THE JOURNAL OF ECONOMIC HISTORY

Economic Sources of Inventive Activity

The fundamental conclusion of this paper is that technological
progress is intimately dependent on economic phenomena. The
evidence suggests that society may indeed affect the allocation of
inventive resources through the market mechanism somewhat as it
affects the allocation of economic resources generally. If this is true,
then technological progress is not an independent cause of socio-
economic change, and an interpretation of history as largely the attempt
of man- kind to catch up to new technology is a distorted one. Cultural
lags undoubtedly exist in social history. The automobile-to use an
obvious example-rendered obsolete many pre-existing social
ar r angem e nts and behavior patterns. But the reverse is also true. New
goods and new techniques are unlikely to appear, and to enter the life
of society with- out a pre-existing, -albeit possibly only latent-demand.
Even a long- standing demand may have been intensified shortly before
a technique to satisfy it is invented. In addition to cultural lag, there
exists technological lag-a chronic tendency of technology to lag behind
demand.1
The problem considered in this paper may be viewed in still another
way. If one charts over time a firm's research and development
expenditures leading to a given product (for example, a given model of
railway car) and the volume of sales of the resulting product, one would
expect
• An earlier version of this paper entitled "Growth, Cycles, and Invention" was read at the meetings
of the American Association for the Advancement of Science, Section K, New York City, Dec. 26,
1960.
I wish to acknowledge the helpful criticisms of Simon Kuznets and of my colleagues at the
University of Minnesota; the financial assistance of the National Science Foundation, the Ford
Foundation, the John Simon Guggenheim Foundation, and the University of Minnesota; the co-
operation of the United States Patent Office in the derivation of the patent statistics used; and the
research assistance of Manmohan S. Arora, Heinrich H. Bruschke, A. Luis Darzins, Irwin Feller,
Sushila Gidwani, Donald W. Katzner, and Allan L. Olson. The opinions expressed arc mine.
1 Cf. S. C. Gilfillan, The Sociology of Invention (Chicago: Pollet Publishing Co., 1935), esp. ch. 1.
in the nature of the case to find-if the product were marketed at all- that
the research expenditures were made largely before the product was
marketed.2 What would one find if total research and development
expenditures of all firms on a given class of product, such as railroad
passenger cars, were compared over a period of time with sales of all
products in the class? On this question, the data examined in this paper
throw some light. Inventive effort, it would appear, usually varies
directly with the output of the class of goods the inventive effort is
intended to improve, with invention tending to lag slightly behind
output. The explanation of this pattern, in my judgment, is that
variations in invention are a consequence of economic conditions with
which output is also positively correlated. While the discussion and
evidence relate to classes of products already in existence, the conclusion
may well apply with almost equal force to basic, industry-establishing
inventions.

A few remarks about the basic data used-statistics of patents granted,

classified by industry-are in order at the outset. Beginning with 1874,
the data represent patents counted as of the time the application for the
patent was filed. This means, according to members of the patent bar,
that generally the inventions represented were reduced to patentable,
although not necessarily commercially useful, form on the average
about six months beforehand. For the period prior to 1874 the patents
are counted as of the time of granting. During this earlier period the
interval between application and grant averaged only a few months.
Such data are subject to criticism because their use entails three un-
proved assumptions: (1) that the ratio of patents to inventions within the
field is constant over time, (2) that the ratio of the number of patented
inventions in the field to the number of patents covered by the particular
series is constant over time,8 and (3) that the average invention in one
period in the field is of the same importance, in some relevant sense, as
the average invention in any other period. The validity of these
assumptions cannot be established. Moreover, since
2 Cf. Gordon K. Johnson and Inez M. Turner, "Use of Transfer Functions for Company Planning,"
Operations Research, 5 (Dec. 1956), 705-10.
8 The statistics used in this paper were compiled by taking all the patents in those Patent Office
subclasses in which two thirds or more of the patents included in the subclass pertained to the industry.
For the fields discussed in this paper, at least 95 per cent of the patents pertain to the industries to which
they have been assigned. However, patents pertaining to the industry assigned to other subclasses arc
not included. The data will be published and discussed more
fully in a forthcoming book.
1940, there may have been a marked decline in the proportion of inventions paten
provide the most effective means for a comprehensive historical investigation of the p
The fact that the data seem to behave reasonably, albeit remarkably, suggests that
are tolerable, at least for the series dis- cussed here.
 rther preliminary points need to be made about the data:
(1) The proportion of patented inventions used commercially is apparently larg
Patent Foundation in Washington, which are consistent with considerable indep
indicate that between 40 and 50 per cent of inventions patented by independent inven
55 and 65 per cent of those patented by corporations are used commercially. In shor
reflect the vaporing’s of disordered minds.
(2) In the two industries for which we have been able to compile the most relia
of major inventions-railroading and petroleum refining-the number of major inv
vary directly with the number of patents.4 This suggests that the inferences drawn f
of the minor inventions, which unquestioned- ably dominate the patent statistics
in considerable measure to major inventions too.

