How to Conduct an OFCCP-Style Compensation Analysis with Microsoft Excel(Will Begin Momentarily) Jim Higgins, Ed.D. www.bcginstitute.org Visit BCGi Online While you are waiting for the webinar to begin, Don t forget to check out our other training opportunities through the BCGi website. Join our online learning community by signing up (it s free) and we will notify you of our upcoming free training events as well as other information of value to the HR community. 1
HRCI Credit BCG is an HRCI Preferred Provider CE Credits are available for attending this webinar Only those who remain with us for at least 80% of the webinar will be eligible to receive the HRCI training completion form for CE submission i About Our Sponsor: BCG Assisted hundreds of clients with cases involving Equal Employment Opportunity (EEO) / Affirmative Action (AA) (both plaintiff and defense) EEO Litigation Support / OFCCP (federal contracting) Audit Support Compensation Analyses / Test Development and Validation Published: Adverse Impact and Test Validation, 2 nd Ed., as a practical guide for HR professionals Editor & Publisher: EEO Insight an industry e-journal Creator and publisher of a variety of productivity Software/Web Tools: OPAC (Administrative Skills Testing) CritiCall (9-1-1 Dispatcher Testing) AutoAAP (Affirmative Action Software and Services) C 4 (Contact Center Employee Testing) Encounter (Video Situational Judgment Test) Adverse Impact Toolkit (free online at www.disparateimpact.com) AutoGOJA (Automated Guidelines Oriented Job Analysis ) Industry Leader 4 2
How to Conduct an OFCCP-Style Compensation Analysis with Microsoft Excel Jim Higgins, Ed.D. www.bcginstitute.org Jim Higgins, Ed.D. JHiggins@biddle.com Biddle Consulting Group, Inc. 193 Blue Ravine, Ste. 270 Folsom, CA 95630 1-800-999-0438 1-800-999-0438 8 www.biddle.com 3
The Pay Grade Theory The Pay Grade Theory: Rejected However The interpretive standards are not intended to restrict OFCCP s use of pay grade information or any other information as an indicator of potential discrimination. the interpretive standards only foreclose the use of the pay grade theory as the basis upon which OFCCP will allege and establish systemic compensation discrimination in violation of Executive Order 11246 and OFCCP regulations. OFCCP uses pay grade (or other aggregated compensation) information submitted in response to Item 11 of OFCCP s Scheduling Letter. 4
Comparing Compensation in the Past Problems With Simple Comparisons of Group Averages Assumes all employees are equally qualified Fails to account for legitimate reasons for pay differences Too simplistic: Fails to find actual discrimination and finds discrimination where there none really exists 5
Compensation Analyses and the OFCCP OFCCP s Compensation Analysis Strategy 6
What is a Similarly Situated Employee Group (SSEG) Understanding Regression Analyses: The Foundation 7
It All Begins With Correlation Compensation Experience Job Performance Correlation And Regression Think Co Relation Com mpensation Time with Company 8
Interpreting Correlation Coefficients r = +1.00 Perfect positive relationship (Allows perfect predictions) +0.50 0.00-0.50-1.00 Moderate positive relationship (Allows better predictions but there will be error) No relationship (Does not help in making predictions) Moderate negative relationship (Allows better predictions but there will be error) Perfect negative relationship (Allows perfect predictions) Interpreting Correlation Coefficients r = +1.00 +0.50 0.00 The closer to +1.00 or -1.00 the stronger the relationship between two variables -0.50-1.00 The stronger the relationship between two variables, the better the ability to predict one if given the other 9
The Major Limitation of Correlation The Correlation Coefficient is a statistical tool that is limited to looking at two variables at a time. You can look a the relationship between Experience and Compensation You can look at the relationship between Education and Compensation But you can t look at both education and experience and their relationship to compensation at the same time. Correlation DOES NOT EQUAL CAUSATION. Enter Linear Regression 10
Correlation and Regression r =.53 r 2 = 28.09% Com mpensation Y Time with Company X Regression/Prediction Line Correlation and Regression r =.21 r 2 = 4.41% Com mpensation Age (as proxy for experience) Regression/Prediction Line 11
