With the increase in class action lawsuits and government investigations involving alleged employment discrimination, many companies engage outside counsel and economists to proactively audit potential compensation differences between groups of employees. In such an audit, as in litigation, the most commonly used technique for assessing unexplained differences in salaries is a linear regression analysis (regression). Regressions allow an economist to identify average pay differences after accounting for various factors that determine pay.
Regressions estimate average relationships between pay and explanatory factors. For example, a regression might indicate that each additional year of tenure is associated with, on average, an additional $500 in salary. The regression does not speak to any single employee’s pay increases. Instead, it estimates the average return to tenure across the population holding constant other factors that affect salary. Regression differences can help identify areas where there are unexplained pay differences, and therefore potential pay equity issues. If a company chooses to adjust salaries based on the audit findings, however, caution should be exercised in moving from the aggregate results of the regression to the individual’s salary.
The regression can be used to predict the salary of an employee. This prediction can be interpreted as the expected salary of an employee given his or her specific characteristics. The difference between an employee’s actual and predicted salaries is referred to as the residual. A positive residual indicates that an employee’s actual salary is higher than predicted, whereas a negative residual indicates the opposite.
A common interpretation of negative residuals is that the employee is “underpaid,” perhaps because of discrimination. However, this interpretation is not necessarily correct, as the differences between actual salaries and predicted salaries may result, in part, from the omission of explanatory factors from the regression. Such omissions may be especially important when analyzing salaries because each salary is the result of decisions made by the employer and employee, possibly over many years, as well as skills and abilities that are difficult to measure. In fact, an examination of the differences between actual and predicted salaries by someone familiar with the employees may reveal valid explanations for the differences. For example, an employee with an actual salary much higher than his or her predicted salary may hold a certification or license not captured in the available data.
If a company chooses to adjust an employee’s salary, basing adjustments on predicted salary is a tempting option. Nevertheless, salary adjustments must be carefully reviewed to determine whether omitted factors explain the differences identified by the regression. This review can be time consuming, but the effort should result in adjustments that are better tailored to individual employees.