Consider first the hypothesis that variations in the output of a class of products ar
corresponding variations in intention. Figure I shows the annual number of patents in
1846 to 1950, with the patents subdivided into two groups: those pertaining to railr
other railroad patents. These two comprehensive components of railroad invention
similar behavior. They reach their all-time peaks within three years of each other, th
are similarly timed, and even minor fluctuations in each tend to be duplicated i
pervasive similarity suggests that common external forces shaped the course of i
railroad fields.
We get an inkling of what those common forces may be from Figure which show
moving averages of four variables total railroad patents, net changes in miles of r
formation in 1929 prices, and the real price of railroad shares. The variables displa
suggest at first glance that the railroad industry became increasingly profitable until 19
end of railroad expansion and the coming of highway transportation, its profitability
As profits rose and fell, so did investment in the industry-as indicated by changes in
in the industry's gross capital formation. Invention in the industry rose and fell, too, a
and the output of railroad equipment, which the capital formation series indicates.
The next question is whether this apparent long-run relationship holds also within
Figure 3 reveals that it does, and with considerable fidelity. This graph shows dev
year moving averages from the seventeen-year moving averages of the preceding g
turning points in the various series, summarized in Table 1, usually come clos
patents usually lagging be- hind the economic indicators. Even when the major tu
• Beginning with 1874 patents are counted as of the year of application. For earlier years they are counted as of the

FIGURE 2: RAILROAD PATENTS AND RAILROAD INVESTMENT, UNITED STATES, SEVENTEEN-YEAR MOVING AVERAG
• Beginning. with 1874 patents are counted as of the year of application. For earlier years they are counted as of the y
SOURCES: See Appendix.

spread out, minor irregularities in the individual sense reveal the persistence of c
conformity between the sense.
TABLE I
LONG SWINGS IN RAILROAD INVESTMENT AND INVENTION-PERCENTAGE DEVIATIONS OF SEVEN
YEAR MOVING AVERAGES FROM SEVENTEEN-YEAR MOVING AVERAGES •

• The deviations for gross capital formation are those of a nine-year moving average from a seventeen-year moving
source provided a nine-year average.) The rest are based on seven-year moving averages.
nvestment indicators for the railroad industry as a whole is also found between inven
of specific kinds of railroad equipment. Each of the next three graphs shows the ann
major variety of rail road equipment coupled with the annual number of patents pe
equipment.
Figure 4 shows railroad rail output and patents annually from 186o

FIGURE 4: RAILROAD RAILS: 0uTPUT AND PATENTS, UNITED STATES, 1860-1950, ANNUAL D A T A •
• Beginning with 1874 patents are counted as of the year of application. For earlier years they are counted as of the ye
Source s: See Appendix.

