Daniel M. G. Raff *
Earlier versions of this paper were extemporized at the First International Meeting of GERPISA and laid out more carefully at the 1993 Annual Meetings of the Economic History Association. I found the comments on those occasions of Robert Boyer and Ross Thomson particularly valuable, I thank for long-standing encouragement and useful advice Alfred Chandler and David Landes, Sidney Fine and David Hounshell, Kim Clark and Robert Hayes, and Richard Nelson. My French colleagues in GERPISA, most notably Robert Boyer, Yves Cohen, Michel Freyssenet, and, above all, Patrick Fridenson have also been very helpful. I equally owe thanks to my collaborator Timothy Bresnahan and to Donald Critchlow, Thomas De Fazio, Takahiro Fujimoto, Bruce Greenwald, Mr. Jozlin the Packard buff, Mike Scherer, Steven Tolliday, and Daniel Whitney. John Sutton has also been an ongoing source of stimulus. The underlying research has been supported by the Harvard Business School, the Columbia Business School, and the National Science Foundation. The usual disclaimer applies.
Are there really alternative industrial models for automobile
manufacturing? Are there really different trajectories? These
two questions are central to the projects of GERPISA and of this
working group respectively. They are closely and, in fact, intimately
related to one another. But they are not the same question. Getting
clear on how they are related seems to me to be a very
important step in our producing a useful final version of our
work.
To begin at the beginning, the only interesting version of the
question about alternative models at a strictly formal level is
whether there are distinct alternative models--that in alternative
means of organizing work, logistics, and physical production and
sales--that are actually viable in some relevant competitive environment.
One might want to assess viability in the context of a war of
all against all or in some more limited but still realistic context.
But the point is that one needs to recognize that our subject
matter is business, not some sort of physics of the factory, and
therefore that no system that fails to turn a reasonable profit
will persist. If a system loses too much money--more precisely,
too much money relative to the costs of scrapping it and starting
over in some other way, the owners will simply insist upon change.
It is worth being clear that the criterion identified in the preceding
sentence is not the naive economic one that all systems produce
identical rates of return in equilibrium. It is not even the somewhat
more sophisticated simple idea that the rates of return should
be equal adjusting for risk, somehow defined. It is that incremental
change seems profitable over some specific horizon given
the operating circumstances (the states of the world) likely to
arise in that interval.
All this being so, the question about models is a question about
steady states (or at least about something like the ergodic states
of a stochastic system). It is a question about where, in a parametrized
space of possible systems, actual firms might settle. It is a
question, posed in a somewhat abstract way, about what the possibilities
are.
Let us suppose that a careful examination of this question reveals
that there really are distinct possible alternatives. Firms might
in principle attain different destinations. The question about
trajectories is then a question about how firms get to wherever
they might be going. Given the opportunities and costs as these
will be perceived by firm decision-makers, is it true that all
viable end-points are attainable from all actual starting points?
Or will some apparently distinct paths inevitably turn into apparently
different ones over time? Will some paths inevitably (or with
a high probability) lead to extinction i.e. exit rather than to
a steady state with positive employment and output and with positive
enough cash flows?
The main thesis of this paper is that there may be a logic to
the economics of trajectories, or at least to the interaction
between operating circumstances and that logic, that constrains
which potentially viable end states are in fact possible. Given
the motives of firm decision-making, it may well be true of some
pairs of starting points and potential end points that,
as the old joke has it, "You can't get there from here".
In order to say something about practical future possibilities,
one just understand this logic and what it might rule out.
I propose to make this point not as a piece of the high theory
of dynamical sytems but rather in terms of the history of the
automobile industry. That history is less transparent than it
is generally taken to be. The usual analysis of the explosion
of output in American motor vehicle manufacturing between, say
the manufacturing censuses of 1909 and 1929 refers to the appearance
of mass production techniques at Ford and says little more. The
usual story of the Depression years that followed, abrupt and
stark contraction, followed by a slow return to something like
pre-Crash conditions--is told in representative firm language,
as if the industry was one big factory adjusting at the margin.
