i

I appeals or it doesn't. Personally I belong to both schools. If it

16 R. Solow
does, I think one can draw some crude but useful conclusions
Technical Change and the Aggregate Production Function1 from the results.
i
Theoretical basis
I
R. solow, 'Technical change and the aggregate production function', I I will first explain what I have in mind mathematically and then
Review of Economics and Statistics, August 1957, pp. 312-20. 1 give a diagrammatic exposition. In this case the mathematics
I
I seems simpler. If Q represents output and K and L represent ,
I capital and labor inputs in 'physical' units, then the aggregate
i production function can be written as
I
In this day of rationally designed econometric studies and super iI Q = F(K,L;t). 1
input-output tables, it takes something more than the usual
'willing suspension of disbelief' to talk seriously of the aggregate i The variable t for time appears in F to allow for technical change.
production function. But the aggregate production function is I It will be seen that I am using the phrase 'technical change' as a
only a little less legitimate a concept than, say, the aggregate 1 shorthand expression for any kind of shift in the production func-
consumption function, and for some kinds of long-run macro-
models it is almost as indispensable as the latter is for the short ! tion. Thus slowdowns, speedups, improvements in the education
of the labor force, and all sorts of things will appear as 'technical
run. As long as we insist on practicing macroecononlics we shall I change '.
II
need aggregate relationships. I It is convenient to begin with the special case of neutral tech-
I
Even so, there would hardly be any justification for returning nical change. Shifts in the production function are d e h e d as
to this old-fashioned topic if I had no novelty to suggest. The neutral if they leave marginal rates of substitution untouched but
new wrinkle I want to describe is an elementary way of segregat- simply increase or decrease the output attainable from given
ing variations in output per head due to technical change from inputs. In that case the production function takes the special
those due to changes in the availability of capital per head.
Naturally, every additional bit of information has its price. In
j! form

this case the price consists of one new required time series, the Q = A(t)f(K,L), 1a
share of labor or property in total income, and one new assump-
tion, that factors are paid their marginal products. Since the
11 and the multiplicative factor A(t) measures the cumulated eRect
of shifts over time. Differentiate l a totally with respect to time
former is probably more respectable than the other data I shall
use, and since the latter is an assumption often made, the price I and divide by Q and one obtains

may not be unreasonably high. 1 Q-

-
Q
-A+A--+A-
-
A
2 f
aKQ
~ a f -t
aLQ
Before going on, let me be explicit that I would not try to I
justify what follows by calling on fancy theorems on aggregation 1 where dots indicate time derivatives. Now define
and index numbers.' Either this kind of aggregate economics I

1. 1 owe a debt of gratitude to Dr Louis Lefeber for statistical and other wk = (aQlaK>KlQ
assistance and to Professors Fellner, Leontief and Schultz for stimulating I
suggestions. Eco~zornicStudies, vol. 23, no. 2). Were the data available, it would be better
2. Mrs Robinson (1954) in particular has explored many of the profound to apply the analysis to some precisely defined production function with
difficulties that stand in the way of giving any precise meaning to the many precisely defined inputs. One can at least hope that the aggregate
quantity of capital, and I have thrown up still further obstacles (Review of analysis gives some notion of the way a detailed analysis would lead.

544 Technical Change and the Aggregate Production Function i R. Solow 345
and easily handled graphically. The production function is com-
pletely represented by a graph of q against k (analogously to the
WL = (2QIaL)LIQ
fact that if we know the unit-output isoquant, we know the
the relative shares of capital and labor, and substitute in the whole map). The trouble is that this function is shifting in time,
above equation (note that i;'Q/bK = A 8f/2~,'etc.) and there so that if we observe points in the (q, k) plane, their movements
results : are compounded out of movements along the curve and shifts of
the curve. In Figure 1, for instance, every ordinate on the curve

