therefore y here stands for the ratio between capacity in terms of consumptiongoods and capacity in terms of capital goods). Hence
ak1= [I-y(T+r)]ak

PgQ= rPcC+
= rak WC

MW L W P0C+ - WQ MnW Q

I -

ak WC

ag

WQ

and since C=Q

[(

r?L)ak?aq]WQ

which corresponds to the expression in the text.

which means that if rates of profit and life of capital goods are stable over time (possible changes in y can be obtained from our own analysis and can be accounted for), changes of ak, are exactly proportional to changes of ak. Should we think, however, that we do not have

CONCEPTS AND MEASURES OF CHANGES IN PRODUCTIVITY

TABLE I. BASIC DATA FOR TECHNOLOGICAL CHANGE IN U.S., I929 TO I950
I929 1950

277

Personal consumption expenditure ($ million) Gross private investment

78,952 I7,002

I94,026 49,0I8

($ million)
Total

95,954

243,044 I06.2
I44.3

Implicit price deflator for consumption expenditure or Capital in I929 prices ($ .million) Structures

73.6
Ioo
I69,7I9

Producer durables Inventories

Totals Capital at current prices

39,138
37,9I5 246,772

I64,166 64,329
48,o62 276,557

($ million)
Structures
334,22I I04,172

Producer durables Inventories

Totals Full-time equivalent employees, private sector (thousands of
man-years) 32,712
2.9333

77,219
5I5,6I2

labor and capital. Starting from the basic data11 gathered in Table i, I proceeded to compute the quantities describedin the previous sections. In order to estimate the number of employees in the consumption goods sector and the number of man-years which would be necessary for reproducingthe existent stock of capital I have made two assumptions: (i) that employees are distributed between the consumptiongoods and capital goods sectors in the same proportionsas gross national product, (2) that gross income per employee is the same in the two sectors. Certainly, these assumptionsare not correct but they introduce the same type of bias both in the I929 and in the I950 figures and therefore not in their ratios. The results are presented in Table 2. They
TABLE 2. -TECHNOLOGICAL CHANGE IN U.S., I929 TO I950
I929 I950

4I,332
5.88029

Gross income per full time employee ($ thousand)

SOURCES: Data on capital are taken from Goldsmith, A Study of Saving in the United States, Vol. III, 20-2I; sum of columns 5, 6, 7, 8, 9 for structures; column I2 for producer durables; columns I5, I6, I7 for inventories; pages I4-I5 sum of columns 4, 5, 6, 7, 8 for structures; column II for producer durables; columns I4, I5, i6 for inventories. All the other data are from Department of Commerce, National Income, 1954 edition, I62-63, 2I6-I7, I96-97.

Employees in the consumption goods sector (thousands of man-years) 26,9I5.79 Man-years in the capital stock (thousands): Structures 57,857.83

32,996.0I

56,837.50
I7,7I5.45 I3,I3I.83

Producer durables

yet the right approximation,we can repeat the same computations, one step back in the production process, for the capital goods industry. Technological Changein the U.S., 1929-50 I present here an estimate of the nature and relative magnitude of technical change which occurred in the United States private economy between I929 and I950. The data are those given by the Department of Commerce for income (going back to I929) and by Goldsmith for capital (up to the beginning of I950). The year I929 was one of full employment and perhaps of over-utilization of capacity. On the other hand, I950 was a year of recovery following a mild recession. A correction could be attempted, but it would entail some arbitrary elements which are better left out. So we shall simply keep in mind that, owing to some cyclical elements affecting the data, the results will slightly underestimatetechnical change. I am considering only the private sector of the economy and two factors of production-

Inventories Totals Real consumption goods production = capacity ($ million I929) aq = input of labor in

I3,342.29 I2,925-36 84,I25.4-8

87,684.78

78,952

I34,469.47

one unit of output

ak = input of labor in

*3409I33 I.0655269
2.933297I
{

.2453792 .6520795
4.0753250

one unit of capacity

I/aq
=

QIL

Index

I/ak = C/N
Index Capital-output ratio: 8
P1929/P1950

ioo
.9385027 ioo
3.I25595

I38.93 I5335553
I63.40 2.6574374

X I00

II7.62

show that between I929 and I950 in the United States private economy there have been (i) a neutral increase in productivity (i.e., of both capital and labor) of 39 per cent, (2) a further increase in productivity of capital alone of 24 per cent. Evidently, the neutral effect has affected both the numeratorand the denominator of the capital-outputratio, leaving it unchanged. The capital-saving effect on the other hand" I have kept the terminology of the statistical sources I have used.