But in 1929, the industry was not Ford writ only slightly larger.
As I will discuss below, production techniques were quite heterogeneous
in 1929. And there was a significant amount of entry into the
population of establishments even over the next six years. The
patterns of exit were quite complex. There was a lot of activity
going on behind the aggregates. Furthermore, manufacturers weren't
stuck with techniques in the same way a farmer is stuck, at least
inframarginally with the quality of his land. The story of the
diffusion of mass production methods turns out to be a much more
complicated and thought-provoking one. I will argue that the story
of what is to come next seems likely to involve the same issues
and principles .
In exposing this hidden history, it is helpful to begin with the
product. Automobiles are at once familiar and strange to us. Few
among the many who use cars on a daily basis have thought much
about how the car parts fit together, never mind how they are
made or about the relationship between the two.
We might divide the car's components into three broad groups. The car is a vehicle that moves forward by virtue of various parts moving round and round. The power for this comes from an engine in which linear motion is converted into rotary motion. The drive train transmits this rotary motion to the wheels. Thus there is an entire category of parts that rotate. These parts are entirely metal. They are both dense and heavy. The second broad category is coach- and interior work. This is the part of the car visible to the customers. The materials are fabric, leather, wood, light metal, and paint and varnish. The residual category we might call the frame.
It comprises all the parts that hold the components and assemblies
of the first two together. Its elements are rigid but invisible.
A reasonable caricature of the old production process is as follows.
The firms that assembled the cars did not make the parts. The
parts may or may not have been made with American System methods;
but in general the tolerances were not very tight. Parts were
delivered to the assembly firm in relatively large quantities.
The parts sat until they were needed. Sub- and final assembly
took place in poorly lit general purpose rooms. There were typically
workbenches around the perimeter, on which could be found vices,
mallets, files/ and other instrument useful in making recalcitrant
pieces of metal fit togethex. There were also general purpose
machine tools around the perimeter, used for the same purpose.
(Fancier equipment on site appears to have belonged to the workers
rather than the firm.)
The assemblers of metal parts were highly skilled machinists.
They needed to be. While there was a single basic design to the
product they were turning out, heterogeneity in the actual shapes
of the parts rendered their work highly unroutinized in its detail.
Their jobs required a great deal of judgement and discretion.
Those employees who worked in leather or wood faced a similar
idiosyncrasy in material shapes and pliability. Their jobs also
were judgement- and discretion- intensive. In this, the three
groups of components were all being treated the same. The process
was one of fitting.
Monitoring this process was quite difficult. Light was not always
good. Caches of raw materials and work in process obscured sight
lines. The evolution of these caches was so irregular that there
was no indirect monitoring through them. More fundamentally, control
was difficult because the primary object was not pace per se.
It was extraordinarily difficult to judge fit from a distance.
In a broad sense, judging fit was a lot of what these workers
were being paid for. Given this, mechanisms for controlling the
process were inevitably also very crude. Almost all establishments
tried to hire skilled artisans who took pride in their work and
then simply paid piece rates.
The development of mass production methods had three salient elements.
The first and most difficult involved procuring interchangeable
parts. It is well known that ideas and methods that were in principle
practical for manufacturing rifles with interchangeable parts
date from the early nineteenth century. But the development of
methods and organizational infrastructure for actually doing it,
and doing it on a large scale, is really a phenomenon of the twentieth
century. The first auto company to do this on any scale left no
records about its parts sourcing and in any case did not do it
for very long
1.
The idea lay dormant for another decade until bursting
into flower at Ford in the years 1912-1914.
The second key element was progressive assembly. Instead of leaving
the men to go to the work--at the sawhorses at the center of the
room when the part easily, off at the perimeter fitting when they
did not, hunting down some part- or subassembly hoard when more
were required, and occasionally far from the sawhorses sharpening
tools and the like--the men were fixed in position and the work
came to them. The cycle time of individual jobs dropped radically.
