Possible measurement bias in aggregate productivity growth

Monthly Labor Review, Feb, 1999 by William Gullickson, Michael J. Harper

By examining industry multifactor productivity in more detail, researchers can gain new insights into the hypothesis of measurement bias in aggregate output and productivity

Output per hour in the business sector has grown about 1 percent per year since the late 1970s, according to data published by the Bureau of Labor Statistics. Some scholars in the productivity research field have suggested, however, that productivity might have grown faster.

One line of reasoning that supports this faster growth theory hinges on the decomposition of productivity trends by industry. When business sector output and hours are allocated to manufacturing and nonmanufacturing, the nonmanufacturing trend in output per hour appears to be very low. When output and hours are further allocated by industry, some of the resulting productivity trends appear to be negative. These trends are difficult to reconcile with anecdotal evidence of productivity improvements.

This article sheds some new light on these issues by using measures of multifactor productivity. The multifactor productivity framework is well suited to sorting out many of the issues because it allows us to account for capital inputs and for intermediate flows between industries. With these measures, we can compare industry and sectoral productivity trends.

The multifactor productivity measures that we present in this article are derived from various published and unpublished government data sources. Using these measures, we are able to conduct two main data exercises--one which examines aggregate manufacturing and nonmanufacturing multifactor productivity and another which examines nonmanufacturing multifactor productivity at the level of two-digit sic industries.(1) Many of the measures that we present are unpublished, and we do not consider them to be prototypical BLS measures. The point of our data exercises is to examine possible problems with the data.

To estimate multifactor productivity for two-digit nonmanufacturing industries, we used input-output tables and other published and unpublished data. (See the appendix, which explains how we assembled the data.) In an earlier study, we describe how an "ideal" set of data, comprised of input-output tables and price deflators, might be used to construct a set of multifactor productivity measures which were in turn consistent with the economic theory of firms.(2) In this article, we emphasize that available data actually fall short of the "ideal" in a number of respects. Nonetheless, the data come close enough in concept to the "ideal" to make the industry multifactor productivity framework a useful tool for analyzing aggregate multifactor productivity data.

An advantage of this approach is that it allows us to rule out certain sources of productivity bias. Specifically, those biases resulting from an incomplete definition of productivity and those biases resulting from an improper allocation of productivity to industries can be evaluated separately from other sources of bias. A further advantage is that multifactor productivity comes closer than output per hour to reflecting the phenomena which people usually have in mind when they think of productivity such as "technological change," "efficiency gains," "increased returns to scale" and "quality change." If there is anecdotal evidence that these phenomena have been operating with a positive influence, then we might expect multifactor productivity to have a positive trend.

Aggregate productivity measures

Manufacturing and business: divergence in productivity. From 1960 to 1973, productivity grew around 3 percent per year in each of the aggregate sectors. Since 1973, however, trends have been lower, causing the United States to experience a major "productivity slowdown." (See table 1.) Nevertheless, post-1973 trends also show a major divergence between the trends for manufacturing and those for business and nonfarm business. Manufacturing was not affected as much by the slowdown between 1973 and 1979. By 1979-98, the divergence intensified: manufacturing productivity nearly returned to its pre-1973 growth rate, while the trends languished in the other sectors. A closer look into this period shows that while all productivity trends increased from 1994 to 1998, the divergence also intensified.

Table 1. Trends in output per hour for major sectors, compounded annual rates of change, selected periods, 1949-98

[In percent]
                            Nonfarm
    Year        Business    business    Manufacturing
1949-98           2.3         2.0            2.7
  1949-60         3.3         2.6            2.0
  1960-73         3.3         3.0            3.0
  1973-79         1.3         1.1            2.1
  1979-98         1.3         1.1            3.1
    1979-90       1.2         1.0            2.6
    1990-94       1.2         1.1            3.2
    1994-98       1.7         1.6            4.3

Is something wrong with this picture? Noting the output per hour divergence between manufacturing and business, some users of BLS productivity data have pointed out that nonmanufacturing productivity growth must be quite low. Furthermore, it has been suggested that this result stands in contrast to abundant anecdotal evidence of remarkable changes in many nonmanufacturing industries. BLS has never published productivity measures for the aggregate nonmanufacturing sector because of concerns about such measures. However, as some users have pointed out, the aggregate productivity measures BLS does publish could be biased. Some have gone even further in speculating as to the implications of these low productivity trends for other government data from which they are derived. Larry Slifman and Carol Corrado, staff members of the Board of Governors of the Federal Reserve Board system, suggest that the low trends could be a manifestation of the Consumer Price Index (CPI bias alleged by the Boskin Commission.(3)