278

THE REVIEW OF ECONOMICS AND STATISTICS mentioned before. In order to evaluate the impact of these changes in productivity and substitution on productivity of capital, I have weighted them with the proportions in which each type of capital goods enters into one unit of capacity. After setting the physical capital producedin I929 by one man-yearin the capital goods industry equal to ioo, I have obtained for I950 an index of io6.5I when I have adopted I929 capital goods proportions,and an index of I08.9I when I have adopted the capital goods
proportions of I950.15 This means that the joint

24 per cent increase- has influencedthe entire numerator but only a fraction of the denominator (that fraction representing replacements and profits), resulting in a fall of I7.6 per cent in the capital-output ratio. The total effect of technologicalchange has been capital saving. Our analysis can now be pushed further to inquirehow changes in capital productivity took place. Looking first at the composition of one unit of capital in I929 and in I950 it clearly appears that between the two years there has been a substitutionof equipmentfor structures.'2 This substitution is very marked in physical terms, less pronouncedin current values, which means that prices of structures have increased relatively to equipment prices. The following computations therefore are also a way of enquiring whether changes of productivity in the capital goods industries give an explanation of this structuralchange in prices. By definingone unit of physical capital as that quantity which made up one unit of capacity13 in I929 and then computing labor inputs and changes in productivities, Table 3 has been obtained. It shows that between I929 and I950 labor productivity decreased i per cent in construction (structures) industries,' increased24 per cent in equipmentproducing industries, and increased 25 per cent in inventory-producingprocesses. This pattern of change has evidently determineda rise in the price of structures with respect to other capital goods and has induced the kind of substitution 12Here is the percentage composition of one unit of capital in I929 and in I950:
1929

effect of labor productivity increases- in capital goods industries - on capital productivity has been +6.5 per cent. Substitution of some capital goods for others has brought about a further increase in capital productivity of + I.5 per cent. The results can now be summarized as follows: (i) better utilization of labor and
physical capital (the neutral effect) 39% (2) joint effect on capital productivity of improvements in capital6.5% producingindustries (3) further increase in capital productivity due to substitutions I 5% among capital goods (4) further improvements in utilizai 6 %o tion of physical capital Total increase in capital productivity 63%

Structures

Physical capital current values 68.8%

I5.9%

Physical capital in I929 units 59-4%

23.39

I950

Current values 64.8%

20.2%

I950

Technical Change in the Automobile Industry, 1939-47

durables Producers'

% Inventories I5 I7.3% I5.3% 1aI have made this change of unit of measure - with order to make the respect to the one used in Table 2 -in labor inputs of Table 3 directly comparable with those of Table 2. The results, in terms of indexes, would have been the same anyhow. ' The decrease in productivity in the construction industry is a little puzzling. The explanation must be sought principally in the nature of the capital data. The proper data for our type of analysis are data on capital at reproduction cost. What we normally have, on the other hand, are either historical costs or current evaluation. Only the latter are likely to reflect reproduction costs. Now, Goldsmith's estimates have been made in such a way that they are closer to historical costs than to current evaluations, and this represents a serious drawback for our computations, especially for investment in structures, which are the most durable.

I now present an example of estimation of technologicalchange in a particularsector-the

United States automobile industry, I939-47.-1

The interval of time is quite short and therefore there is the possibility that some temporary or cyclical variations may obscure the technical
' The index of physical capital - in the units of Table 3 - produced by one thousand man-years in the capital goods industries is .938507 in I929; .998593 in I950 (using I929 capital goods proportions in one unit of capacity); again in I950 (but using 1950 capital goods proI.022090 portions). 16I am very grateful to Prof. W. W. Leontief and to Dr. E. W. Gilboy for having put at my disposal many unpublished data which the Harvard Economic Research Project is working on, and for giving me permission to use some of them in the present section.

