Why Did Productivity Fall So Much During the Great Depression?

This study assesses five common explanations for the large decline in U.S. total factor productivity (TFP) during the Great Depression: changes in capacity utilization, factor input quality, and production composition; labor hoarding; and increasing returns to scale. The study finds that these factors explain less than onethird of the 18 percent TFP decline between 1929 and 1933. The rest of the decline remains unexplained. The study offers a potential explanation: declines in organization capital, the knowledge firms use to organize production, caused by breakdowns in relationships between firms and their suppliers, for example. As some firms failed during the Depression, efficiency in surviving firms decreased; managers had to shift time away from production in order to establish new relationships, and firms had to shift to unfamiliar technologies that initially were operated inefficiently. This article originally appeared in the American Economic Review. © 2001 by the American Economic Association. The views expressed herein are those of the author and not necessarily those of the Federal Reserve Bank of Minneapolis or the Federal Reserve System. The Great Depression brought a striking short-run productivity change to the U.S. economy. Between 1929 and 1933 in the United States, real output per adult fell more than 30 percent, and total factor productivity (TFP)— changes in output not accounted for by changes in measured inputs—fell about 18 percent. This TFP decrease is much larger than expected from just extrapolating the TFP decrease that typically has occurred during postwar U.S. recessions. During the average postwar downturn (between 1947 and 1992), output has fallen about 2 percent and TFP, 0.3 percent. This relationship suggests that TFP should have fallen only about 4–5 percent during the Depression, rather than 18 percent. It is unlikely that this decrease is due to technological regress, which is the simplest interpretation of this productivity change. If that is not the cause, however, then what is? The Depression remains one of the most important and enduring mysteries in macroeconomics, and identifying the causes of this productivity decrease may shed new light on this mystery. Here I present productivity data from the Depression and assess how much of the TFP decrease can be explained by five commonly suggested factors: two types of errors in the measurement of inputs—changes in capacity utilization and in the quality of factor inputs—plus three other factors—changes in the composition of production, the hoarding of labor, and increasing returns to scale. I find that all of these factors combined explain less than one-third of the 18 percent decrease. I conclude by suggesting that decreases in organization capital (the knowledge firms use to organize production) may be a promising candidate for explaining the productivity decrease. But as yet that decrease remains a tantalizing puzzle. Factor Mismeasurement? My analysis uses John Kendrick’s (1961) TFP measure, which is the ratio of real gross national product (GNP) to an index of total factor input. This input measure is a factor share–weighted average of aggregate capital input and labor input. Table 1 shows Kendrick’s 1930–33 values for the TFP measure, output, capital, and labor relative to their values in 1929. According to this measure, TFP fell throughout the Depression and was about 18 percent below its 1929 level in 1933. I begin my analysis by estimating how much of the productivity decrease is due to factor mismeasurement. Microeconomic studies indicate there were changes in capital utilization and in the average quality of capital and labor input during the Depression. Capital utilization fell, and the average quality of employed capital and labor rose as the least productive inputs were idled. These changes are not all captured in Kendrick’s TFP measure, so I adjust his input measures to take account of them. Adjusting capital input requires estimating how much of the capital stock (measured in efficiency units) was idle during the period. Since there is no standard aggregate measure of idled capital, I estimate it using manufacturing data from the work of Timothy Bresnahan and Daniel Raff (1991). They report that the number of active manufacturing plants fell one-third between 1929 and 1933. There are at least three reasons, however, that one-third is too large an estimate of the fraction of the aggregate capital stock idled. First, the manufacturing sector contracted more than average in the period 1929–33, which suggests that a greater fraction of its capital was idled than the capital in other sectors. Second, the idled plants tended to be much smaller than the plants that remained active (Bresnahan and Raff 1991). Third, the idled plants tended to be the least productive plants (Bresnahan and Raff 1991). This indicates that the idled plants (measured in efficiency units) were much smaller than the operating plants. While a detailed analysis of idled capital is beyond my scope here, these three facts suggest that