The man who worked the piece was not the man who kept the input
stocks, nor was he the man who kept in good working order any
machines or tools required. Work tasks themselves were routinized
and rendered more easily monitored. Being more easily monitored,
they were more easily controlled.
If the pace of work tasks could be controlled centrally, through
job design and standard setting, so too could the pace
of part requirements. Enhanced logistical control was the third
element of the innovation. This was not carried anywhere near
so far as Ohno and his associates at Toyota did in these early
years; but control certainly was improved and tasks certainly
were consciously more tightly coupled. This was in fact what made
progressive assembly an easy vehicle for monitoring.
Designing and controlling the parts production process so that
the parts came out truly interchangeable was difficult. And the
word "truly" here can be overemphasized. The parts could
be far from identical and still not require fitting: this was
simply a matter of design tolerances. Anyone who has ridden in,
listened to, or even watched a Model T under its own power will
know that the ride was not smooth. And the standard of coachwork
and interiors was intentionally not fine. Nonetheless, the significance
of assemblers' having adequately interchangeable parts was simple.
Assembly ceased being skilled work. A fortiori, it ceased being
judgement- and discretion-intensive. Relatively unskilled employees
could do it. Managers knew what they could expect each worker
to do and how long it ought to take. Given that degree of knowledge
of the basic production process tasks, implementing progressive
assembly seems, at least in the few documented cases, to have
been a relatively minor task. Logistical control does not appear,
at least in the surviving sources, to have in itself been a major
trial. The real practical difficulties on the manufacturing end
seem to have lain in conceiving the complementarities between
these activities and in implementing them more or less all at
once.
The output consequences at Ford were certainly dramatic.
Production was soon an order of magnitude larger than that of
his closest competitors and two orders larger than the industry
norm. The operational consequences of the new system were both
more profound and more complex than the shopfloor ones identified
above. Those costs, after all, were almost entirely fixed. There
were sunk costs too. Machines and tools needed to be designed.
Ford made extremely large-scale investments in machine- and tool-building
expertise and facilities and in machines and tools itself to get
the parts it needed. The machines and tools it developed were
sufficiently revolutionary that they were the subject of extensive
description and photography in the leading metal-working trade
journals as well as those of the automobile industry itself. To
the extent the company was not to be backward integrated, relationships
with vendors needed to be set up. A new human resource management
system needed to be designed and implemented. Perhaps most important,
a marketing niche needed to be created. Demand for a branded production
was not discovered but created. The public had to be educated
as to what a Ford car was and convinced that that particular bundle
of attributes and price was a desirable one. A network of dealers
needed to be set up and an infrastructure to support them designed
and implemented. In the event, these last two tasks were, respectively,
easy and not so difficults. But no one in the organization
seems to have thought them trivial matters.
Ford himself became--Marshall would surely want to note--very
publicly extremely wealthy. The techniques were widely thought
to have made the money. It is natural to ask whether there was
swift entry into the use of mass production methods.
The famous outward signs of the new practices, at least in the public's eye, were conveyor belts. It is indisputable that these became common in American automobile plants within a year or two of the publicity showered on Ford methods in connection with the Five-Dollar Day. On the surface, then, the diffusion of mass production methods in automobiles has a very different pattern from the slow and halting nineteenth century history chronicled by Hounshell 2. It is, howeverr extremely unclear from this publicity--even down to articles on the subject in trade journals--whether the factories in question had adopted the underlying Ford production practices identified above. Such sources and the small number of intra-company sources offering glimpses into the operating and investment decisions suggest, rather, that the conveyors were simply a substitute for unskilled labor 3.
The traditional literature on the history of the american automobile
industry leaps from these traces and the output data recorded
in the Manufacturing Censuses conducted in 1909, 19l4, and 1919
to the conclusion that Ford methods took the industry by storm.
The operational end of Ford methods certainly was better known
by 1919, at least in all shops that spent the war years manufacturing
standard issue products in large volumes. But the fact that the
strikes that so threatened the industry in 1919 appear to have
been crat strikes did not register
4.