CONCEPTS AND MEASURES OF CHANGES IN PRODUCTIVITY

TABLE 3. - BREAKDOWN OF CHANGE IN CAPITAL PRODUCTIVITY IN U.S., I929 TO I950
I929 I950

279

TABLE 4. - BASIC DATA FOR TECHNOLOGICAL CHANGE IN U.S. AUTOMOBILE INDUSTRY, I939 TO I947
I939 I947

rstructures

54,3I8.98

52,557-94

Physical

producer

durables

I2,553.37 I2,079.65
78,952

20,6I6.I6
I5,307.28 88,48I.38

inventories

totals L
gstructures

I.065I494
I.0628452

I.08I4I70
.8592992

Labor inputs

producer durables
linventories

I.0700I I I .93.88354 I00 .9408707

.85788I3
.9247I266

rstructures index Reciprocals of labor inputs producer durables (labor proindex ductivities)
inventories index

98.96
I.I637390 I23.69 I.I656624
I24.73

ioo
.9345697 ioo

a The unit of measurement is that quantity of capital which makes up one unit of capacity ($ million) in I929.

change. However, in dealing with a particular sector of the economy, at least temporaryvariations due to idle capacity and to inter-sectoral differencesin profit and wages can be accounted for. The basic data for the computations have been gathered in Table 4. They have to be adapted to our purposes. The production process which is considered here is the transformation into automobiles of the materials the industry buys from other sectors. Therefore, our inquiry does not extend to the production of those materials. First we must correct the price index for output, because we need a price index which refers only to value added. The correction has been made by utilizing the input-output table for I947.*7 I have isolated the most important inputs of the industry (rubber, 3.5 per cent of the total; iron and steel, 7.8 per cent; metal products, 6.7 per cent; machine tools, 4.4 per cent; electrical machinery, 4.2 per cent); then I have considered indexes of their prices (respectively 209, I83, i88, I42, I5); and finally I have made a correction of the price index for automobiles, which has changed from I89 to I93.77For capital I have utilized coefficients estimated by the Harvard Economic Research Project. These estimates refer to capital-output
' W. Duane Evans and Marvin Hoffenberg, "The Interindustry Relations Study for I947," this REviEw, xxxiv (May I952).

Total productiona ($million) 4,047.87 Capacity a ($ million) 5,4I9.60 Value addedb ($ million) I,84 No of full-time equivalent employeesb (thousand man-years) 467 Price index for automobiles a IOO Capital coefficients a (with respect to total production): Construction .II70 Machine tools .2009 Other tools .0368 Electrical machinery Other items .0004 Totals .355I Price indexes a Machine tools IOO Other industrial equipment ioo ioo Electrical equipment Construction IOO Value added per full time equiv. employee b Automobile industry ($ thousand) 2.53533 Construction ($ thousand) I.90566 Machinery ($ thousand) 2.254I6
SOURCES:

I2,855 I4,850
3,522

749
I89

JI755 *I57I .0292 .0I9I

.oo68

.3877
I42

I4I I5I
I90.3

4.7022 4.07420 3.99742

a Harvard Economic Research Project, Reports on Research for I954, II9-20, and for I955, I6; Estimates of the Capital Structure of American Industries, I939; 67, Estimates of the Capital Structure of American Industries, I947, 376. (All this material is unpublished.) b Department of Commerce, National Income, I954 edition, I77 and I97- Department of Commerce composite construction index, from Survey of Current Business.

ratios for I947 and to capital-capacity ratios for I939. The notion which is needed here, however, is still another one: namely a capitalcapacity ratio where capacity refers not to total production but to value added. The particular way in which the estimates have been done has induced me, moreover, to make some aggregations. The capital coefficientof Table 4 is composed of an item representingconstruction and of three other items representingmachine tools of various types. For these machine tools, classifications of I939 and I947 do not coincide completely. Since, however, prices of different types of machine tools between I939 and I947 have substantially changed in the same way, I have lumped them in a single item, "equipment."18 I have also confined myself to fixed
18Neither the construction industry nor the equipmentproducing industry is completely integrated. They also buy raw materials from other industries. However, I have not considered it necessary to go further back into the production process of capital goods because in both industries the proportion of value added to total production is very high (45 per cent in the construction industry, 6o per cent in the machine tools industry, while it is, for example, only 28 per cent in the automobile industry).