to 1950. Inspection of the graph reveals the existence of

substantially parallel major and minor fluctuations in the
two variables. Output reaches its all-time high in 1906,
and patents in 19n. A tentative though somewhat
arbitrary list of timing of major troughs in each series is
provided in Table 2, columns (1) and (2). The
corresponding columns of Table 3 offer similar
estimates of major peaks. Rail output generally leads
patents at the troughs, but not necessarily at the peaks.
Figure 5 compares railroad passenger car output and
patents annually from 1871 to 1949. Again both
variables move in marked harmony over both the long
and the short run. Railroad passenger car output reaches
its all-time peak in 1907, railroad passenger car patents
in 1908. At the major troughs tentatively identified in
Table 2, output generally leads patents, but again, leads
and lags are roughly even at the peaks shown in Table 3.
Figure 6 offers a similar comparison for freight car
output and freight car patents, with much the same
results. In this case, the all-time peak in patents occurs
in 1907, the same year as that in output. At the major
troughs patents again usually lag, but again honors are
even at the major peaks.
The examples cited thus far all relate to the railroad
industry; does the same pattern exist in other industries?
Figure 7 shows that the all- time peak of American
invention in another part of the transport industry-
horseshoes-was reached around 1900, at the time that the
steam traction engine on the farm, and the auto, truck,
and bus began to displace the horse and just before the
output of horse-drawn vehicles
reached its peak around 1904. The output of the product
and inventions pertaining to it apparently moved
together, just as in railroads. graph also shows that
inventions pertaining to the horseshoe calk, a device
attached to the shoe of workhorses to prevent slipping,
creased paving of roads. This pattern of invention
casts doubts on the notion that invention declines in
an industry because the possibilities for further
improvement are exhausted, and that an industry's rate
of growth declines because technical progress
slackens.5 Quite the reverse-the decline in the
industry's rate of growth apparently induced the
decline in invention and technical progress.
Figure 8 presents a more important series on petroleum
refining investment and invention. These investment
data are less complete than those for railroads. Beginning
with Drake's well in 1859, investment in the industry
probably underwent two major growth cycles,
5 Cf. Simon Kuznets, Secular Movements in Production and Prices
(Boston: Houghton Miffing, 1930), ch. 1, esp. pp. 30-32; Arthurs F.
Burns, Production Trends in the United States Since 1870 (New York:
National Bureau of Economic Research, 1934), ch. 4, esp. pp. 141-45;
Robert K. Merton, "Fluctuations in the Rate of Industrial Invention," Quarterly
/journal of Economics, XX.XIX (1935), 454-74; and W. E. G. Salter,
Productivity and Technical Cliange (Cambridge: The University Press,
1960), ch. X, esp. pp. 133-34.