Correlation and Regression r =.36 r 2 = 12.96% Com mpensation Regression/Prediction Line Performance Appraisal Score A Note About Error Prediction Error! r =.53 r 2 = 28.09% Com mpensation Time with Company An employee s actual compensation An employee s predicted compensation 12
A Note About Outliers Skewed regression line r =.30 r 2 = 9.00% Com mpensation Time with Company The correct regression line What Is Multiple Regression And How Does It Help? Correlation Linear Regression Multiple Regression is like linear Regression on Steroids! 13
Why Multiple Regression? Experience? Performance Differences in Compensation Job Market Factors Tenure Education What Exactly Is Multiple Regression? Experience Performance Multiple Regression Examines the relationships between various potential explanatory factors and compensation. Tenure Objective The goal is to create a statistical model that explains why employees receive the compensation they do. Compensation 14
How Does Multiple Regression Help Find Signs of Pay Discrimination? Multiple Regression All variables together become the basis for a prediction model known as a regression model. The regression model predicts a certain percentage of what makes up an employee s compensation. R =.67 R 2 =.45 p Compensation Age/Experience Time In Company Perf. App. Score 15
Multiple Regression Q: So how does regression help to identify discrimination in pay? A: If the prediction model becomes significantly better after including the protected variable. Gender Age/Experience R 2 =.45 without gender Time In Company Compensation Perf. App. Score R 2 =.51 with gender Actually It Is MUCH More Complicated The illustration we have been using assumes that each of the variables is ONLY correlated with compensation and are NOT correlated with each other. In the real world, experience is correlated with age, age is correlated with education and therefore experience, experience is related to job performance and therefore age, etc. In other words, the real world is a mess! 16
Meanwhile, Since We Don t Live Anywhere Near Perfect Compensation Previous Experience Job Performance Gender Age Using Microsoft Excel to Conduct a Simple Regression Analysis 17
How to Run Multiple Regression in Microsoft Excel Step 1 Install the data analysis toolpak Step 2 Gather the appropriate data that you believe your organization uses to make compensation decisions Step 3 Run Compensation Analyses Step 4 Consider anecdotal data/cohort analysis Step 5 Calculate amount needed to eliminate statistical significance if necessary How to Install the ToolPak 18
How to Install the TookPak How to Install the TookPak 19
What the ToolPak Does For You Gather the Appropriate Data Go for what is easy unless you are preparing for litigation. The OFCCP usually starts their analysis with a standard set of variables: Time in Company (a tenure variable) Time in Job (a tenure variables) Age (as a proxy for experience) Job Performance (be careful here!) The Protected t variables (white/minority, i Male/Female, Over/Under 40) 20
A Word About Tainted Variables Tainted Untainted Formatting your Data Gender (Female = 0, Male = 1) Minority Status (Minority = 0, White = 1) Age (Years? Months?) Performance (Scale? 1 to 5?) Time in Job (Months? Years?) Time in Company (Months? Years?) Compensation (Hourly? Annualized?) 21
The Data Compensation TIJ TIC Age Performance Gender $33,000 2 9 36 3 0 $37,200 6 8 40 5 0 $29,200 1 2 35 3 0 $31,666 5 5 42 4 0 $34,888 6 7 34 4 0 $40,175 8 9 44 5 1 $38,216 5 8 31 3 1 $42,876 7 10 39 4 1 $36,989 6 8 46 3 1 $39,998 7 7 50 5 1 $43,155 9 9 48 4 1 Based solely on comparing average salaries there is a difference! t-test: Two-Sample Assuming Equal Variances Female Male Mean Compensation $33,190.80 $40,234.83 Observations 5.00 6.00 t Stat -4.25 Statistical Significance 0.00 t Critical two-tail 2.26 22
Correlations Compensation 1.0000 Compensation TIJ TIC Age Perf Gender TIJ 0.8680 1.0000 TIC 0.7809 0.6128 1.0000 Age 0.4772 0.6222 0.2078 1.0000 Perf 0.4543 0.6393 0.2379 0.5185 1.0000 Gender 0.8170 0.6589 0.5333 0.4765 0.1257 1.0000 Look for the variables that are related to compensation (Those that correlate with compensation with values that are closer to 1.00 than 0.00) Multiple Regression Analysis SUMMARY OUTPUT Regression Statistics Multiple R 0.966 R Square 0.934 Adjusted R Square 0.868 Standard Error 1638.692 Observations 11.000 ANOVA df SS MS F Significance F Regression 5.000 189312363.721 37862472.744 14.100 0.006 Residual 5.000 13426564.279 2685312.856 Total 10.000 202738928.000 Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Intercept 26045.716 4747.446 5.486 0.003 13842.038 38249.394 TIJ 626.876 487.345 1.286 0.255-625.882 1879.634 TIC 635.045 315.320 2.014 0.100-175.509 1445.600 Age -68.415 118.636-0.577 0.589-373.378 236.548 Perf 857.636 992.784 0.864 0.427-1694.393 3409.665 Gender 3914.397 1613.516 2.426 0.060-233.273 8062.066 23