280

THE REVIEW OF ECONOMICS AND STATISTICS

capital (inventorycoefficientsfor I939 and I947 in the two years, I have made a conservative are not comparable). estimate of qualitative change in the following For the computation of the number of man- way. I have put i car I947 equal to I.25 cars years which would be needed for reproducing I939. Table 6 has then been obtained on this the stock of capital, I have proceeded in the same way as in the previous section, with the TABLE 6. -TECHNOLOGICAL CHANGE IN U.S. AUTOMOBILE INDUSTRY, I939-47 difference that I have been able to use here particular data about values added, for each I947 '939 sector. Table 5 contains the modified data. BeTABLE 5. DERIVED DATA FOR TECHNOLOGICAL CHANGE IN U.S. AUTOMOBILE INDUSTRY, I939-47
I939 I947

"Real" output "Real" capacity a,.= input of labor in one unit of output ak = input of labor in one unit of capacity
I/a, Indexgio

I,I84

I,5-85

2,272.02 2,624.90

.394426
.57I219

.329662
.470939

Price index for value added by automobile industry Io0 Ratio of total production to value added 3.42 Percentage of utilization of capacity Q X ioo/C 74.69% Capacity (of value added) I,585 ($ million) Stock of capital at reproduction cost ($ million) Construction 634.22 Equipment I,290.67 Totals I,924.89 Man-years in capital stock (thousands) Construction 332.8I Equipment 572-57 Totals 905-38 Capital coefficient relative to capacity (of value added) Construction 4002 Equipment .8I43 Totals I.2145

2535332
i23

3.033409
I9.65 2.I234I6 I2I.29

I93.77

Ila

I.750643
I00

3.65

Index

86.56%
4,o68.38

2,256.II 2,727.89

4,983
553.75
682.4I 1,236.I6

.5544
.6705
1.2249

fore performingon them the same computations as in the previous section, an important element here cannot be overlooked. Between I939 and I947 the quality of cars has changed quite a lot. Taking this into account is a difficult task because there is no objective way of putting it into quantitative terms. However, after considering carefully the characteristics of cars19produced
for example shipping weight and taxable horse-power of the three most popular cars (they represent more than 6o per cent of total production-each type had
'1Consider

assumption. The results show that between I939 and I947 the United States automobile industry has benefited by an increase of I9.5 per cent in labor productivity and an increase of 2I per cent in capital productivity. In other words technical change has been on the whole roughly neutral-or slightly capital-saving. It is worth noticing that a different estimate of qualitative change would influencethe level but not the proportionof these percentages. We would expect from these findings- if mobility of factors were perfect-a slight diminution of the capital coefficient. But markets have not been perfect and the capital coefficient (see Table 5) has remainedthe same-or more precisely it has increased o.9 per cent. The explanation is given by the pattern of profit and wage changes in the different industries between I939 and I947. Wages and salaries have increased more in the capital goods industries and our findthan in the automobile industry,20
cate strong differences in quality. Comparisons of other minor details show similar results. Note, however, that all these changes are not mutually exclusive criteria for evaluating qualitative changes - they add up. The assumption of I car I947 = I car i939 appears therefore to be a conservative one. '2Here I report some indexes which I have computed

two modelsin I939):

Chevrolet
Ford Shipping weight (lbs.) 1939 Change I947 2,805 3,170 +I3% 2,875
2,525 2,750 3,2I3 3,245 +27% +I8%

Taxable HP Change I947 I939

29.4 29.4 2I.1 32.5 +2I% + 8%

from the Department of Commerce estimates (National Income, 1954 edition, supplement to Survey of CurrentBusi-

ness):
1 I947 Wages and salaries per employee (indexes) IOO Automobile industry I75-93 Construction industry ioo 2I8.93 Machine tools industry Ioo I84.32 It is more difficult to say how profits behaved; the sum of profits, wages, and salaries however-i.e. value addedfollowed roughly the same pattern of wages and salaries:

29.4
2I.6 30.-

Plymouth

2,839
2,909

3,o82

+ 9%

23.4

25.3

8% and

23.4 SOURCE: AutomotiveIndustries,February 25, I939, March I5, I947, I26-27.

214-I5,

The remarkable increases both in weight and power patently indi-

CONCEPTS AND MEASURES OF CHANGES IN PRODUCTIVITY

TABLE 7. - BREAKDOWN OF CAPITAL PRODUCTIVITY CHANGE IN U.S. AUTOMOBILE INDUSTRY, I939-47
'939 I947

28I

construction

($ million
Physical capital a
I939) 522.2575

976.2566

equipment

($ million
I939) i,062.7425 I)593.02I9

jotals Labor inputs Reciprocals of labor inputs {construction

equipment

I 585 .637252 .538768 I.569238 i0o

2,569 .2 785
.567222

.428377 I.762978
II2.35 2.33439I I25.77 I20.I9

Cconstruction index equipment Lindex Units of capacity produced by i unit of labor (index)

i.856o86
I00 i00

a The unit of measurement is that quantity of capital which made up one unit of capacity in I939.

