Industrial Dynamics in a Historical Setting
Cliometric Sessions at 1990 ASSA Meetings--December 30, 10:15 AM


4 Intra-Industry Heterogeneity, Composition Effects,
and the Supply Response to the Great Depression:
Lessons from a Seven-Industry Panel

Timothy F. Bresnahan
Stanford University


Daniel M.G. Raff
Harvard University

The subject of this paper is the microeconomic behavior of the American industrial economy in the period 1929-1935. The period spans the largest decline in demand ever to strike the manufacturing sector. Understanding the response of the supply side to the shock, both in output markets and in factor markets such as the market for labor, is of enormous analytic importance and central to understanding the economic history of the period.

By and large, data availability has in the past limited analysts to and aggregated view of the economy. This is unfortunate. We have known for fifty years that there was a very wide diversity of output, price, and employment experience across industries in the 1930s. This diversity presumably has it roots in the variety of economic environments--most importantly, the technologies in place and so the cost structures--facing individual firms. Examination of the origins of diverse economic outcomes in terms of the heterogeneity of individual productive units has been very fruitful for studies of the postwar period and for studies of the nineteenth century economy. We believe that extension of these methods to studies of the Great Depression will provide substantial knowledge.

The novelty of our enterprise lies in part in the attention we pay to economy-wide extreme behavior of the official statistics. Briefly, these trends were the emergence of large establishments (typically furnished with expensive capital goods) and multi-plant firms, employing semi- skilled labor at relatively routinized tasks under centralized, professional managerial control. Scale economies in all of this led to relatively concentrated industries. But these secular trends were very strong in some particular industries and nearly absent in others. The experience of changes in price, output, and employment was, as one might then expect, extremely heterogeneous across industries.

A simple comparison can fix ideas about the inter-industry contrasts. In motor vehicles, the mass-production methods and organization dominated in the operations of the leading three firms. These firms accounted for a substantial majority of the industry's output. They accounted, however, for only about a third of the establishments; and these establishments behaved quite differently from those outside the Big Three in a number of economically important respects. In the influence of mass production methods upon the bulk of the industry's production, the motor vehicle industry was a leading sector in an economywide trend. But the trend was hardly uniform. Cotton textiles was another substantial industry. It was not without change in this period: production was still moving south, and new vertical relationships were developing. But its technology was relatively static and its industry structure was in all other respects scarcely evolving at all. Thus there is considerable variety beneath the macro aggregates. Careful statistical methods and thorough research into the technological context of the numbers are clearly prerequisite to sorting out what was actually happening.

Aggregate or industry level data cannot support such research: firm- or establishment-level panel data is required. We have discovered and begun to augment and exploit in our research a new statistical data source which can supply it. This data source is the manuscript returns to the Censuses of Manufacturing conducted in 1929, 1931, 1933, and 1935. We are coding, editing and panelizing records from this establishment-level source.

We presented a preliminary study of one industry's worth of the data at the EHA meetings in Montreal.1 The industry was motor vehicles. It is well-known that the Great Depression hit this industry quite hard. The fact about the incomplete diffusion of mass production methods is much less widely appreciated. This partial diffusion of the technology and related organizational changes formed the initial conditions for motor vehicle competition in the Great Depression. Those firms whose plants and organizations embodied mass production were at a competitive advantage.

There were two separate routes by which this advantage could have played out. Low average cost firms can survive bad times where less efficient ones fail. Second, low marginal cost firms can have high market shares among survivors, as competition assigns more output to the lower- cost firms. We found that only the first of these two routes was important for motor vehicles in the Great Depression. But it was very important, at the downturn caused a major shakeout of the inefficient.

Our evidence for the shakeout view seems very solid. Half of plants, and more that half of firms, closed between the 1929 peak and the 1933 trough. The closures accounted for almost a third of the decline in industry employment. Output per worker in the industry did not decline nearly as much as output per worker at a typical continuing plant from 1929 to 1933. Since the existing plants were low in labor productivity and since their numbers were large in the aggregate, they account for this large composition effect. This was truly a shakeout; the following upturn was largely accomplished by expansions by existing plants and firms, not by replacement of the former exiters with new concerns.

Evidence for the technical heterogeneity view of the shakeout also seems solid. A group of variables, plausibly proxying mass production techniques and organization, predicted plant closure. Were these really mass production proxies? The same linear function of them that predicts closure also predicts 1929 size in the cross section of plants. Thus, they do appear to be related to large plant scale, which is in effect induced by the techniques.

Plant shut down was the only clear size-related source of composition effects in the data. Conditional on survival, we calculated, all plants tended to contract proportionally. Both large and small surviving facilities tended to employ about half as many wage earners at the trough as at the peak.

This paper extends our reach to a more diverse group of industries.2 The further industries are3 cotton textiles, linoleum, matches, blast furnace products, rubber tires, and cigarettes. We had a number of criteria for inclusion in our first panel. We wanted to select Census industries that contained some 4-digit subsets whose definitions did not change over the panel. We wanted a group that spanned the range of competitive conditions from roughly perfectly competitive through loose oligopolies through tight ones and dominant firms to the other extreme of virtual monopoly. We wanted a range of technology types represented, from capital-intensive and tightly integrated through labor-intensive and fragmentary. We sought a mix of producer and consumer goods and of durables and non-durables. We wanted semi-manufactures and finished goods, and industries that drew their raw materials from a variety of sources--agriculture as well as mines and forests. The sample is obviously preliminary from the point of view of formal hypothesis testing. But we expect on the basis of it to be able to speculate in a relatively structured fashion.

The broad themes suggested by our research to date are as follows. There was technological heterogeneity of a sort that had significant competitive and dynamic consequences in important sectors. Sectors were also changing in importance vis a vis one another with respect to output and employment. Thus technological history matters for good empirical microeconomics; and composition effects matter in measuring structural and other quantitative relationships at both industry and more aggregated levels. Representative firm approaches to the empirical economic relationships in the economy in this conspicuously interesting period can be, and for specifications we have estimated certainly are, seriously misleading.

1 Copies of this paper are available from Raff at Cotting House 215, Harvard Business School, Soldiers Field, Boston, MA 02163.
2 We are pursuing our study on the intensive margin as well. Copies of a first draft of a further motor vehicles paper (incorporating, among other things, interpretation of the interesting cross-sectional pattern of short-run labor demand) should be available in December or early January. Again, please write Raff if you would like a copy.
3 (as of this writing--there may be more by December)