the fraction of the aggregate capital stock idled is much less than one-third. For this study, I assume that the fraction idled is 20 percent. I next examine changes in the average quality of labor input during the Depression. I focus on two types of quality changes: intersectoral changes and intrasectoral changes. Intersectoral quality changes arise from shifts in the composition of production across sectors. These shifts change average labor quality because labor quality varies by sector. For example, agricultural workers at the time of the Great Depression were less skilled, on average, than manufacturing workers. Kendrick’s labor measure adjusts for this source of quality change by multiplying sectoral hours by the sectoral wage. Intrasectoral quality changes arise from changes in the average quality of workers within sectors. Kendrick’s labor measure does not adjust for this type of quality change. But we can get a rough idea of its size from other studies. Stanley Lebergott (1993) reports that employee quality rose during the Depression; employment loss was concentrated among low-wage workers, and the most productive workers worked the longest shifts. This suggests that the average quality of individuals who continued to work during the Depression was higher than the average quality of individuals working before the Depression. Harold Cole and I (2001) use macroeconomic data to estimate how much measured wages were biased upward by layoffs of low-wage workers during the Depression. That estimate suggests that the quality of workers may have increased as much as 15–18 percent during this period (Cole and Ohanian 2001, p. 204). Lebergott (1993) also reports microeconomic data suggesting that the average quality of workers at the two largest firms in the electrical equipment industry (General Electric and Westinghouse) rose about 10 percent during just the first two years of the Depression. Given these estimates, I assume that average worker quality rose 7 percent during the Depression. This is a more conservative adjustment than either of the two preceding estimates and thus will produce a relatively small revision to Kendrick’s TFP measure. I recompute aggregate TFP with these capital and labor adjustments. I find that these adjustments explain only about two percentage points of the 18 percent TFP decrease. This is because the change in labor input, multiplied by labor’s share, offsets much of the change in capital input, multiplied by capital’s relatively small share. Production Shifts? Since these factor mismeasurements do not explain much of the decrease in aggregate TFP, I now examine sectoral data to see if less-aggregated productivity measures also fell during the Depression. The first column of numbers in Table 2 shows TFP values in 1933 relative to TFP values in 1929, for the five sectors Kendrick reports. These five sectors account for about half of 1929 GNP. The data show that these sectoral productivities fell during the Depression much less than aggregate productivity did. Manufacturing and railroads are the only sectors that show substantial TFP declines, and these declines are only about half as large as the decline in aggregate TFP. The fact that aggregate productivity fell more than these sectoral productivities raises the possibility that shifts in the composition of production from sectors with a high value of marginal product to sectors with a low value of marginal product contributed to the aggregate TFP decrease. Labor and relative wage data are also consistent with this view. The second column of numbers in Table 2 shows the level of sectoral hours worked in 1933 relative to its level in 1929, while the third column shows the 1929 average wage in the sector relative to the 1929 average wage in all sectors. These labor and wage data show that the agricultural sector, which pays relatively low wages, had only small declines during the Depression, while the manufacturing and mining sectors, which pay relatively high wages, both had substantial declines. How much did these shifts in the composition of output decrease aggregate TFP? Kendrick tries to correct his aggregate TFP measure for the effect of compositional shifts by multiplying sectoral inputs by sectoral factor prices. He estimates that compositional shifts reduced aggregate TFP by about 2.5 percent. Without the compositional correction, Kendrick’s aggregate TFP measure would have decreased 20.5 percent rather than 18 percent. Kendrick’s 2.5 percent adjustment seems small, however, relative to the large expansion of the low-value agricultural sector. As a robustness check, I independently estimate the size of the compositional effect. I begin by constructing a model to understand the connection between sectoral productivities and aggregate TFP. The model specifies that sectoral outputs Yi are produced from constant returns to scale production functions using capital Ki and labor Li that differ only by their TFP level. The TFP is denoted by Ait: (1) Yit = AitF(Kit,Lit). Aggregate output is the sum of sectoral outputs multiplied by base-year sectoral prices, which are denoted as pi: