Estimates of Productivity Change by Industry

An Evaluation and an Alternative

Edward F. Denison
Release Date: March 10, 1989

Government figures show that from 1948 to 1973 productivity increased at the same rate in the manufacturing and nonmanufacturing portions of American business. The same government series indicate a dramatic shift in the 1980s, when productivity increased far more in manufacturing industries than in nonmanufacturing. Denison’s analysis challenges the reality of this reported change.

Denison focuses on the meaning and reliability of productivity series for industries within the business economy. His analysis divides into three parts.

First, he finds that a large part of the difference between manufacturing and nonmanufacturing growth rates of output per hour stems from the computer industry. He argues that the effect of recent rapid productivity growth in computer production is greatly overstated because the weighting system used to combine products in output measurement exaggerates the importance of computers. Use of an output measure that does not deduct depreciation has the same effect. Denison also questions the way that output of computers is measured.

Denison next examines the way changes in output per hour are allocated among industries. His evaluation leads to two major conclusions. One is that the information underlying the estimates is insufficient to warrant confidence in indicated differences among industries. The other is that, quite apart from the computer, the estimated increase in output per hour in the U.S. manufacturing in recent years is almost surely overstated and the increase in nonmanufacturing is correspondingly understated. The size of bias has, in all likelihood, increased over time.

Denison recommends that the government introduce an alternative way of dividing the economy for productivity measurement, one that measures productivity in the production of different types of final goods instead of by industry. He describes a methodology that could provide such estimates. The results, Denison contends, are more enlightening and the problems encountered in estimating productivity by industry can be avoided.