A Retrospective on the Postwar Productivity Slowdown

A Retrospective on the Postwar Productivity Slowdown

William Nordhaus

Including an Appendix with Alexandra Miltner

Yale University

September 27, 2004

Abstract

The present study reviews the well-documented “productivity slowdown” of the 1970s and 1980s. The study also develops a new data set – industrial data available back to 1948 – as well as a new set of tools for decomposing changes in productivity growth. The major result of this study is that the productivity slowdown of the 1970s has survived three decades of scrutiny, conceptual refinements, and data revisions. The slowdown was primarily centered in those sectors that were most energy-intensive, were hardest hit by the energy shocks of the 1970s, and therefore had large output declines. In a sense, the energy shocks were the earthquake, and the industries with the largest slowdown were near the epicenter of the tectonic shifts in the economy.

The sources of the productivity slowdown in the 1970s have been a long-running thriller in macroeconomics and policy studies. It is widely accepted that the rapid productivity growth of the postwar period ended sometime around 1973. It was followed by a dismal productivity record and accompanying stagnation in real incomes until the mid-1990s, at which point productivity rebounded sharply.

The present paper returns to that episode in the spirit of economic archaeology. The first section puts the 1970s slowdown in historical context by asking whether the slowdown is “unusual” by historical standards. The second section introduces a new data set on industrial value-added data and then uses that new data set to help understand the sources of the slowdown by industry and by composition shift.

I. Was the Productivity Slowdown of the 1970s an Unusual Event?

The first question addressed is whether the productivity slowdown of the 1970s was historically unusual. In other words, were the magnitude and duration of the slowdown something that was frequently seen in the historical record?

In answering this question, I turn to long-term data on productivity growth for the United States. The most appropriate data for such comparisons are data on productivity per hour in the non-farm sector, which are available back to 1889. There are some serious concerns about the historical comparability of these data, which are spliced together from a variety of sources. The major concern is that the data before 1947 are based on fixed-year-weighted output data, while that since 1947 uses the new chain-weighted indexes. Additionally, early data also have different source data and use different price indexes (or input indexes) to measure real output in several sectors.

A second issue concerns the appropriate technique for comparing contemporaneous and past productivity slowdowns. The approach used here is to consider periods of slowdown of different lengths – from 5 to 20 years – and to determine how many slowdowns of that magnitude occurred during the period 1889-2004. More precisely, I construct dummy variables that were 0 outside of the slowdown period and 1 in the slowdown period. I constructed a set of rolling dummy variables beginning in 1889 for lengths of 5, 10, 15, and 20 years. Note that these tests attempt to measure the deviation of productivity growth from its 1889-2004 trend.

It is tempting to perform statistical significance tests on these series. However, it is clear that the underlying series have non-stationary variances, are not normally distributed, and we therefore cannot use standard tests because of the overlapping samples. The best approach is probably to examine the pattern of results.

Figure 1 shows the plots of the coefficients on the dummy variables in the rolling regressions. The productivity slowdown of the 1970s is not unusual, but it is also not unique at any frequency. The 1970s is, however, dramatically different from periods since World War II – a period when the data are surely more reliable. We can also use these graphs to ask whether the productivity resurgence after 1995 was particularly unusual. For most of the time windows, the latest observations barely make it back to the trend line. Hence, the latest productivity figures would be characterized as return toward normal rather than unusually high.

Table 1 shows a compilation of ranks of the 1970s slowdown. By any standards, the 1970s slowdown was unusual but not unique. It ranks between fourth and sixth of slowdowns over the 1889-2004 period. For the longer period windows, the productivity slowdown that began around 1900 was slightly larger than the one that began in the 1970s. On the other hand, there was nothing remotely as large a slowdown in the period since World War II.

