My recent column on public sector unions caused something of a kerfuffle among my loyal critics. Appropriately, they note that comparing average compensation between groups does not establish impropriety; that is, differences between average male/female salaries does not mean those differentials are unjustified.
In the public sector unions column, I noted there are two types of factors that must be considered when comparing compensation differentials: individual worker characteristics and job characteristics. If you have adequate controls for both differences in people and jobs and you still have a compensation differential; then, and only then, do you have evidence that the compensation differentials are unjustified.
Brian Cooney brought the public/private compensation study by Bender and Haywood, two University of Wisconsin-Milwaukee economists, to my attention. As he indicates, they conclude state and local workers are paid less than private sector workers in terms of both hourly wages and total compensation.
As is always the case, however, the devil is in the details. The authors note that the average public employee is twice as likely to have at least a college degree compared to the average private sector employee. When Bender and Haywood conduct their analysis of hourly wages, they control only for worker characteristics; there are no controls for job characteristics. This means their model has a serious case of “omitted variable bias.”
Why is this important? Bender and Haywood note that “The most common occupations in state and local sectors include teachers, social workers, nurses, and university professors” (p 7).
Teachers and university professors have life-time job security through tenure, and all public employees are protected by complex civil service regulations. Public employees are not employed at will; hence, their employment risk is negligible. Private sector workers are subject to layoffs. Private sector workers bear a higher employment risk and should be compensated for that risk. Not controlling for these differences in employment risk biases Bender and Haywood’s results. This is what is meant by “omitted variable bias.”
When researchers fail to include all of the factors that should be in the model, they can be accused of selecting the variables in order to get the result they desire. I have no idea if this is what motivates Bender and Haywood. I only know that critical variables are missing from their analysis, and this has very precise statistical consequence for their analysis, and these consequences are all bad.
Further, consider Bender and Haywood’s hourly wage variable. They state: “The earnings data identify usual weekly earnings, which we convert to hourly wages by dividing by usual weekly hours” (p 17).
Using this as the hourly wage variable across public and private workers ignores the fact that teachers and college professors work only nine months out of the year, while private sector workers must work 12 months each year. It also ignores the differences in the number of paid holidays. This is most certainly not “apples to apples.”
The proper way to calculate the average hourly wage variable is to multiply the computation of the average hourly wage by the number of weeks in a year and then divide by the number of weeks each year actually worked. This yields a much higher average hourly wage for teachers and college professors, and it puts the hourly wage rates on an equal footing.
Next, consider the other half of total compensation: benefits. Bender and Haywood measure benefits by employer contributions to insurance, retirement and health care.
Since public sector workers typically have defined benefits coverage and private sector workers have defined contributions coverage, the measurement of public sector benefits compensation is significantly understated. Defined benefits programs lead to serious unfunded benefits problems, and this is a large part of the state and local financial crisis. State and local governments contribute too little to these funds. If you add in the unfunded pension liabilities, the real cost of public sector benefit packages are much higher.
Controlling for different employment risk and modifying the measurement of hourly wages and compensation all tend to make the disparity in average compensation between public sector and private sector workers worse. That is, they tend to make total compensation going to public sector workers much higher than private sector workers.
Bob Martin is emeritus Boles Professor of Economics at Centre College.