one associated with the kerosene phase of the industry

and the other with the gasoline phase. Petroleum
refining patents traversed a pair of similar growth cycles
about the same time. Thus the same long-run relation
between investment, or investment goods output, and
invention exists in petroleum as in railroads.
In the building industry, too, I have been able to
compile a rough but
usable long-term record of annual economic data.
Here the indicator of building activity has been
provided by splicing series which are not wholly
comparable; trend comparisons are therefore not
warranted, and Figure 9 depicts only the behavior of
deviations of seven- from seventeen-year moving
averages for building patents and building activity.
The picture here is much the same as in railroading.
Patents on building components-doors, walls, roofs,
chimneys, etc.-also exhibit long swings similar to
those observable in building activity itself. The
turning point dates, listed in Table 4, come within
two years of each other nine times out of twelve, with
patents lagging slightly more often than they lead.6
Just as in railroads, the patenting patterns in
individual subfields within the building industry, not
shown here, bear a strong family resemblance,
suggesting again that common external forces
there is a marked tendency for output and invention to
move together in the long run, in the short run, and in
intermediate periods. Since it seems implausible,
despite the lead-lag relationship, that output variations
directly produce vanat1ons in invention, two possible
explanations for this phenomenon exist: (1) the
variations in invention cause those in output; or (2)
both variables fluctuate in response to a third set of
forces as yet unspecified.
The first explanation conforms to popular belief
and to the indisputable fact that research on any given
product must precede its production and sale.
Nonetheless, three reasons count heavily against this
as a general explanation. First, it offers no explanation
for the parallel behavior of invention in the different
subfields within an industry. Further, the lead-lag
relations between output and invention are generally
in the wrong order. If the fluctuations in invention
cause those in output, the all-time peaks in invention
should come before those in output. Instead, the all-
time peaks in railroad equipment invention usually
follow, or at best coincide with, those in equipment
output. Second, if the fluctuations in invention cause
those in output, then the major troughs in invention
should precede those in output by a few years-to allow
time for inventions to reach a commercial stage, for
plants to get tooled up, and for the necessary
marketing effort to made. Instead, inventions usually
increase after rises in output, or occasionally tie or
lead by only a very short period. Finally, the
correspondence between invention and output is so
close that if we are to argue that the fluctuations in
invention cause those in output, we are in danger of
explaining too much. We in effect assign only a minor
role to the state of the economy, the rate of population
growth, and other such factors as determinants of the
output of an industry. Schumpeter, who made so
much of the role of innovation as a determinant of
economic development,7 thought investment in most
industries was ordinarily governed by system-wide
determinants. He argued only that the state of the
whole system was primarily determined by
innovations in a few industries-industries which
differed from one period to another. If the line of
causality runs neither from production of the good in
el movement of both variables must mean that both
are effects either of the same or of correlated causes.
The causes of fluctuations in the output of
commodities have long been studied by economists
and economic historians,
6 The substantial lead of building activity over patents at the start of the series
may reflect deficiencies in the economic data. This portion of the Riggleman series
is regarded as particularly unrealistic. Cf. Historical Statistics of that: United
States: Colonial Times to 1957 (Washington:
U. S. Census Bureau, 1960), p. 376, discussion of series N64. The frequency
with which patents lead building activity (four times out of twelve) may be an
illusion induced by aggregation. The long swings in building activity have
somewhat different timing in different sections of the country. Since inventive
activity may be more heavily concentrated in the leading sections, if the
inventions from each section were compared with the building activity in
that section, patents might be found to lead even less often than appears to be
the case.
and we cannot hope to shed further light on the matter here. It is sufficient
for our purposes to state that in an economic system, the whole tends to
dominate in its parts, and that more often than not variations in the output
of a given industry are traceable to developments elsewhere in the system.8
On this basis, our problem is not to account for the fluctuations in output,
but to understand the inventive process and its socio-economic connections
well enough to explain why inventive and economic activity are as closely
correlated as they appear to be. While this problem cannot yet be
definitively solved, it does seem possible at this stage, on the basis of
private discussions with and published accounts of inventors, applied
research scientists, and research directors, to suggest what may prove to be
some of the key factors at work.9
Even if the germinal ideas for inventions occurcostlessly and at
random, which is perhaps seldom the case, it often takes talent, hard work
and money to carry them out. These are not likely to be forth- coming in
volume without commensurate prospective rewards in income or prestige.
Clearly, prospective rewards will be favorably affected by high volume of
sales of the commodity to be improved. According to Harrel, "The
choosing of projects must be as precisely timed as a well-planned military
maneuver. Not only must they be precisely timed
-they must also be the correct project properly developed at a
utilizable place," 10 and quoting R. P. Soule: "The history of research
indicates that the odds in favor of the project are most favorable
when. first, the product to be developed is riding a rising rather than
a de-
s This general theoretical premise finds support, not surprisingly, in a comparison of the long
swing turning points in railroad investment and building activity. The interested reader who
compares the turning points given in Table I for railroad investment with those in Table 4 for
building activity will find that out of eleven occasions, the two variables turn in the same year
three times, within a year of each other four times, within two years of each other twice, and within
three years twice. Since the trough-to-trough length of a long swing in each industry is about
nineteen years, the presence of common external influences on both industries is strongly indicated.
(For a fascinating suggestion as to the possible nature of the causal mechanism at work, cf. Simon
Kuznets, "Long Swings in the Growth of Population and Related Economic Variables,"
Proceedings of the American Philosophical Society, Feb. 1958.) It is perhaps also pertinent to
direct attention to the substantial, but again not surprising, similarity in the timing of major peaks
and troughs in the output of major varieties of railroad equipment (Tables 2 and 3).
