Salary Regression Analysis
The Salary Regression Analysis is a Multiple Linear Regression analysis method of testing for possible compensation disparity within an organization. This analysis can take into account multiple factors that influence salaries (e.g. seniority, performance, skills etc.) and determine if there is a significant correlation between race or gender and compensation. It cannot be overemphasized that the salary regression analysis performed constitutes only the first step in what may be a necessary comprehensive study of your compensation practices. While providing useful insight to potential wage disparities, the results of this salary analysis should not be construed as conclusive evidence of either the existence of, or lack of an impermissible compensation practice.
The Salary Regression Analysis report tests if race or gender status significantly influences compensation after accounting for other independent variables included in the analysis, such as: number of years employed with the company, number of years in the current job, and performance ratings. Each organization may utilize up to six factors to help explain potential pay disparities. The analysis can be broken down by department, jobs, job groups, EEO categories, salary grades, SSEGs, or by all employees. We normally choose to complete the analysis by job or SSEG, because these most closely follow the Equal Pay Act and Title VII guidelines.
The Salary Regression Analysis report lists the job (or salary grade, department, job group, etc.) being tested, the total number of employees in the job and the average salary of the employees in the job. Note: Employees who report to offsite managers are not included in the Salary Regression Analysis report.
The "Avg Salary" column shows the average salaries for each of the groups that are analyzed. The"Significant" and "Std Dev" columns tell us whether the difference in average pay by race or gender is statistically significant. A standard deviation of -1.96 or less is considered statistically significant and "Yes" will be reported in the "Significant?" column.
The next three columns provide an indication of the strength or reliability of the analysis. The R² column indicates the percent of variance in compensation that is explained by the variables used in the analysis. For example, an R² of .112 for Females means only 11.2% of the difference in pay between females and males is explained by the included variables. Adding more variables, such as performance rating, experience, and date in job, may increase this percentage and help eliminate a finding of statistical significance. The Significance of F is a measure of how strongly correlated or predictive the variables are to compensation. A value greater than .05 indicates the analysis is not reliable and a different combination of variables should be used. The Significance of F Change measures the influence race or gender has on compensation after the other explanatory variables have been accounted for in the analysis. A value greater than .05 indicates that race or gender status does not significantly impact compensation.
The Salary Regression Analysis uses group size to determine if the analysis is statistically valid. The analysis is deemed statistically insignificant and "N/A" is reported in the "Significant?" column if the total group size is less than 30 or if any subgroup is less than 5. For example, if there were 31 males and 4 females in the group, the total group size of 35 is large enough, but the Female subgroup is too small to render valid results.
An explanatory variable that is multicollinear with another variable has been excluded from the analysis if indicated in the footnotes. Variables are considered multicollinear if they contain the same or close to the same values. For example, a company may have Original Hire Date and Most Recent Hire Date available as explanatory variables. If these two variables contain substantially the same dates in the particular group being analyzed, one of the variables should be excluded or the analysis is less reliable.
Only those variables that significantly correlate to salaries are included in the analysis if that option is checked. A variable is excluded from the analysis if it has a standard deviation that is not greater than 1.96 or less than -1.96. Unchecked, this option will force all variables to be included in the analysis. All variables that are included or excluded from the analysis are listed in the footnote section of the report.
The Race/Gender Coefficient column displays the average difference in compensation after accounting for all included independent variables. For example, a coefficient of -8709 for females indicates the average pay for females is $8709 less than that of the males in the group after adjusting for any explanatory variables included in the analysis. Adding more variables, such as performance rating, experience, and date in job, may reduce this difference.
See Also
Publish Your Reports
Report Options - Salary Regression
Understanding Your Reports
© Copyright Yocom & McKee, Inc.