From time series of Q/Q, w k ,k/K, W L and L/L or their discrete

year-to-year analogues, we could estimate A/A and thence A(t)
itself. Actually an amusing thing happens here. Nothing has
been said so far about returns to scale. But if all factor inputs are
classified either as K or L, then the available figures always show
1 v and
~ wL adding up to one. Since we have assumed that factors
are paid their marginal products, this amounts to assuming the
hypotheses of Euler's theorem. The calculus being what it is, we
might just as well assume the conclusion, namely that F is
homogeneous of degree one. This has the advantage of 111aki11g
everything come out neatly in terms of intensive magnitudes.
Let Q/L = q, K/L = li, 1vL = 1 - w ~ note; that q/q = Q/Q- L / L
etc., and 2 becomes

Now all we need to disentangle the tech~iicalchange index A ( t )

are series for output per man-hour, capital per man-hour, and the
Figure 1
share of capital.
So far I have been assurning that technical change is neutral. for t = 1 has been multiplied by the same factor to give a neutral
But if we go back to I and carry o~ztthe same reasoning we arrive upward shift of the production function for period 2. The prob-
at something very like 2a, naniely lem is to estimate this shift from knowledge of points Pi and P2.
Obviously it would be quite misleading to fit a curve through raw
observed points like PI, P2and others. But if the shift factor for
each point of time can be estimated, the observed points can be
!'t can be shown, by integrating a partial differential equation,
corrected for technical change, and a production function can
that if F/F is independeni of K and L (actually under constant
returns to scale only K/L matters) then P has the special form l a then be found.3
3. Professors Wassily Leontief and William Fellner independently
F~F -
and shifts in the production function are neutral. If in addition
is constant in time, say equal to o, then A(t) ear or in
-
discrete approximation A(t) (i +o)f.
pointed out to me that this 'first-order' approximation could in principle
be improved. After estimating a production function corrected for tech-
nical change (see below), one could go back and use it to provide a second
The case of neutral shifts and constant returns to scale is now approximation to the shift series, and on into further iterations.

346 Technical Change and the Aggregate Production Function

 The natural thing to do, for small changes, is to approximate would surely increase. But the maximal flow of capital services
the period 2 curve by its tangent at P2(or the period 1 curve by its would be constant. There is nothing to be done about this, but
tangent at PI). This yields an approximately corrected point PI*, something must be done about the fact of idle capacity. What
and an estimate for AA/A, namely E23/q1. But belongs in a production function is capital in use, not capital in
klP12 = qz- aql8kAk place. Lacking any reliable year-by-year lneasure of the utilization
-
and hence PI2P1= q2q 1 cq/ CkAk -- Aq- aq/akAk
--

and AA/A = PI2Pl/ql= Aq!q- ?qlZk(k!q)Ak/k =

Ag/q- ~ ~ A k j k
which is exactly the content of 2a. The not-necessarily-neutral
case is a bit more complicated, but basically similar.
An application to the US: 1909-49
In order to isolate shifts of the aggregate production function
from moverilcnts along it, by use of 2a or 2b, three time-series
are needed: output per unit of labor, capital per unit of labor,
and the share of capital. Some rough-and-ready figures, together
with the obvious conlputations, are given in Table 1.
The conceptually cleanest measure of aggregate output would
be real net national product. But long N N P series are hard to
coine by, so I have used G N P instead. The only dift'erence this
makes is that the share of capital has to include depreciation. It
proved possible to restrict the experiment to private non-farm
econon~icactivity. This is an advantage (a) because it skirts the
problem of measuring governillent output, and (b) because
eliminating agriculture is at least a step in the direction of homo-
geneity. Thus my q is a time-series of real private non-farm G N P
per man-hour, Kendrick's valuable work.
The capital time-series is the one that will really drive a purist Figure 2
mad. For present purposes, 'capital' includes land, mineral
deposits, etc. Naturally 1 have used Goldsmith's estimates (with of capital I have simply reduced the Goldsmith figures by the
government, agricultural and consumer durables eliminated). fraction of the labor force uneinployed in each year, thus
ldeally what one would like to measure is the annual flow of assuming that labor and capital always suffer unemployment to
capital services. Instead one must be content with a less Utopian the same percentage. This is undoubtedly wrong, but probably
estimate of the stock of capital goods in existence. All sorts of gets closer to the truth than making no correction at aIL4
conceptual problems arise on this account. As a single example, The share-of-capital series is another hodge-podge, pieced
if the capital stock consisted of a lnillion identical machines and 4. Another factor for which I ha\e not corrected is the changing length
if each one as it wore out was replaced by a more durable machine o l the work-week. As the work-week shortens, the intensity of use of
of the same annual capacity, the stock of capital as nleasured existing capital decreases, and the stock figures overestimate the input of
capital services.