9 The obvious possibility that variations in the proportion of inventions patented occur in response to
variations in economic prospects will be ignored in what follows. This is justified by the fact that while
such variations may explain minor year-to-year variations in patents, the economic incentives to invent
are the same as the incentives to patent. Hence, the major movements in patenting, which have been
the focus of our discussion, can be reasonably construed as reflecting movements of invention in the
same direction, though not necessarily of the same magnitude.
10 Cf. C. G. Harrel, "Selecting Projects for Research," in C. C. Furnas (ed.), Research in lnd11stry:
lets Organization and Management (New York: D. Van Nostrand Co., 1948), p. u6.
dining industrial trend, and, second when it is started in the earlier
rather than the later phases of that trend." 11
However, the relation of invention to sales may be non-linear in the
short run; if sales get too high relative to plant and personnel, potential
inventors may be kept too busy in current production, and potential
innovators may lack both the required slack resources and the pressure
to innovate.12
To the degree that inventions are made either by producers or consumers
of a commodity, more money will be available for invention when the
industry's sales are high than when they are low. Increased sales imply that
both the producing firms and their employees will be in a better position
than before to bear the expenses of invention. Large purchases of a product
also suggest that its buyers are better able to finance invention. Finally, the
current business practice of setting re- search budgets at a fixed percentage
of sales tends to assure, in recent years, the relation we have observed.
Certain aspects of the psychology of invention are noteworthy in this
context. Effective inventive activity apparently requires an intense, almost
obsessive preoccupation with its object. Buyers, for example, are more
likely to show such a preoccupation when they are shopping for a product,
with standards and expectations about performance and cost beginning to
form in their minds. The number of potential inventors among the buyers
or producers of a product, then, might vary with sales and employment.13
Another psychological requirement for invention, which is also involved
in new purchases, is a certain dissatisfaction with existing post sessions.
Every new purchase indicates such dissatisfaction. New buildings are
erected because old ones are crowded or inadequate. New machines are
purchased because old ones are breaking down or cannot handle an
increased volume or produce unsatisfactory products. The same conditions
which lead some men to buy existing goods may lead others to try to
improve them. Such dissatisfaction with existing products may stem from
several sources whose influence usually varies directly with sales. First, in
periods of general economic expansion and rising sales, when proportions
of many kinds are changing, products which were satisfactory under earlier
conditions may be unsatisfactory
11 Ibid., p. 117.
12 This consideration may explain the frequency with which railroad patents declined
cyclically before output.
13 This may explain why invention in some branches of agriculture has declined with the number
of farmers, rather than with the sales of farm implements.
under the new. To meet the changed requirements of customers, the
producers or the customers themselves may invent new or modified
products. Second, increased sales often mean that some people have be-
gun to use a product for the first time. Since the circumstances and
preferences of new customers may differ from those of old ones, either the
producers or the new customers themselves may invent modifications in
the product to suit the altered requirements.
Finally, newly purchased goods embodying recent improvements are
also likely to have the "bugs" which seem almost the inevitable
accompaniment of progress. These emerge after use of a product has begun,
and provide the occasion for corrective invention by the user or the
producer. Such invention will tend to vary directly with, but lag behind,
sales of the product involved.
To summarize, the evidence strongly suggests that the output of a
commodity and invention relating to it vary together, with invention
tending to lag. The relation cannot be explained by the hypothesis that
the variations in invention cause those in output.14 Rather, it seems
probable that expected profits from invention, the ability to finance it,
the number of potential inventors, and the dissatisfaction which in-
variably motivates it-are all likely to be positively associated with
sales. That major and minor inventions in an established industry tend
to go together is a further indication that inventive effort is responsive
to economic pressures and opportunities.
One may suggest further that even basic inventions which establish
new industries are often, and perhaps usually, induced by economic
forces like those which appear to operate in established industries. Even
when a scientific discovery underlies an invention, the discovery may
contain the seeds of many potential inventions, and economic factors
may then determine which potential applications are selected for
exploitation. These economic factors may take the form either of new,
or newly intensified, latent demands, or of a greater intensity of the
latent demands satisfied by the inventions actually made, as compared
to that of the demands which remain unsatisfied. For example, while the
invention of wireless telegraphy cannot be understood without reference
to prior scientific discoveries in electromagnetism, it cannot be
understood with reference to those discoveries alone. Other inventions
based on the same discoveries could presumably have been made, and a
substantial and probably increased demand for rapid and low- cost
communication techniques probably influenced Marconi's efforts. The
essential point is that the incentive to make an invention, like the
incentive to produce any other good, is affected by the excess of
expected returns over expected costs. Scientific progress may reduce
expected costs and so increase the probability that a given invention will
be sought and made. However, every invention represents a fixed cost,
and the expected benefits from it vary with circumstances. Those
circumstances, arising from changes in the prospective market for a
commodity or a process, depend not on scientific discovery, but on
socio-economic change-urbanization, declining family size, changing
status of women, changes in relative factor costs, increases in
population and per capita income, etc. Antecedent scientific discoveries
are sometimes necessary, but seldom sufficient, conditions for
invention. The historical shifts in inventive attention appear to reflect
the inter- play of advancing knowledge, which opens up new inventive
opportunities for exploitation, and the unfolding economic needs and
opportunities arising out of a changing social order. The evidence
adduced in this article suggests that the influence of the latter has been
substantial, at least in established industries.
14 It is essential that the meaning of this statement should be clear. Of course, as noted earlier,
inventions affect production. It is only asserted here that their effect does not take the form of inducing
fluctuations in production which are synchronized with fluctuations in invention,

