INTER-INDUSTRY WAGE DIFFERENTIALS: Empirical Evidence 9

Industry rents are particularly difficult to measure. The NBER data provides a measure of value added. Theoretically this variable represents employee wages, other employee benefits such as social security which are not directly included in the wage bill14, ex-ante capital rental costs, industry rents if any, and finally firm-specific rents which accrue to the owners of the capital,
w6626-28
The coefficient a represents the per worker cost of non-wage benefits plus average rents, /3 represents the capital rental costs, and 8 is the rent residual. Since it is impossible to separate from the constant that part which represents average rents, we use only the estimated rent residuals smoothed over seven periods to form estimates of sectoral long run rents.

Table 3.8 reports a number of basic statistics for the additional data. The average education level for production workers was slightly less than 12 years in 1993, up from only 10 years in 1960. There are distinct differences in the educational attainment of workers across sectors. In 1990 the average years of schooling varied

Table 3.8: Supplemental Statistics

Additional explanatory variables for wage and hour regressions computed per production worker.
w6626-29
across sectors from 9 years to 13.5 years. Around 20% of production workers were actively enrolled in unions or covered by a union contract in 1993, a significant decrease from the 42% enrollment in 1960. Again there appears to be a fairly large variation in union participation across sectors, with a maximum of 60% to a minimum of 2% in 1993. In 1960 this range stretched between 12% and 82%. The bottom of table 3.8 provides basic statistics on the computed industry rents. Note that these data have been converted to real values by dividing by the PPI deflator. The average rent is close to zero in each period, a function of the regression technique employed to construct this measure. Interestingly there has been a sharp increase in the variance of rents since 1980.

Except for the percentages, unionization rate and female rate, all these variables will be entered into our equation in logarithmic form. Table 3.9 reports the correlations between different variables within a data set and also between the same variables from the CPS data and the NBER data. The correlation between average weekly hours and average weekly wages across the two data sets are respectively 0.55 and 0.75 implying that the patterns seen in the NBER data are in part replicated in the CPS data despite the large differences in the collection techniques employed by the two sources.

Of the other variables, education appears to be significantly correlated with both wages and hours. Not surprisingly educated workers are also over-represented in capital-intensive sectors. Unionization is highly correlated with wages and positively correlated, albeit weakly, with weekly hours. This surprising result may be linked to the high positive correlation between union activity and the capital intensity of the industry. Rents are also positively correlated with capital intensity. These correlations make it possible that what we are seeing in the initial set of wage and hour regressions is only omitted variable bias not a wage-effort offer curve. These new variables will now be included in another set of regressions.