Another tricky issue concerns the dating of the productivity cycle. The following shows the dates of the maximum slowdown according to the length of the window:

Window length Period of Slowdown and Midpoint

5 year 1978-1982 (1980)

10-year 1973-1982 (1977-78)

15-year 1977-1992 (1984)

20-year 1977-1996 (1986-87)

The different windows provide slightly different answers as to the inception of the slowdown, ranging from 1973 to 1978, with the end point being from 1982 to 1996.

In summary, the productivity slowdown of the 1970s does appear to be a major distinguishing feature of the last century. At long frequencies (more than a decade), it is one of two major productivity slowdowns of the last century. At all frequencies, it is a major outlier for the period since World War II.

Figure 1. Coefficients on Rolling Regressions

This figure shows the coefficients on dummy variables of differing lengths in rolling regressions for the period 1889-2004. For example, the rolling regression of the growth in labor productivity on a constant and a 5-year dummy variable beginning in 1978 has a coefficient of 0.020, indicating an estimated deceleration of 2.0 percentage points. ______

Length of slowdown window / Rank of 1970s slowdown period and beginning year / Years with slowdowns larger than 1970s
5 years / 4 (1979) / 1910, 1929, 1928
10 years / 6 (1973) / 1908, 1924, 1925, 1905, 1906
15 years / 4 (1977) / 1900, 1903, 1901
20 years / 5 (1977) / 1896, 1898, 1893, 1995

Table 1. Performance of 1970s Productivity Slowdown

Table shows the rank of the 1970s productivity slowdown relative to other slowdown periods. The 1970s was “unusual” but not “unique” in terms of other slowdown periods of the 1889-2004 period.

______

II. A Close Look at the Productivity Slowdown

In this section, we focus an analytical and empirical microscope on the 1970s productivity slowdown. The occasion for this analysis is the development of a new set of detailed industrial accounts that have been extended back to 1948. These data, which are described in the Appendix, comprise an integrated set of accounts that aggregate to total GDP. The purpose of this exercise is to identify alternative ways of measuring productivity and to determining the detailed sources of the 1970s slowdown.

A. Analytical Measures of Productivity Growth

To begin with, we summarize alternative approaches to measuring productivity growth. The customary approach to measuring productivity growth is (a) the difference of growth rates approach. This defines productivity growth as the difference between the growth rate of output and the growth rate of inputs. In a companion paper, I showed that this is not an appropriate welfare-theoretic measure of productivity growth. I proposed two alternative measures: (b) a welfare-theoretic measure, which is defined as the current-weighted average of productivity growth where the weights are the shares of nominal output and (c) a fixed-weight measure, which has the same basic construction as the welfare-theoretic measure except that it uses nominal output weights of a given year. The difference between (b) and (c) is a complex set of factors including a Baumol effect, that captures the impact of changing shares of nominal output.

We can summarize the earlier work in terms of a decomposition equation for the growth in productivity:

(1) g(At) = i g(Ai,t)i,0 + i g(A it) [i,t - i,0 ] + i g(S it)[i,t - wit]

where g(At) is the growth of output per unit input, g(Ai,t) is the growth of output per unit input in industry i, i,0 is the share of industry output in total nominal output in the base period, [i,t - i,0 ] is the difference between the current share and the base-period share of nominal output of industry i, and (i,t - wit) is the difference between the share of nominal output and of inputs in industry i. The first term in equation (1) is the fixed-weight measure of productivity growth, the second term is the Baumol effect, and the third is a set of factors I have called the Denison effect which captures the effect of different levels of industry productivity on total productivity. The first two terms of equation (1) are the welfare-theoretic measure of productivity.

B. A New Data Set for Industrial Output, 1948-2001

At present, comprehensive industrial data for the United States are limited to recent years. Before the most recent revision, the U.S. Bureau of Economic Analysis (BEA) published detailed industrial data on quantity and price indexes for major industries for two subperiods, 1977-1987 and 1987-2001. The present author and Alexandra Miltner have developed an approach that uses earlier fixed-weight data by industry to produce a comparable set of data for the period starting in 1948. The derivation of the data are explained in the appendix.