348 Technical Change and the Aggregate Production Function

R. Solow 349
together from various sources and ad hoe assunlptions (such as change, at least looks reasonable. But on second thought I
Gale Johnson's guess that about 35 per cent of non-farm entre- decided that I had very little prior notion of what would be
preneurial income is a return to property). Only after these 'reasonable' in this context. One notes with satisfaction that the
computations were complete did I learn that Edward Budd of trend is strongly upward; had it turned out otherwise I would
Yale University has completed a careful long-term study of not now be writing this paper. There are sharp dips after each of
factor shares which will soon be publisl~ed.It seems unlikely that the World Wars; these, like the sharp rises that preceded them,
minor changes in this ingredient would grossly alter the final can easily be rationalized. It is more suggestive that the curve
shows a distinct levelling-off in the last half of the 1920s. A
sustained rise begins again in 1930. There is an unpleasant saw-
tooth character to the first few years of the AA/A curve, which I
imagine to be a statistical artifact.
The outlines of technical change
The reader will note that I have already drifted into the habit of
calling the curve of Figure 2 AA/A instead of the more general
AF/F. In fact, a scatter of @/Fagainst K/L (not shown) indicates
no trace of a relationship. So I may state it as a formal conclusion
that over the period 1909-49, shifts in the aggregate production
function netted out to be approximately neutral. Perhaps I
should recall that I have defined neutrality to mean that the shifts
were pure scale changes, leaving marginal rates of substitution
unchanged at given capital-labor ratios.
Not only is M I A uncorrelated with K/L, but one might almost
conclude from the graph that AA/A is essentially constant in
time, exhibiting more or less random fluctuations about a fixed
Figure 3 mean. Almost, but not quite, for there does seem to be a break
at about 1930. There is some evidence that the average rate of
results, but I have no doubt that refinement of this and the capital progress in the years 1909-29 was smaller than that from 1930-
time-series would produce neater results. 49. The iirst twenty-one relative shifts average about 9/10 of
In any case, in 2a or 2b one can replace the time-derivatives 1 per cent per year, while the last nineteen average 2$ per cent
by year-to-year changes and calculate Aq/q--wKAk/k. The per year. Even if the year 1929, which shows a strong downward
result is an estimate of A F / F or AAIA, depending on whether shift, is moved from the first group to the second, there is still a
these relative shifts appear to be neutral or not. Such a calcula- contrast between an average rate of 1.2 per cent in the first half
tion is made in Table 1 and shown in Figure 2. Thence, by arbi- and 1.9 per cent in the second. Such post hoe splitting-up of a
trarily setting A (1909) = 1 and using the fact that A(t+l) = period is always dangerous. Perhaps I should leave it that there
A(t)
. - (1 +
. AA(t)/A(t))
. . one can successively reconstruct the A(t) is some evidence that technical change (broadly interpreted)
time series, which is shown in Figure 3. may have accelerated after 1929.
I was tempted to end this section with the remark that the The overall result for the whole forty years is an average up-
A(t) series, which is meant to be a rough profde of technical ward shift of about 1.5 per cent per year. This may be compared

350 Technical Change and the Aggregate Production Function

 Table 1 Data for Calculation of A(t) -

Priv. izon- E~lzployed

, Yeor % labor Capital Col. 1 x Col. 2 Share of farm G N P capital per
force stock ($ mill.) propert}, in
iizcorrzcr per man-hour. n~arz-hour.
(4) (5) (6)