ings suggest that in the process of passing productivity increases on to the customers, capital producing industries, considered together, have passed on about 2 per cent less than the automobile industry. One more step completes our analysis, namely, the splitting of the total increase in productivity of capital into components reflecting its origin. I have followed exactly the same procedure of the previous section and the results are given in Table 7. The findings about technological change in the United States automobile industry can now be stated in the following way. As a whole the automobileindustry has benefitedbetween I939 and I947 by a roughly neutral increase in productivity of the order (with the assumption of i car I947 = I.25 cars I939) of I9.5 per cent. This change has been devoted entirely to improvements in quality; moreover, the improvements in quality have gone somewhat beyond the increase in productivity, so that in I947 the industry produced, from comparable units of capital and labor, about 5 per cent fewer automobiles than in I939. This last estimate is independent of any measure of qualitative improvements. Although the change has been roughly neutral on the whole, it came from a complex comI939 I947

bination of different factors acting in different industries. In the construction industry there has been an increase in productivity of I2 per cent and in the equipment-producingindustry an increase in productivity of 26 per cent. The joint effect of these changes on the productivity of capital for the automobile industry has been +20 per cent. Because of differencesin changes of wages, salaries, and profits, not the entire 20 per cent but only about i 8 per cent of the increase has become relevant for the automobile industry. On the other hand, within this industry, technical change has been almost completely labor-saving. Labor productivity has increased I9.5 per cent, while production from the same physical capital has increased only 3 per cent. This has been compensated by the increase in productivity in capital goods industries so that on the whole the result of technological change has been almost neutral and the capital-outputratio has remained practically unaltered. Summary I have been concerned mainly with the economic interpretation of technological change and I have centered my attention on capital. The contention has been that, since capital comes from the production process itself, on which technical change operates, it cannot be dealt with in the same way as labor and land. My critique has been directed: (i) toward the most common measures of technological change which run in terms of labor productivities, because they neglect completely all other factors of production, (2) toward the attempts at evaluating technological change with models based on production functions of neo-classical inspiration, because they need a "physical" or "real" notion of capital. I have discarded such a notion because the same physical unit of capital has differenteconomic meaning according to the state of technical knowledgewhich prevails. My proposalshave been in the direction of an over-all estimate which includes both the direct production process and the production process of capital. In the simplest case, the procedure reducesto the computationof two ratios Q/L, output per man, and C/N, capacity over labor necessary for reproducingit. These two ratios

Value added per employee (indexes) Automobile industry Construction industry Machine tools industry

iOO
IOO

I85.46
2I3.79

iOO

I77-33

282

THE REVIEW OF ECONOMICS AND STATISTICS into the analysis, and finally two examples of empirical computations have been given - one for the United States economy as a whole from I929 to I950 and one for the automobile in-

and their changes throughtime can give a rough but complete evaluation of the nature and relative magnitudeof technical change. More complicated cases and relations have been dealt with, some refinements introduced

dustryfromI939 to I947.