This data set has the desirable property that it is consistent with the data on nominal output for the industries and aggregates to income-side gross domestic product. Availability of the industrial data allows us to extend analyses of productivity to include the period before the productivity slowdown of the 1970s and to develop a decomposition of the slowdown by industry. Unfortunately, because BEA has shifted from the earlier SIC industrial classification to the NAICS system, the detailed industrial data have been discontinued, the NAICS data go back only to 1998, and it is virtually impossible for private scholars to map the SIC into the NAICS system. Therefore, it is unlikely that a continuous series can be developed beyond 2001.

Figure 2 shows one comparison of the new data set. This figure shows annual productivity growth for the business sector using the new income-side data set and BEA’s hours data as compared with the standard productivity growth rate from the BLS. The overall pattern is similar, but cyclical movements differ. We will also see below that the trends are somewhat different.

Figure 2. Comparison of Productivity Growth for Business Sector using BLS and Income-Side Output Data

C. Measures of the 1970s Productivity Slowdown

Using the new data set and the techniques described in section A above, we can examine the productivity slowdown using different concepts. Tables 2 through 5 show the alternative measures for the overall economy, for the private business sectors, for the private nonfarm business sectors, and for what I have defined as “well-measured output,” or WMO.[1] The major results of the exercise are the following:

1. We first compare labor productivity measures under the current (income-side) aggregate as compared to the standard (product-side) measure used by the Bureau of Labor Statistics. The productivity growth estimates for the entire period of 1948-2001 are very similar between the two sources. For the BLS business sector, average productivity growth was 2.39 percent per year compared to 2.00 percent per year for the present construct. For the BLS nonfarm business sector, average productivity growth was 2.14 percent per year compared to 1.96 percent per year for the present construct. (We have been unable to reconcile the two measures, but it appears that the estimates on hours worked is the major source of the discrepancy.)

2. The productivity slowdown can be seen in each of the sectors (GDP, business, nonfarm, business, and well-measured) for each of the three measures (total productivity growth, welfare productivity growth, and fixed-weight productivity growth). For total productivity growth, the slowdown ranges from 1.01 percent per year for GDP to 1.26 percent per year for the non-farm business sector.

3. In general, the income-side measures of output show a slightly smaller productivity slowdown from 1959-73 to 1973-85 than do the BLS measures. The two BLS measures show a productivity growth slowdown averaging 1.60 percent per year, while the total productivity measure (which is the comparable concept) shows a slowdown of 1.24 percent per year.

4. The traditional measure of productivity growth (chained output per hour worked) shows a larger slowdown than does either of the other concepts. The welfare-theoretic measure shows a productivity slowdown of 1.06 percent per year for business output, which is 0.17 percent per year less than the number for total productivity growth. Similar differences are seen for nonfarm business and well-measured output.

5. Well-measured output shows a consistently higher rate of productivity growth than do the other concepts. Over the entire period from 1948 to 2001, for the total productivity concept, WMO has a welfare productivity growth of 2.54 percent per year as compared to 2.00 percent per year for private business. The difference is due to the poor measured productivity performance in many (poorly measured) service and finance sectors.

6. One of the major questions that can be addressed is whether productivity growth in the rebound period after 1995 has returned to the pace of productivity growth before 1973. According to our figures, the 1995-2001 productivity performance was on a par with that in the 1959-1973 period; by contrast, the BLS data shows a slight deterioration from 1959-73 to 1995-2001. Each of our measures for business, nonfarm business, and well-measured output shows more rapid productivity growth in the post-1995 period than in the 1959-73 period. A substantial part of the difference between the BLS and the income-side data is that the statistical discrepancy gives a boost of 0.25 percent per year to the growth of income-side GDP over the rebound period; subtracting that from the income-side, or adding that to the product side, erases a large fraction of the difference in productivity growth between the BLS and the income-side measures over the 1995-2001 period, although it erases only half of the difference in the measures of the rebound between the two sources.