Notes and sources

Column 1 :Percentage of Iabor force employed. 1909-26, from Douglas, Real Wages itt the United States (Boston and New York,
1930), p. 460. 1929-49, calculated from The Economic Almatrac, 1953-4 (New York, 1953), pp. 426-8.
Column 2: Capital stock. From Goldsmith, A Strtdy of Saving in the United States, vol. 3 (Princeton, 1956), pp. 20-21, sum of
columns 5, 6, 7, 9, 12, 17, 22, 23, 24.
Column 3 : 1 x 2.
Cohmn 4: Share of property in income. Compiled from The Economic Almanac, pp. 504-5; and Burkhead (1953), pp. 192-219.
Depreciation estimates from Goldsmith, p. 427.
Column 5: Private non-farm G N P per man-hour, 1939 dollars. Kendrick's data, reproduced in The Economic ~lttlanac,p. 490.
Column 6: Employed capital per man-hour. Column 3 divided by Kendrick's man-hour series, ibid.
Column 7: AA/A = A5/5-4 x A6/6.
Column 8: From 7.
[The last seven observations in column 8, for the years 1943-9, are incorrect duc to an arithmetic slip in the calculations.
This was pointed out in a note by Hogan (1958, pp. 407-I]).]
with a figure of about 0.75 per cent per year obtained by Stefan over the period 1871-1 951 about 90 per cent of the increase in
Valavanis-Vail (1955) by a different and rather less general outputper capita is attributable to technical progress. Presumably
method, for the period 1869-1948. Another possible comparison this figure is based on the standard sort of output-per-unit-of-
is w i t h t h e output-per-unit-of-input computations of Jacob input calculation.
Schmookler (1952) which show an increase of some 36 per cent It might seem at first glance that calculations of output per unit
in output per unit of input between the decades 1904-13 and of resource input provide a relatively assumption-free way of
1929-38. Our A(t) rises 36-5 per cent between 1909 and 1934. llleasuring productivity changes. Actually J think the inlplicit load
But these are not really con~parableestimates, since Schrnookler's of assumptions is quite heavy, and if anything the method pro-
figures include agriculture. posed above is considerably more general.
As a last general conclusion, after which I will leave the in- Not oi~lydoes the usual choice of weights for conlputing an
terested reader to his own impressions, over the forty-year period aggregate resource-input involve something analogous to my
output per man-hour approximately doubled. At the same time, assumption of competitive factor markets, but in addition the
according to Figure 2, the cun~ulativeupward shift in the vroduc- criterion output divided by a weighted sun1 of inputs would seenz
tion function was about 80 per cent. It is possible to argue that tacitly to assunze (a) that technical change is neutral, and (b) that
about one-eighth of the total increase is traceable to increased the aggregate production function is strictly linear. This explains
capital per man-hour, and the remaining seven-eighths to techni- why numerical results are so closely parallel for the two methods.
cal change. The reasoning is this: real G N P per man-hour We have already verified the neutrality, and as will be seen subse-
increased from $0.623 to $1-275. Divide the latter figure by quently, a strictly linear production function gives an excellent
1-809, which is the 1949 value for A(t), and therefore the full fit, though clearly inferior to some alternative^.^
shift factor for the forty years. The result is a 'corrected' G N P
per man-hour, net of technical change, of $0.705. Thus about The aggregate production function
eight cents of the sixty-five cent increase can be imputed to Returning now to the aggregate production function, we have
increased capital intensity, and the remainder to increased earned the right to write it in the form la. By use of the (practic-
prod~ctivity.~ ally unavoidable) assumption of constant returns to scale, this
Of course this is not meant to suggest that the observed rate of can be further simplified to the forn~
technical progress would have persisted even if the rate of invest-
ment had been milch smaller or had fallen to zero. Obviously
much, perhaps nearly all, innovation must be embodied in new which formed the basis of Figure 1. It was there noted that a
plant and equipment to be realized at all. One could imagine this simple plot of q against k would give a distorted picture because
process taking placewithout net capital formationas old-fashioned of the shift factor A(t). Each poinl: would lie on a different
capital goods are replaced by the latest nlodels so that the capital- member of the fanuly of production curves. But we have now
labor ratio need not change systen~atically,But this raises prob- 6. For an excellent discussion of some of the probiems. see Abrarnovitz
lems of definition and measurement even more formidable than (1956). Some of the questions there raised could in principle be answered
the ones already blithely ignored. This whole area of interes~has by the method used here. For example, the contribution of improved
been stressed by Fellner. quality of the labor force could be handled by introducing various levels of
For comparison, Sololnon Fabricant (3954) has estimated that skilled labor as separate inputs. 1 owe to Professor T. W. Schultz a height-
ened awareness that a lot of what appears as shifts In the production
5. For the first half of the period, 1909-29, a similar computation function must represent improvement in the quality of the labor input, and
atiributes about one-third o r the obser\ ed increase in G N P per man-hour therefore a result of real capital formation of an imporlact kind. Nor ought
to increased capital intensity. it be forgotten that even straight technical progress has a cost side.