COMMENT Robert M. Solow Mr. Pasinetti's carefully thought-out article gives clear expression to one important view of productionand technologicalchange. I enjoyed reading it, but I must admit that I disagree with much of it. Differences here seem quickly to ramify into other more important matters, such as the theory of capital and the determinants of distribution. For this reason I thought it might be useful to elucidate the ultimate sources of the difference. In the process, I hope to strike a blow for our side. With this end in view I shall start on a very general level - I won't say "methodological,"for that is the kiss of death -and work my way down to some more specific comments. i. Mr. Pasinetti begins by suggesting that the neo-classical theory of production and pricing does not account for the basic facts of capitalist economic development, and may actually have led to quite opposite expectations. But what does neo-classical theory lead one to expect about relative shares, the rate of profit, and the other aggregates?In general, very little. The whole tone of the theory is much more agnostic, much less willing to state iron laws of this and that, than were the classical economists. I should think that the basic facts quoted by Pasinetti, far from contradicting neo-classical expectations, could all be interpreted plausibly within the framework of the theory. Of course the interpretationis after the fact. I had thought of my own paper on this subject as doing just that - tracing out what the shifts in the aggregate production function must have been if the macroeconomicfacts are to be explicable on the hypothesis of approximately competitive imputation. Whether this extreme generality is the beginning of wisdom or the end of relevance is another matter. Without much more reliable empirical knowledge than we now have, neoclassical theory does not yield many definite predictions (about aggregates). And it is cold comfort to be able sagely to accept the universe after the fact. It is certainly possible to construct alternative models which will yield definite predictions. Whether the predictions are correct is also an unsettled matter of some importance. And in this respect the neo-classical framework,suitably modernized,may not have been exploited to the end. - 2. This leads to another matter which has been much discussed. General equilibrium theory is fundamentallymicroeconomic. It runs in terms of the prices and outputs of individual goods and services. Aggregates appear only accompaniedby their exact definitions as functions of observables,and index numbers do not appear at all. A notion like "the stock of real capital" is no problem for the theory because the notion never occurs; "real capital" is not bought or sold on any market. However, to make the theory usable some amount of aggregationis indispensable. At this point two tacks are possible. One is to try to preserve the rigor of the theory by making special assumptions and defining special aggregates, so that the underlying microeconomic relationshipsentail the existence of corresponding macroeconomicrelationships. The other is to give up the rigor in favor of empirical practicality, to use conventionallydefinedaggregates and index numbers, and to construct macroeconomic systems more or less by analogy with microeconomics. Whether this device works can be discovered only by trying it out. But it seems to me to be a mistake to ask deep philosophical questions of such loose concepts. Index numbers will usually not behave precisely like the physical commodities they are supposed to

CONCEPTS AND MEASURES OF CHANGES IN PRODUCTIVITY mimic. They are not trying to do so; they are empirical compromises. They need scrutiny, of course, but of an appropriatekind. 3. We come now to the production function and the distinction between shifts of it and movements along it. This distinction Mr. Pasinetti proposes to abandon, at least for present purposes. With it goes and must go any possibility of dealing with technical change in traditional terms, for the traditional treatment is based squarely on this distinction. The notion of a production function is not simply a matter of formality or convenience. It is meant to express the fact (if it is a fact) that at any time a firm or industry or economic system is faced with a range of technological alternatives any one of which might be chosen, and that given enough time any previously made choice can be changed without appealing to new knowledge. Apart from laboratory situations only one choice is in fact made at any one time and that is all we can directly observe. But this by itself is no reason for abandoning the traditional distinction. Nobody has yet proposed giving up the distinction between shifts of and movements along demand and supply curves although all we ever observe for one market at any one time is one p and one q. The analogy is quite close, for in both cases it is the possibility of virtual movements that is supposed to determine what it is that actually takes place.1 4. To see the importance of this consider Pasinetti's first proposal for measuring and classifying technical change, which he describes as "satisfactory as far as land and labor are
concerned."2

283

reflect exclusively or primarily impulses from the side of the supply of saving and the growth of population. Then whatever happens to Q, Pasinetti's judgment about the nature of technical change will in fact reflect nothing about technology but only the relative speeds at which population and capital accumulate. In fact an economy with a stationary labor force, positive accumulation, and absolutely stationary technique would be described, as it slid into the Ricardo-Mill stationary state, as experiencing neutral technical regress accompanied by only partially compensating labor-saving technical progress! There is one case in which the simple comparison of average productivities will yield accurate results. It is when Q is produced by K and L in fixed proportions, and no one ever wastes any K or L. Then the technology is describedcompletely by the two coefficientsof production (reciprocal input-coefficients)Q/K a, Q/L= b, and the observed K/L ratio must be b/a. If subsequently Q'/K' = a' and Q'/L' = b' it seems reasonable enough to say that technology has become more capital-using if b'/a' > b/a and to measure the extent of progress by b'/b and a'/a. By abandoning the production function Mr. Pasinetti has in effect abandonedthe possibility of substituting capital for labor as of given technical knowledge. If technology were really like that, no one could complain. But to believe this seems to me to be folly, especially for aggregates and especially for periods as long as a
decade.3

It amounts to comparing propor-

tional changes in Q/K and Q/L. Technical change is said to be labor-saving or capitalsaving according as Q/L increases more or less than Q/K, i.e. according as K/L increases or decreases. Suppose that changes in K and L over time
'Professor Melvin Reder has insisted to me that what does blur the distinction is the regularization of research and the fact that resources are expended in "planned" shifts of the production function. One can get around this difficulty formally by enlarging the production function to include research as well as production activities and making future production possibilities depend on present and past investment in research. But that this raises a serious practical problem I do not doubt. 2 He also describes it as roughly like my method, which I roundly deny.