Statistics about the industry's
growing output never subtract off Ford, whose giant plant at the
River Rouge, in its time the largest industrial establishment
on the continent, was constructed during this period. They do
not form estimates of the growth of per-firm output exclusive
of Ford, let alone take advantage of the data collected through
1919 on capital stock to estimate productivity growth exclusive
of Ford and compare it to the Ford performance. The capital stock
data is probably not to be taken too literally. But taking it
as given and making the most straightforward intermediate assumptions
about the timing of investment, the rest of the industry looks
nothing like Ford.
The overt history of diffusion is sketched in Hounshell's book
and is given more detail in a recent article of mine
5. The first
large-scale implementation was in relative isolation, in the Chevrolet
Division of General Motors from about 1924
6. The executive behind
it had been hired away from Ford for the purpose. The methods
began their diffusion through the product lines of the Corporation
starting in 1926 with the design and production of the Pontiac.
Sloan's challenge to the Pontiac designers was to create a new
car, to as large an extent as was feasible, out of Chevrolet
and Oldsmobile parts with already existing dies and tooling.
Internal GM documents reveal his exuberent belief they had succeeded
handsomely. The common parts strategy diffused slowly through
the Corporation's other lines through 1929. The company's main
operational response to Depression conditions was to increase
the extent. The competitive value of this approach was so obvious
to Walter Chrysler when organizing the company that now bears
his name that he wrote it into the organizational diagram
7. GM
had a separate engineering staff for each of its five divisions.
Sloan had to entreat them to share designs. Chrysler cut this
Gordian knot by establishing a single engineering office for his
entire multi-divisional corporation.
This is a story of slow but purposive diffusion. It took fifteen
years; but on the other hand, it was a matter of superior (actually,
inferior) unit costs steadily winning out. Yet such an interpretation
follows only from assuming that in 1929 Ford, GM, and Chrysler
were the industry. The overt history is the only history there
was. It is in fact true that the Big Three were responsible for
about three-quarters of the 1929 output of the Census motor vehicle
industry. Yet it comes as something of a shock to the people to
discover that the Big Three operated only one quarter of the establishments
in that year
8.
The non-Big Three establishments appear to have
been organized according to very different principles; the modal
car was not produced in the modal type of factory.
Most motor vehicle manufacturers, and indeed most car manufacturers,
ran single establishment firms with relatively short production
runs. On most of the measures deriveable from the 1929 Census
and contemporary Moody's data, their establishments look very
different from those of the Big Three. A glance at a randomly
selected issue of Automotive Quarterly will reveal that
their output looked quite different also.
This would have been obvious enough to their customers. Details
of the Census forms reveal that these firms sold principally to
two markets. The first was for utilitarian vehicles specific to
fairly idiosyncratic purposes. These were unlovely but distinctive.
They were made substantially by hand because the market was too
small to support the infrastructure of interchangeable parts.
The second was for relatively luxurious vehicles. The purchasers
of these paid for visible hand work in the bodies and interiors
and for the consequences of hand fitting in the less visible parts.
That work was patient and extremely careful balancing of all rotating
parts and then dynamic balancing of components and sub-assemblies.
While this work was not visible to the silk-sleeved customer,
anyone could feel its consequences riding in the car. The following
is from a trade journal, not advertising literature.
"To attain the degree of refinement, length of life, freedom from noise, and absence of vibration required in a car selling in the highest price range, it is essential to work to unusually close limits, and to finish many parts all over that otherwise might be left rough on some surfaces. Grinding, polishing, and lapping operations represent a considerable portion of the total machining costs of the Cadillac engine; and only a few of those operations can be done (i.e. to the requisite standard) on fully or even semi-automatic machines" 9.
The requirements of selling such output had imposed strong constraints
on factor inputs and so on the structure of costs. Much of the
equipment required tended to be general purpose rather than special
purpose. This was in itself a major contrast to the Ford-style
system. But the human capital required to make this sort of output
was very different too. The workers needed very sophisticated
stills. They rapidly developed highly firm-specific (indeed, design-specific)
knowledge.
