Adverse Impact and Test Validation Book Series: Multiple Regression Using Multiple Regression to Examine Compensation Practices Introduction Reasons for Investigating Pay Equity: The Equal Pay Act of 1963 Title VII of the Civil Rights Act of 1964 Employers may wish to examine Compensation between groups Tools available: o Independent Samples t-test (t-test) o Multiple Regression (MR) Comparison of Compensation using t-test: Pros: Simple to conduct Easy to understand Cons: Doesn t account for any legitimate reasons for differences in pay
Comparison of Compensation using MR: Cons: Requires much more data Requires deeper understanding of statistics/statistical analyses Pros: Includes explanatory variables Better reflects real-world compensation structure Endorsed by OFCCP Steps for Conducting MR Analyses Consider Purpose: Proactive or Reactive? Proactive: o Typically Employer-initiated o Data build is simpler Reactive: o Typically externally-initiated (Plaintiff or Government Agency) o Data build must be thorough Step 1 Identify and Review Available Data Establish a solid understanding of the data available and what will be used
Step 1 Identify and Review Available Data Standard variables to consider: Employee ID Job Grouping Information (e.g. Job Title, Job Group, other Similarly Situated Employee Group (SSEG) information) Race/Ethnicity Gender Date of Last Degree Earned Highest Degree Earned Date of Birth Time in Company Time in Current Position (Time in Job) Step 1 Identify and Review Available Data Standard variables to consider (con t): Current compensation (Annual Salary or Hourly Wage) Part-time vs. Full-time status Exempt vs. Non-exempt status Job Title Grade Level or Salary Band Employee Location Prior Experience data (many types) Performance Ratings Etc. Step 1 Identify and Review Available Data Select variables which best mirror on-the-ground compensation decisions Are variables Categorical or Continuous? o E.g. Education: Any Degree? vs. Highest Degree Choose Variable Formats appropriately Text Date Number Currency Etc.
Step 2 Create and Verify Variable Coding Focal Group: Females Total Minority Individual Minorities Reference Groups: Males Whites Step 2 Create and Verify Variable Coding Focal Group should be coded as 1 ; Reference Group should be coded as 0 e.g. Female 1, Male 0 e.g. Minority 1, White 0 e.g. African American 1, White 0 Step 3 Conduct Preliminary Data Analysis Generate Correlation Matrix In SPSS o Click: Analyze Correlate Bivariate o Identify variables In Excel o Data Analysis ToolPak: Select Correlation o Identify range containing variables
Step 3 Conduct Preliminary Data Analysis Examine Correlation Matrix Is there a statistically significant correlation between Gender/Race and Compensation? What other variables are correlated with Compensation? Are there any large correlations between any of the explanatory variables? (This could lead to multicollinearity, discussed later) Note: This is done without accounting for SSEG s, and as such is informative rather than analytical. Step 4 Create Groups of Employees for Analysis Size of Employee Groups plays a large role in Statistical Power (i.e., the ability to detect a statistically significant difference if it exists to be found) Larger Groups favor those who seek statistically significant differences (e.g. Plaintiffs, OFCCP, EEOC) Smaller Groups favor those who seek no statistically significant differences (e.g. Defendants, Employers) Step 4 Create Groups of Employees for Analysis Groups should represent Employees that are similarly situated with respect to: Work performed Levels of responsibility required Skills needed Qualifications needed
Step 5 Conduct MR Analysis Using Microsoft Excel Using SPSS Excel Verify the Analysis ToolPak is Installed Excel Prepare Data worksheet Dependent (Compensation) Variable in Column A Independent variables in Columns B and on Grouping variable (Gender or Race) in the column following all explanatory variables
Excel To bring up the Data Analysis Menu: For Excel 2007/2010, select the Data tab on the Ribbon, then click Data Analysis (at the right end of the ribbon) For Excel 2003, click on the Tools menu, then the Data Analysis menu option On the Data Analysis Menu, Select Regression Excel Select Range containing Compensation information for Input Y Range Select Range containing explanatory information up to Gender/Race for Input X Range Check the Confidence Interval Box If the data has field names, also check the Labels box Click OK Excel: Interpret Initial Results Under Regression Statistics, check the R-Square value In the ANOVA section, check the Significance F value for Regression Evaluate the p-values for the independent variables
SPSS Run Analysis Regression Diagnostics and supplemental statistical procedures SPSS Interpret Initial Results Review the model summary Review the ANOVA Report Review the Coefficients Report Multicollinearity The key to the interpretation Interaction Terms: When two variables create a third Step 6 Conduct a Cohort Analysis For all SSEG s with an statistically significant differences, investigate whether there is additional information which may also contribute to Compensation decisions. Examples include: Previous Salary Prior Experience Publishing History