354 Technical Change and the Aggregate Production Function

provided ourselves with an estimate of the successive values of the cedure was designed to purify thosepoints from shifts in the func-
shift factor. (Note that this estimate is quite independent of any tion, so that way out would seem to be closed. I suspect the
hypothesis about the exact shape of the production function.) explanation may lie in some systematic incomparability of the
capital-in-use series. In particular during the war there was almost
It follows from 3 that by plotting q ( t ) / A ( t )against k(t) we reduce
certainly a more intensive use of capital services through two- and
all the observed points to a single member of the family of curves
in Figure 1, and we can then proceed to discuss the shape of three-shift operation than the stock figures would show, even
with the crude correction that has been applied. It is easily seen
that such an underestimate of capital inputs leads to an over-
estimate of productivity increase. Thus in effect each of the
affected points should really lie higher and toward the right. But
further analysis shows that, for the orders of magnitude involved,
the net result would be to pull the observations closer to the rest
of the scatter.
At best this might account for 1943-5. There remains the post-
war period. Although it is possible that multi-shift operation
remained fairly widespread even after the war, it is unlikely that
this could be nearly enough to explain the whole discrepancy.'
One might guess that accelerated amortization could haveresulted
in an underestimate of' the capital stock after 1945. Certainly
other research workers, notably Kuznets and Terborgh, have
produced capital stock estimates which rather exceed Goldsmith's
at the end of the period. But for the present, I leave this a mystery.
In a first version of this paper, I resolutely let the recalcitrant
observations stand as they were in a regression analysis of
Figure 4, mainly because such casual amputation is a practice I
deplore in others. But after some experimentation it seemed that
to leave them in only led to noticeable distortion of the results.
So, with some misgivings, in the regressions that follow I have
omitted the observations for 1943-9. It would be better if they
Figure 4 could be otherwise explained away.
Figure 4 gives an inescapable impression of curvature, of
f(k,l) and reconstruct the aggregate production function. A persistent but not violent diminishing returns. As for the pos-
scatter of q / A against k is shown in Figure 4. sibility of approaching capital-saturation, there is no trace on this
Considering the amount of a priori doctoring which the raw gross product level, but even setting aside all other difficulties,
figures have undergone, the fit is remarkably tight. Except, that is, such a scatter confers no particular license to guess about what
for the layer of points which are obviously too high. These happens at higher KIL ratios than those observed.
maverick observations relate to the seven last years of the period, 7. It is cheering to note that Professor Fellner's (1956, p. 92) new book
1943-9. From the way they lie almost exactly parallel to the main voices a suspicion that the post-war has seen a substantial increase over pre-
scatter, one is tempted to conclude that in 1943 the aggregate war in the prevalence of multi-shift operation.
production function simply shifted. But the whole earlier pro-
. --
J 1
'i The results of fitting these five curves to the scatter of Figure 4
1 As for fitting a curve to the scatter, a Cobb-Douglas function
are shown in Table 2.
i
J
comes immediately to mind, but then so do several other para-
metric forms, with little to choose among them.8 I can't help The correlation coefficients are uniformly so high that one
feeling that little or nothing hangs on the choice of functional hesitates to say any more than that all five functions, even the
form, but I have experimented with several. In general I limited linear one, are about equally good at representing the general
I shape of the observed points. From the correlations alone, for
myself to two-parameter families of curves, linear in the para-
meters (for com?utational convenience), and at least capable of what they are worth, it appears that the Cobb-Douglas function
i exhibiting diminishing returns (except for the straight line, which 4d and the semi-logarithmic 4b are a bit better than the other^.^
t on this account proved inferior to all others). Since all of the fitted curves are of the form gw = a+bh(x),
b
one can view them all as linear regressions and an interesting test
1 The particular possibilities tried were the following:
of goodness of fit proposed by Prais and Houthakker (1955,
I q = a+Pk p. 51) is available. If the residuals from each regression are
i q a + P log li
= arranged in order of increasing values of the independent variable,
1 q a-p/k
=
then one would like this sequence to be disposed 'randomly'
i logq = a+Plog k about the regression line. A strong 'serial' correlation in the
j