It is natural, comparing the production processes, to speculate
about the shape of cost curves. This raises the question of which
process had the greater fixed costs. This raises the subtler question
of which fixed cost were sunk and which were not. The Ford-style
operations had large investments in plant and equipment. But much
of that was dedicated to specific products and therefore sunk.
The equipment in the artisanal-type plants tended not to be sunk:
there was, even in the Depression, a secondary market for general
purpose machine tools. Those plants' labor forces, on the other
hand, did represent sunk capital investment. There was no real
secondary market in the early 1930s for knowledge about how swiftly
to balance the rotating parts of Packard engines or gearboxes,
and there were, in any case, no labor market institutions to enable
Packard to sell the knowledge.
Bresnahan's and my recent study of a panel of establishment data
drawn from the Censuses conducted in 1929, '31, '33, and '35 mobilizes
these ideas to develop systematic clues to the covert history
of how and why this population of heterogeneous techniques and
firms evolved as it did. We do this by estimating a model of the
plant shut-down decision in which we can distinguish the effects
on those decisions of cash flow (and its drivers) from those of
recoverable (i.e. unsunk) fixed costs. Since we work from an establishment-level
database, we can also match the estimated values of unsunk fixed
costs to particular establishments and correlate those values
to estimates of, for example, the speed with which short-run labor
demand responds to shocks.
Over the course of the first half of the thirties, the industry
did have a shakeout. The representative firm vision suggests a
Marshallian (survivorship) explanation, with an adverse demand
shift as the animating mechanism. The sector of this industry
whose production process was relatively intensive in highly skilled
workers was indeed essentially wiped out during the Depression.
But it was not wiped out, as the auto history traditional literature
suggests, because its average cost curve was too high. High-end
plants had better, not worse, margins than the mass production
facilities. (Presumably product differentiation protected them.)
Shut-down was presumably driven by total revenues being insufficient
to cover the recoverable fixed costs. On our calculations, those
costs were in fact highest for the relatively artisanal plants.
Our investigation concluded that the economic source of those
costs was the treatment of blue-collar labor as a quasi-fixed
asset. The mass production plants behaved very differently, of
course.
What actually happened then was this. The Depression struck. Most
of the fixed costs of the mass production firms were sunk i.e.
avoidable. So they contracted like mad at the margin and
rode out the storm. The artisanal firms could not do this. Most
of their fixed costs were effectively unsunk. They could only
contract by letting their workforce go i.e. dissipating many of
the assets on the basis of which they competed. No one thought
the Depression would last as long as it did. Optimistic, the artisanal
firms hung on to their workforces as long as they could. Eventually,
they ran out of money and went out of business. Other sunk investments
of the Big Three had created serious barriers to entry. Once out,
the artisanal firms were never able to return.
Why did the non-Big Three producers not switch over when demand
was booming? To ask the question so bluntly is to suggest they
faced the same choices as the Big Three did. But they didn't.
They had not made the Big Three's sunk investments in manufacturing.
They had made very different ones downstream. For brand names
are also capital investments. Cadillac had so heavily invested
in its brand identity and distribution infrastructure that it
could significantly change its manufacturing methods in the early
thirties without causing any radical revolt from customers or
dealers
10.
Launching a new model or moving into a new market required
new and substantial sunk expenditures. Knute Rockne's name and
a well-produced vehicle were not sufficient to get consumers to
take a mass-produced Studebaker seriously
11. Packard also tried
to launch a car for a more popular market in 1937. Their dealers
did not know how to attract the new target buyers; and the old
ones viewed the old lines with new suspicion. The two groups of
producers thus faced fundamentally different choices.
The meaning of this story, it seems to me, is that the history
of industrial organization in automobiles is not simply a history
of the coming of machine production, played out in terms of fixed
and average costs. The history is one of path dependency, which
turns on the sunkness--an attribute not often studied--of investments
themselves not often focused upon by economic historians. Sunkness
makes for differences over time and between early- and late-comers.