5. Next Mr. Pasinetti proposes to distinguish the production of consumables and the production of capital goods. This doublingthe number of commoditiesin the model increases its realism by IOOper cent. This is all to the good but I
would point out
-

only half in jest

that if

there are really i,ooo commodities worth distinguishing we only decrease the unrealism by about one-tenth of one per cent. In any case I would have welcomed an extended analysis of a two-sector model. It is not quite true to say
x In correspondence Mr. Pasinetti has suggested that instead of assuming fixed proportions he might compare points on different production functions with the same rate of profit. I think this either reduces once again to fixed proportions or fails to meet the difficulty.

284

THE REVIEW OF ECONOMICS AND STATISTICS accordingto Pasinetti, one can judge the factorsaving incidence of technical change by looking .Q/L at the ratio C/N' the average productivity of labor in the consumption goods sector divided by the amount of labor required to produce a capacity-unit of robots. Since he has already put Q = C, this ratio reduces simply to N/L. But this is just a measure of the capital-intensiveness of production in the consumer sector, the ratio of the labor-content of the stock of robots to the direct labor input into consumption goods. The argument I raised in paragraph 4 above against his first measure applies equally here. Any event that changes N/L will be registered as a technical change. Suppose that the rate of capital formation increases with no alteration in technological knowledge but as a result of a change in time preferenceand savings propensities. Then labor has to be withdrawn from the consumersector and put to work making robots; L will decrease and N will eventually increase. Pasinetti will say that a laborsaving or capital-using technical change has occurred. I will say that no technical change has occurred. Of course this story cannot occur if consumer goods are produced by labor and robots in fixed proportions. In that case and only in that case the ratio N/L is indeed dependent only on technical coefficients. So once again the view I am criticizing seems to rest on the hypothesis that one cannot substitute labor for robots in the production of final commodities. An even stronger case can be made. If we step outside the simplified model and suppose that there are two consumer goods each produced under fixed proportions but one using relatively more robots and less direct labor than the other, then a shift in consumers'tastes from one to the other will also eventually change the ratio of indirect to direct labor in the production of final output without anything that one would recognizeas a technical change having occurred. 6. The same point can be made in terms of the capital-output ratio as described in Mr. Pasinetti's third section. From his equation (io), remembering that C = Q, it is apparent that the capital-output ratio is nothing but the price of a unit of capacity in terms of consumption goods. In the super-simplecase where con-

that by failing to distinguish between consumables and capital goods I ignore that capital is in fact produced. At best, we are dealing with simplified models. I had in mind the simplest of all capital models, what might be called the Ramsey model, in which the same commodity or composite commodity serves both as consumptiongood and as capital good (wheat which can be eaten or used as seed). The next step is surely to do as Pasinetti does and consider a two-commodity model, which might be called the Wicksell model (from his "Mathematical Analysis of Akerman'sProblem"). In that case I would simply proceed as I did before with each of the industries separately and try to measure shifts in the production function for capital goods and shifts in the production function for consumergoods. The only obstacle in the way of this procedure is the difficulty of obtaining time series on inputs of labor and capital into the two sectors separately. Pasinetti is after bigger game, however. He wants to be able to say somethingabout an overall increase in productivity. This entails a shift in emphasis from my more modest objective. Considerhis sentence: "However-

and this is the important point -

what-

ever form technical change may take in the capital goods industry - whether neutral, labor-saving, or capital-saving- its effect is always capital-saving for the consumption goods industry." What does this mean? To see it clearly, suppose consumer goods are produced by robots alone and robots are produced by labor alone. Imagine an innovationwhich makes it possible to produce robots more cheaply, without changing the nature of the robot at all nor its characteristics in use. Then we would all agree that a labor-saving invention had occurred in the capital goods sector. In what sense can it be said to representa capital-saving improvementin the consumersector? It doesn't save robots. What it saves is an abstract "waiting." It now takes less saving to add a robot to the stock of capital than it did before. This is a true statement and an interesting one. But it mixes up, as such statements must, technological and non-technological facts. To clinch this point I must go back to the model of Pasinetti's second section, and allow that some labor is required to operate the robots. Then,

CONCEPTS AND MEASURES OF CHANGES IN PRODUCTIVITY sumption goods are produced by robots alone and robots last forever, it is easy to see that the price of a unit of capacity in terms of consumer goods is simply the reciprocal of the rate of profit. Thus anything which changes the rate of profit has to be read as a non-neutral technical change. This is also clear from Pasinetti's

