Changes in federal and state regulations have increased the need for proactive pay equity analyses. The Equal Employment Opportunity Commission (EEOC) plans to begin collecting employers’ pay data in March of 2018, and several states have passed more aggressive equal pay or fair pay statutes. Thus, many employers are considering evaluating (or re-evaluating) their workforce to determine whether there are unexplained gender- or race-based pay disparities and to assess potential liability should a compensation claim be filed.
There are several benefits to conducting an internal pay audit prior to an Office of Federal Contract Compliance Programs (OFCCP) or EEOC investigation or to the filing of a lawsuit. First, the audit allows assessment of potential liability under the protection of the attorney/client work product privilege. Second, an audit reduces the cost of later responding to government investigators, opposing counsel, the courts, or an opposing expert, should such responses be necessary. Third, if a firm does a proactive analysis, it will have sufficient time to develop meaningful statistical models and to research peculiar-looking pay values. Thus, it can refine the data or the model or make any pay adjustments that the analytical results and subsequent research suggest are warranted.
A pay audit must start with a clear picture of the comparisons to be made. Will the analysis examine base pay only or variable compensation too? Is it only interested in examining gender pay differences or race/ethnicity base pay differences? At what level should the analysis be conducted? What data are readily available to conduct the analysis? What other data might be necessary to make the pay comparisons? Once the necessary comparisons have been delineated, it will be possible to develop a statistical model to make those comparisons using regression, a common statistical technique.
The first step in building a statistical model for use in the audit is to group “similarly situated” employees by considering what employees do. Under federal statutes, job title is often used to form these groupings. Job titles, however, may provide categories that are either too narrow or too broad. Recently passed state fair/equal pay regulations use different terminology for the grouping of like workers – “substantially similarly situated,” “equal work,” “comparable work,” etc. That terminology suggests that a broader “job family” (possibly combining job titles) might be relevant, although that will not be clear until the state courts begin to interpret this new regulatory language. If broader job families are used, then data on education and licenses or certifications might become more important methods of measuring differences in employees that legitimately lead to differences in pay. Unfortunately, most companies do not track education and training nor do they update changes after hire. Even those firms that capture education data often do not capture anything more than the highest degree received.
The second consideration in the model is how well employees do their job. Performance is often difficult to quantify, but it may be measured by proxies, such as time-in-job or time-in-grade, company tenure, and age-at-hire. Age-at-hire is a poor proxy for prior experience. Actual years of relevant prior experience is a better measure, but very few companies gather this information, even though it is often considered when setting starting pay. Data from performance evaluations may also be used to measure performance. At a minimum, any analysis should examine the model’s results both with and without performance measures.
A number of additional variables may be worth investigating for possible inclusion in the model. For example, an identifier for the employee’s organization within the company, such as the division or department, may capture the decision-making structure as well as differences in payroll budget. Similarly the employee’s location may capture geographic differences in pay. Other variables of interest may include whether an employee joined via acquisition (and when the acquisition occurred), and whether the employee possesses special skills that necessitate a premium. (sometimes called “hot skills”).Variables may also be included to show that special circumstances have caused an employee to be “red-circled” (paid above grade maximum) or “green-circled” (paid below grade minimum). The variables that should be included vary from company to company. Labor economists rely on discussions with counsel and human resources personnel to further refine the factors to be included in the statistical model to match the company’s pay practices as much as possible.
A key concern in constructing the model is the data. In many instances, the data to be used are determined by availability, and many companies do not maintain the data necessary to run a proper pay analysis. The data must be carefully reviewed for completeness and accuracy. Analysts need to ensure that values make sense. For example, one dataset indicated that almost all employees earned $60,000 to $120,000, but a few earned over $1 million. In fact, those seemingly very high salaries were measured in yen, not dollars.
After the data have been prepared, it will be possible to estimate the model to determine whether there are any potential problems, specifically statistically significant average pay differences between men and women or between members of different race/ethnicity groups. If any potential problems are found, the model will help identify employees in the analysis who are driving the pay difference.
By way of illustration, suppose that the pay difference between men and women is statistically significant and adverse to women among 500 employees in engineering roles. It is possible to isolate the subgroups within engineers having the biggest influence on the overall female/male pay difference. It may be that a large percentage (e.g. 25%) of the overall engineer pay difference is attributable to a small group of 15 petroleum engineers. Research could then focus on the 15 petroleum engineers rather than all 500 engineers.
The model may be used to identify employees having the biggest impact on the gender pay difference and those with the largest gap between expected pay (from the model) and actual pay. Such refined comparisons will enable the company to focus on a subset of employees, which will reduce the time cost of researching the pay differences. (It will be less costly to review 15 employee records rather than 500.) With a narrower focus, the company will be able to take appropriate action, either by refining the data or the statistical model to show the reasons for the pay differences or by adjusting the pay of selected employees where adjustments should be made.