It therefore makes for heterogeneity in cohorts. To decentralized
decision-makers, there may well be more than One Best Way. This
seems on the face of things as likely to be true for the lean
production model as it was for Fordist mass production.
The second main point seems to me to be that while there are important
sources of this sunkness in the realm of operations
in which students of the industry usually look--in manufacturing--there
may be equally important sources in the much less frequently surveyed
domains of upstream supply and distribution. The impulse to theorize
explicitly, to measure, and to test could in principle add a great
deal to the writing of the business and industrial organization
history of the automobile. But to address the current ferment
in thinking about the nature of firms--their competences and their
opportunities--and the evolution of markets and competition, researchers
will have to take such sources seriously. This suggests that an
illuminating history of industrial organization and of organizational
innovation must cast its evidentiary net far more widely than
we are used to doing. The usual statistical data is restricted
to counts and values of inputs and outputs, and this does not
even come close to measuring the scope of the sort of relationships
we now recognize to be critical.
My third concluding point is methodological on a level intermediate between these two. If the first was that sunk costs can matter and the second that we are not yet surveying them--never mind measuring them--systematically, the third is that some sunk costs may matter more than others and we need to sort out which ones are which. To understand the evolution of these production systems is to understand the incentives facing those who make decisions about change. Those decisions are made in the context of a set of expectations about future operating and competitive conditions. The decisions have whatever selection consequences they do in the context of the set of conditions that actually come about. One important characteristic of operating systems is their robustness to an appropriately wide range of these conditions. The sector of the American industry that largely died in the thirties did so because its systems were not robust to the conditions it confronted. As we carry on our company-level work and seek to identify alternative possible models, we should contrast the sorts of shocks and market changes each finds advantageous or is at least relatively robust to. We should try to understand the differential adaptive abilities and opportunities of each. This is the aspect of a technological trajectory that makes it interesting in studying the future.
* - Pennsylvania University, Wharton School
1 - This was the Olds Motor Company circa 1904. All the records and most of the capital stock disappeared in a freak fire. When the works went back into production, it made a very different sort of vehicle with very different methods.
2 - David A. Hounshell, From the American System to Mass Production: the Development of Manufacturing Technology in the United States (Baltimore: Johns Hopkins University Press, l984).
3 - For two examples, see (reading closely) "Conveyor System Aids Big Production", The Automobile, July 20, 1916, pp. 100-104, and (completely on the surface) Minutes of the Studebaker Corporation Finance committee for 9 February, 1914, in the Studebaker National Museum, South Bend, Indiana.
4 - See the articles on the strikes in Automotive Industries for
April 25, May 1, and May 8, 1918 (pp. 841, 972, and 1028 respectively).
5 - Daniel M.G. Raff, "Making Cars and Making Money in the Interwar Automobile Industry: Economies of Scale, Economies of Scope, and the Manufacturing that was Behind the Marketing", Business History Review, 65 (4) (1992), pp. 721-753.
6 - The process is described in convincing detail in William S. Knudsen, "For Economical Transportation: How the Chevrolet Motor Company Applies Its Own Slogan to Production", Industrial Management 76 (XXX) (1927), pp. 65-68.
7 - Personal communication from Bruce Thomas, Chrysler Engineering, 1991.
8 - For this and the following paragraph, see Bresnahan and Raff, "Intra-Industry Heterogeneity and the Great Depression: The American Motor Vehicles Industry 1929-1935", Journal of Economic History, 1991.
9 - Herbert Chase, "Practice in Precision by Cadillac", American Machinist 73 (August 28, 1930), pp. 359-361.
10 - Steve Babson of Wayne State University has interviewed a man who claims to have assembled Cadillac engines by hand in 1931. By the mid-thirties, Cadillac was involved in the common parts strategy described above.
11 - Studebaker's bankruptcy was a strictly financial event, caused by a stubborn dividend policy and set off by the Bank Holiday. The receivership passed smoothly and swiftly precisely because there was plenty of demand for the cars.