Decomposing technical change: a reply.

• Author: Gort, Michael
• Position: Response to article by Frank Song in this issue, pp. 209

The comment on our paper--Gort, Bahk, and Wall [1], henceforth GBW--raises two issues. First, are real wages an appropriate proxy for human capital? Second, is our unconventional approach of introducing labor quality as a separate argument in the production function justified? We discuss each of these issues in sequence.

We assumed that plants generally face a common labor market. Consequently, variations in average wages at a point in time mainly measure differences in skills rather than differences in the prices of identical classes of labor. Empirically, our assumption was supported by three observations: (1) variations in average wages for plants in the same industry are far larger than those that could be attributed to unionization or random influences; (2) we observed no consistent interregional variation in average wages; and (3) within the same industries, variations in average wages were larger for new than for old plants. The third observation is especially decisive since new plants have considerable choice in location and can, therefore, take advantage of local or regional differences in wages.

In contrast, the author of the comment appears to believe that each plant has its own labor market conditions with a unique equilibrium wage. The author cites differences in working environment and length of commute from home to work as examples of sources of variation in wages. To be sure, no two plants are identical in every respect as employers. A friendly manager may persuade some employees to accept a lower wage than a manager with a gruff manner. However, when the range of variation in average wages for the same industry is such that the highest average wage is three times greater than the lowest (as we often observed), we must ask a simple question. Is this plausibly the result of jovial smiles and low bus fares for some plants or does it arise because the plants employ very different types of labor? Obviously, we believe the latter. Individual plant peculiarities such as the author of the comment mentions merely represent the usual noise associated with all large bodies of data.

In previous studies, the approach has been to convert labor inputs into common efficiency units by multiplying physical inputs (hours worked or number of employees) by an index based on such factors as years of education (and possibly some demographic attributes). The resulting measure of labor input renders it difficult to separate out the effects on the...