ik log q = a- B/k. residuals, or a few long runs of positive residuals, alternating

with long runs of negative residuals, would be evidence of just

i
Of these, 4d is the Cobb-Douglas case; 4c and 4e have upper
asymptotes; the semi-logarithmic 4b and the hyperbolic 4c must that kind of smooth departure from linearity that one would like
to catch. A test can be constructed using published tables of
f Table 2 critical values for runs of two kinds of elements.

Ij:
This has been done for the linear, semi-logarithmic and Cobb-
Curve a B i' Douglas functions. The results strongly confirm the visual in-ipres-
4a 0-438 0-091 0.9982 sion of diminishing returns in Figure 4, by showing the linear
b 0.448 0.239 0.9996 9. It would be foolhardy for an outsider (or maybe even an insider) to
c 0-917 0.618 0.9964 hazard a guess about the statistical properties of the basic time series. A few
1 d -0.729 0.353 0.9996 general statements can be made, however. (a) The natural way to introduce
e -0.038 0-913 0.9980
Q -
an error term into the aggregate production function is multiplicatively:
(1 +I~)F(K,L;I).In the neutral case it is apparent that the error factor
will be absorbed into the estimated A(r). Then approximately the error in
cross the horizontal axis at a positive value of k and continue AAIA will be Au/l +it. If rc has zero mean, the variance of the estimated
ever more steeply but irrelevantly downward (which means only AAIA will be approximately 2(1 -p) var rc, where p is the first autocorrela-
that some positive k must be achieved before any output is tion of the ii series. (b) Suppose that marginal productivity distribution does
not hold exactly, so that KlQ6Ql2K = )tlk -Lo, where now o is a random
forthcoming, but this is far outside the range of observation); deviation and )v, is the share of property income. Then the error in the
4e begins at the origin with a phase of increasing returns and ends estimated AAIA will be c Alilk, with variance (Aklk)' var 21. Since K1.L
with a phase of diminishing returns - the point of inflection changes slowly, the multiplying factor will be very small. The effect is to
occurs at k = 812 and needless to say all our observed points bias the estimate of AAIA in such a way as to lead to an overestimate when
property receives less than its nlarginal product (and k is increasing).
come well to the right of this. (c) Errors in estimating A(t) enter in a relatively harmless way so far as the
8. .4 discussion of the same problem in a diflerent context is to be found regression analysis is concerned. Errors of observation in k will be more
in Prais and Houthakker (1955, pp. 82-8). See also Prais (1952-3, pp. serious and are likely to be large. The effect will of course be to bias the
87-104). estimates of B downward.

