TRENDS IN OCCUPATIONAL WAGE DISTRIBUTION 47 of the estimated coefficients. The wage distribution between low skilled and high

skilled workers is represented by the scaled coefficients and can be interpreted in relation to those of medium skilled workers. The differences in payment resulting through different skill levels reflect the degree of inequality in the wage structure.

Results are presented in Table 3.8. All estimations are significant at the 1% level.

The results are robust to the inclusion of several control variables, for example, unemployment by skill level, GPD growth, or exchange rates. The first column gives the estimations results, the second column contains the computed wage dispersion between the different skill levels, respectively.

Results are quite similar for the OECD and the EU, as well as for the United States and Germany. Low skilled workers earn on average about 15 percent less than medium skilled workers, whereas high skilled workers earn about 60 percent more than medium skilled workers. As Germany is a member of the EU, and both, the United States and Germany, are members of the OECD, I test whether the results for the OECD and EU are driven by these countries. Excluding Germany slightly changes the results for the EU: While the scaled coefficient for low skilled workers stays almost unchanged (0.849), the scaled coefficient for high skilled workers decreases (1.533). Thus, wages differences between medium and high skilled workers are driven by Germany. Running the approach for the OECD without Germany does only change the scaled coefficient of high skilled workers (1.510). Excluding both, the United States and Germany from the OECD countries leads to the same result. The scaled coefficient for low skilled workers does not change (0.839), while the scaled coefficient for high skilled workers decreases (1.492). It can therefore be concluded that Germany drives wage inequality with regard to the relation of medium to high skilled workers. The same applies to the OECD for the United States and Germany.

The results are consistent with the assumptions of the theoretical model pre-sented in Section 2. There is a large wage heterogeneity between the different skill levels, which can be explained by different returns to the bundle of skills that is required to carry out an occupation.

However, several authors argue that it is not only the skill level that affects wage distribution and wage inequality, but the task content of jobs (for example Autor et

deviation of the log wage for each task group as a measure of wage distribution (see Table 3.5 for a presentation of the task groups). The results for the member states of the OECD and the EU, as well as the United States and Germany are presented in Figure 3.4. Again, the evolution of the standard deviations of log wages is quite stable in the OECD and the EU. It is remarkable that Germany is characterized by a almost constant wage distribution, while there is a large wage heterogeneity in the United States. The results for the OECD are robust to the exclusion of Germany and the United States. This is also valid for the exclusion of Germany from the sample of the EU.

Using task classification instead of the skill level allows a more nuanced view of wage the distribution. The development of the standard deviation of log wages supports the assumption of the theoretical model that wages are determined by the bundle of skill that is required to carry out an occupation, as there is hardly any change in the wage distribution within the task groups. With the exception of the United States, wage distribution within the five task groups is more or less constant.

I use a fixed effects approach to determine the effect of the different task clas-sifications on the log wage distribution (in US-Dollar). The estimation equation is quite similar to the one presented in equation (3.3), as I change only the explana-tory variable ”skill level” into ”task classification” of occupation o. I choose the log wage level of workers who performNonroutine Analytic as a benchmark. Again, the benchmark is scaled to 1.0. The wage distribution between the five different task classifications is represented by the scaled coefficients, which can be interpreted in relation to the wage level of the Nonroutine Analytic task group.

Again, the scaled coefficients are computed by using exponential function of the estimated coefficients. The differences in payment resulting through different task requirements reflect the wage distribution within each country or country group.

Results are presented in Table 3.9. All estimations are significant at the 1% level.

The first column gives the estimation results, the second column contains the com-puted wage dispersion between the different task groups, respectively. The results are also robust to the inclusion of several control variables, for example,

unemploy-3.4. TRENDS IN OCCUPATIONAL WAGE DISTRIBUTION 49 ment by skill level, exchange rates, or GDP growth. Determining the wage spread within task groups allows to draw a more complex picture about wage distributions within countries or groups of countries. The spread between workers occupied in jobs with different task requirements is considerable larger than the spread between different skill levels. For example, workers in jobs that require routine manual tasks earn on average about 50% of the wage of workers in jobs with nonroutine analytic tasks requirements. The largest spreads can be found in the United States. The results for the EU and the OECD are robust to the exclusion of Germany, as the scaled coefficients change for less than one percent. Excluding also the United States from the OECD-sample leads only to a small change in the result for Nonroutine Interactive tasks.

The results are consistent with the assumptions of the theoretical model. Wages differ with respect to the tasks required to carry out an occupation. There is a large heterogeneity between the different task groups, which can be explained by different returns to the bundle of skills that is required to carry out an occupation.

3.4.3 Occupational Wage Spreads across Industries

The theoretical model presented in section 2 allows to explain, why wages differ across occupations, even if workers are skilled equal. The descriptive analysis of occupational wage spreads as given in the previous section shows that wages in the same occupation differ across countries. I focus on the question, whether there are differences in wages in the same occupation across different industries.

Therefore, I focus on occupations, which are reported for several industries, and analyze whether there is a wage gap in the same occupation between industries in the OECD, the EU, the United States, and Germany. There are two occupations in the dataset, which are reported for several industries: Stenographer-Typist and Laborer. The stenographer-typist is reported for five different industries, the laborer for eight industries. I analyze the relationship between the wage level and the indus-try by regressing occupation-indusindus-try-dummies on the log dollar wage, controlling for country- and year-effects. The results of the regression are labeled as coefficients.

in the OECD as benchmark: Banks is benchmark industry forStenographer-Typist, and Electric light and power for Laborer. The wage level of both benchmark in-dustries is scaled as 1.0, the results for the other inin-dustries can be interpreted in relation to the benchmark and are hereinafter referred to as scaled coefficients. All estimations are significant at the 1%-level. The results are robust to the inclusion of several other control variables.

The first column of Table 3.10 gives the regression results, the second column gives the exponential function of the coefficient and shows the scaled coefficient of the wage level of each industry compared to the benchmark industry. I find significant differences in the wage level for the different industries. Thus, a stenograph-typist in Wholesale trade is on average paid worst in the OECD, the EU, and Germany.

He or she earns at least 15% less than a stenograph-typist who works in a bank. In contrast, a stenograph-typist in the United States earns most in the sectorWholesale trade. The differences in payment for a laborer are not as large as for a typist, but, there are also significant differences. The industry with the lowest wage for a laborer is Spinning, weaving and finishing textile in the OECD, the EU, and the United States. In Germany, a laborer in Iron and Steel Basic Industries earns the least.

The results show, that there is wage inequality within the same occupation across industries. Again, there are two variables in the theoretical model given in equation (3.1) which may explain this variation in wages in the same occupation. First, it is possible that the base payment θ_ot a worker earns in occupation o varies between industries. Therefore, the variable θ_ot should be rewritten with an industry indexi:

θ_oit. Second, even if the same bundle of skills is needed to pursue a particular occu-pation, the returns to skill r_kt may differ between industries and lead to differences in wages.