Figures 2 and 3: Railroad stock prices: F. R. Macaulay, Some Theoretical

Problems Suggested by Interest Rates, Bond Yields, and Stock Prices in the United
States Since 1856 (New York: National Bureau of Economic Research, 1938),
Appendix Table 10, and Moody's Transportation Manual, 1959, p. 3a. These were
adjusted for changes in the B. L. S. wholesale price index of all commodities,
Department of Labor, Bureau of Labor Statistics, Handbook of Labor Statistics
(1941 ed.), p. 715 (1947ed., p. 126, and subsequent supplements thereto).
Net additions to miles of road: calculated by successive subtraction from statistics
on miles of road operated given in Poor's Manual of Railroads.
Gross capital formation in 1929 prices: annual data used for calculating the
seventeen-year moving average from M. J. Ulmer, Capital in
Transportation,
Communications, and Public Utilities (Princeton: Princeton Univ. Press, 1960),
Appendix Table C-1; nine-year moving averages from ibid., Appendix Table K-2.
Figure 4: Railroad rail output: 1860-1914: Edwin Frickey, Production in the United
States, 1860-1914 (Cambridge: Harvard Univ. Press, 1947), pp. 10-rr; 1915-1929:
Arthur F. Burns, Production Trends in the United States Since 1870 (New York:
National Bureau of Economic Research, 1934), pp. 294-95; and for subsequent
years, American Iron and Steel Institute, Annual Statistical Report, various issues.
Figures 5 and 6: Railroad passenger and freight car output: 1871-1914: Frickey,
Production in the U. S., pp. 14-15; 1915: American Railway Car Institute, Railway
Age, Annual Statistical and Outlook Number, Jan. 7, 1939, p. 83, and subsequent
issues. Beginning in 1920 the data include production in railroad repair shops.
Before 1920 they do not.
Figure 8: Annual rate of net capital formation: estimated from Creamer,
Dobrovolsky, and Borenstein, Capital Formation in Manufacturing and Mining
(Princeton: Princeton Univ. Press, 1961), Table A-8.
Index of contracts awarded in petroleum refining and pipelines: data supplied
by F. W. Dodge Corp.
Figure 9: Riggleman's index of the per capita value of building permits given in
Miles L. Colean and Robinson Newcomb, Stabilizing Construction (New York:
McGraw-Hill, 1952), Appendix N, Table 2, was multiplied by the U. S. population,
Historical Statistics of the United States, Colonial Times to 1957 (Washington: U.S.
Census Bureau, 1960), Series A 2, to provide an estimate of total building in current
prices for 1830-1920. By splicing the 1913-based building cost index given in
Colean and Newcomb, Stabilizing Construction, Appendix N, Table 4, to the
American Appraisal Company index of construction costs (U. S. Departments of
Labor and Commerce, Construction Volume and Costs, 1915-1956, nod, p. 54), a
continuous index of building costs (1947-1949 100) from 1830 to 1950 was
secured. The latter was then used to convert the index of total building from 1830-
1920 in current prices into an index of building in 1947- 1949 prices. The post-
1920 index of building activity was derived by subtracting public utility
construction in 1947-1949 prices (ibid., p. 14) from total construction in 1947-1949
prices (ibid., p. 39). Public utility construction was eliminated to make the series
more comparable with that derived from the Riggleman data and with the patent
statistics.