285

is it worth distinguishing between autonomous shifts in technical possibilities and induced reversible changes in factor proportions? If the answer to this question is Yes then it also follows that discussions of productivity and technical change should be carried on in terms of physically identifiable inputs and outputs, or frankly approximate index numbers for these. equation (iO). 7. To sum up, when the difference between If we must talk about things like "real capital" Pasinetti's view and my more traditional one is as distinct from real capital goods, the time to stripped down to essentials it turns out to have do that is at the end not the beginning. To nothing to do with whether capital goods are believe otherwise is to fall into the previously produced or not. The basic question is this: unnamedfallacy of Misplaced Abstractness.

REPLY Luigi L. Pasinetti I am very grateful to Professor Solow for devoting so long a comment to my article. I appreciate his calm and carefully framed criticisms, although I must say that I disagree with most of them. Professor Solow is right in saying that the two views we hold stem from basically different approaches to the problem of production and technological change. Unfortunately, I have no space here to go into this basic issue, so that I shall have to confinemy short reply to Professor Solow's more specific remarks. i. Almost all these remarkshave been aimed, in many different ways, at one single point: to show how the approach I have suggested might yield misleading conclusions in a type of world where there is no technical change and where capital accumulationtakes place only by movements along a well-defined production function (with the rate of profit continuously falling through time). My analysis, however, started just from the realization that, fortunately, the world is not of that type. If it were so I would be perfectly happy with the neo-classical framework (and unhappywith the world). The point I started from, on the other hand, is the plain fact that, apart from short-run fluctuations, by far the largest part of changes in productivity over time have been shown to be due to technical change' and only to a minor extent to changI Much as I disagree with Professor Solow's way of dealing with shifts and movements along a production function,

es in income distribution. I have simply suggested, therefore,an approachwhich focuses the investigation on the first cause, as opposed to the neo-classical analysis which focuses it on the second. This is not equivalent to denying that "at any time a firm or an industry or an economic system is faced with a range of technological alternatives any one of which might be chosen," but only to accepting, for the purpose of evaluating technical change, those solutions of this problem which have been actually chosen and which are the only ones we can observe. To constrain the alternative bygone possibilities in a function which cannot but be arbitrary because non-observable is of no use for our purpose if, "given enough time [for] any previously made choice to be changed,"the firm or the industry or the economic system will be faced with the same rate of interest but with an entirely new set of technical choices. 2. Another point which is brought into the open by Professor Solow refers to the distinction between production processes of consumption and of capital goods. My position on this issue is not one of more or less aggregation, as Professor Solow seems to have interpreted it. I never said that a two-commodity model as opposed to a one-commodity model increases realism ioO per cent! What I do say is that, at whatever level of aggregation our analysis may
I recall here that this is a conclusion which also comes from his own empirical analysis.

286

THE REVIEW OF ECONOMICS AND STATISTICS

be carried on, if we consider a factor of produc- as to whether the "shift" occurs in the first or tion - namely capital - which itself comes in the second function. With the approach I from the process of production, any evaluation have suggested these are differenttypes of techof changes in productivity cannot leave without nical change; they have an entirely different an explicit consideration the technical change impact on the dynamicmovementof the system. which may occur in the production process of Professor Solow would invariably classify them capital. In an investigation at the level of single as of the same type -neutral technicalchanges! industries (i.e., on values added by single in- - no matter what happens to relative prices, to dustries) the point is so clear as hardly to need the capital-output ratio, and - to mention one any further comment- neglecting to consider importantconsequencein a dynamic analysis technical improvementsin the capital-producing to the relation between the propensity to save industries means just forgetting something, i.e. and the rate of capital accumulation. It is surprising that Professor Solow defines a sheer mistake. But also in an analysis which embraces the economy as a whole, even if we my conclusions on this point (namely that any conceive - to make things suitable for Profes- kind of technical improvement in a capitalsor Solow's approach-a production function producing industry means a capital-saving imfor capital goods of exactly the same type as the provement for the industry that uses that capiproductionfunction for consumergoods (so that, tal) as "a true and interesting statement" but analytically, they might even be aggregatedinto then he does not go on to draw the consequences one function) the consequences of technical and falls back once more to the sort of criticisms change on relative prices are differentaccording I have mentionedunder my first point.