Comments and Discussion

Daniel E. Sichel:¹

Ian Dew-Becker and Robert Gordon have put together a very nice paper that covers some important ground. Although I will highlight some quibbles and questions, my overall view is that the paper is quite interesting and extremely well crafted. The issue the authors examine could be summarized as "Who got the benefits from the increases in labor productivity, in the past decade and over the longer haul?" To address this question, the paper covers four broad issues. First, it reviews the key measurement issues that must be understood before proceeding to a comparison of trends in labor productivity and real compensation per hour. Second, the paper tills some new ground on wage-price dynamics and presents a new inflation equation, an updated version of Gordon's "Goldilocks" equation from several years ago. Third, the paper uses the estimated wage-price dynamics to gauge the effect on the labor share of income when trend productivity growth changes. Finally, the paper turns to a micro analysis of changes in the income distribution, using Internal Revenue Service tax data from 1966 to 2001; importantly, the paper links these data to a measurement framework consistent with the National Income and Product Accounts, allowing the authors to examine which income deciles received gains in real hourly compensation that equaled or exceeded the rate of increase in labor productivity growth, and which received less. I will focus on each of these broad topics in turn.

The first point made in the paper's measurement section is that comparisons of productivity growth and gains in real compensation per hour must be made using comparable data. Some analysts compare productivity in the nonfarm business (NFB) sector with average hourly earnings deflated by the consumer price index and find that productivity has risen significantly faster than real wages. As Dew-Becker and Gordon point out, however, such a comparison can be quite misleading, because the consumer price index used to deflate average hourly earnings differs from the deflator used for NFB productivity, and because the average hourly earnings measure covers the earnings of production and nonsupervisory workers only, not all workers in the sector. Dew-Becker and Gordon steer readers to a more appropriate earnings measure against which to measure changes in labor productivity, namely, a comparably deflated measure of compensation per hour from the Bureau of Labor Statistics' Productivity and Cost (P&C) release. This measure does cover all workers in the sector, and by this measure the difference between the growth rate of labor productivity and that of real compensation per hour over the past several decades is rather small.

Although the authors get to the right numbers for comparing labor productivity and real hourly compensation, their explanation of why P&C compensation per hour is more appropriate for this comparison than average hourly earnings is incomplete. They correctly point out the difference in worker coverage, but they do not mention another very important difference, which is that average hourly earnings excludes benefits whereas P&C compensation per hour includes the value of benefits. The value of output used in the labor productivity figure includes benefits, and it is therefore important to use a compensation measure that also includes benefits, especially given that benefits make up nearly 30 percent of compensation and have risen considerably faster than wages and salaries in recent decades.²

The second point emphasized in the paper's measurement section is the well-known fact that the labor share of income has exhibited no discernible trend over the past twenty years or, arguably, the past fifty. Of course, this share could have shifted over these periods if and only if aggregate labor productivity and aggregate real hourly compensation had increased at different rates; thus the stylized fact about constant labor shares implies—and is implied by—the similarity in growth rates of labor productivity and real hourly compensation noted above. I have no quibble with any of this analysis, which is based on the conventional NIPA data. However, this widely accepted result is dependent on the use of those particular data. If one...