358 Technical Change and the Aggregate Production Function

function to be a systematically poor fit. As between 4b and 4d Capital-saturation would occur whenever the gross marginal
there is little to choose.1° product of capital falls to 0.03-0.05. Using 4b, this would happen
at KIL ratios of around 5 or higher, still well above anything
A note on saturation ever observed."
It has already been mentioned that the aggregate production
Summary
function shows no signs of levelling off into a stage of capital-
saturation. The two curves in Table 2 which have upper asymp- This paper has suggested a simple way of segregating shifts of the
totes (c and e) happen to locate that asymptote at about the same aggregate production function fro111 movements a l ~ n git. The
place. The limiting values of q are, respectively, 0.92 and 0-97. method rests on the assunlption that factors are paid their mar-
Of course these are both true asynlptotes, approached but not ginal products, but it could easily be extended to monopolistic
reached for any finite value of k. It could not be otherwise: no factor markets. Among the conclusions which emerge from a
analytic function can suddenly level off and become constant crude application to American data, 190949, are the following:
unless it has always been constant. But on the other hand, there is
no reason to expect nature to be infinitely differentiable. Thus
I. Technical change during that period was neutral on average.
any conclusions extending beyond the range actually observed in 2. The upward shift in the production function was, apart from
Figure 4 are necessarily treacherous. But, tongue in cheek, if we fluctuations, at a rate of about 1 per cent per year for the first
take 0.95 as a guess at the saturation level of q, and use the linear half of the period and 2 per cent per year for the last half.
function 4a (which will get there first) as a lower-limit guess at 3. Gross output per man-hour doubled over the interval, with
the saturation level for k, it turns out to be about 5.7, more than 874- per cent of the increase attributable to technical change and
twice its present value. the remaining 12+ per cent to increased use of capital.
But all this is in terms of gross outprct, whereas for .analytic
purposes we are interested in the net productivity of capital. 4. The aggregate production function, corrected for technical
The difference between the two is depreciation, a subject about change, gives a distinct impression of diminishing returns, but
which I do not feel able to make guesses. If there were more the curvaturc is not violent.
certainty about the meaning of existing estimates of depreciation,
especially over long periods of time, it would have been better to
conduct the whole analysis in terms of net product. ABRAMOVITZ, M. (1956), 'Resource and output trends in the U S
However, one can say this. Zero net marginal productivity of , since 1870', Ariier. t.corz. Rer., Pop. Proc., \!ol. 36.
B U R K H E A D , J. (1953). 'Changes in the functional distribution of income',
capital sets in when gross marginal product falls to the 'mar- 3. An~er.Star. Assac., vol. 38.
ginal rate of depreciation', i.e. when adding some capital adds FABRIA CN T , S. (1954), 'Economic progress and economic change',
only enough product to make good the depreciation on the 34th Annual Reporr oj' the Ntrtiorral Brrrcarr of' Ecorzornic Researcll,
increment of capital itself. Now in recent years N N P has run a New York.
FELL N E R , W. (1956), Trc~r~ds nr~dCycles i r ~Ecoriorriic ..lcril;ity, Holi,
bit over 90 per cent of G N P , so capital consumption is a bit Rinehart & Winston.
under 10 per cent of gross output. Ffom Table 1 it can be read that
capital per unit of output is, say, between 2 and 3. Thus annual I I . And this is under relatively pessimistic assumptions as to how
depreciation is between 3 and 5 per cent of the capital stock. technical change itself affects the rate of capital consumption. A narning is
in order here: I have left Kendrick's G N P data in 1939 prices and Gold-
10. The test statistic is R, the total number of runs, with small values smith's capital stock figures in 1929 prlces. Before anyone uses the B a of
significant. For 4a, R = 4; for 4b, R = 13. The 1 per cent critical value in Table 2 to reckon a yield on capital or any similar number, it is necessary
both cases is about 9. to convert Q and K to a comparable price basis, by an easy calculation.

360 Technical Change and the Aggregate Production Function