Breaking it Down Regression Statistics Multiple R 0.966 RS Square 0.934 Adjusted R Square 0.868 Standard Error 1638.692 Observations 11.000 Adjusted R Square tells you the percent of compensation that is accounted for by the variables in your model when they are taken into account as a whole. The larger this number, the better! Breaking it Down (Continued) ANOVA Significance df SS MS F F Regression 5.000 189312363.721 37862472.744 14.100 0.006 Residual 5.000 13426564.279 2685312.856 Total 10.000 202738928.000 ANOVA is a test of whether the variables you have included in your model do a good job of explaining compensation. If the Significance of F is less than 0.05, you may conclude that your model is working. The F statistic is the ratio of explained variance to error or unexplained variance. 24
The Heart of Multiple Regression Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Intercept 26045.716 4747.446 5.486 0.003 13842.038 38249.394 TIJ 626.876 487.345 1.286 0.255-625.882 1879.634 TIC 635.045 315.320 2.014 0.100-175.509 1445.600 Age -68.415 118.636-0.577 0.589-373.378 236.548 Perf 857.636 992.784 0.864 0.427-1694.393 3409.665 Gender 3914.397 1613.516 2.426 0.060-233.273 8062.066 The key number to look at here is whether your protected variable (e.g., gender) is a statistically significant predictor. You tell this by looking at the p-value for Gender. If the p-value is less than 0.05, it means that even after controlling for legitimate job-related variables, gender still has an effect on compensation. If the p-value is greater than 0.05, it means that after you have controlled for legitimate job-related variables, gender has does not appear to influence compensation. You can t just point and click. Multiple Regression Places Certain Assumptions on Your Data 25
Assumptions of Linear Regression All relevant variables are included in the prediction model All irrelevant variables are excluded from the prediction model The relationship between predictors/explanatory variables is best characterized by a straight line. The relationship between variables is homoscedastic None of your variables is perfectly correlated with any other variable (called colinearity) No combination of variables is perfectly correlated with another of your explanatory variables (called multicolinearity) None of your variables is a constant Confirming Linearity Statistical Techniques Qualitative i (interocular fusion test) 26
Multicolinearity Review correlation matrix looking for perfect correlations (or close to it) between predictors. Run a multiple regression that does not include compensation but instead try using all combinations of predictors against each single predictor and look for near perfect R 2. Age + Performance Education Education + Performance Age Education + Age Performance Homoscedasticity Statistical Techniques Qualitative i (interocular fusion test) 27
Compensation Analysis Pitfalls and Issues to Consider Analysis Pitfalls and Issues to Consider The dog chasing its tail: Making compensation changes to one group can affect others (for better or worse) Rectifying problem areas for women may create problem areas for minorities Rectifying problems in one SSEG may create problems for a department, location, manager, etc. 28
Analysis Pitfalls and Issues to Consider Regression analyses can be very data intensive Statistical Power issues Missing data can easily undermine your analyses Missing variables Missing data within a variable (regression analyses typically require all data for all records) Be sure to analyze your explanatory variables for inequities between groups (e.g., performance appraisal scores) Analysis Pitfalls and Issues to Consider Be sure to evaluate and include an adequate timeframe for your data (e.g., performance appraisal scores, productivity metrics, etc.) Flip-flops in disparities (against men/whites in some circumstances and women/minorities in others) may mean your organization does not systematically discriminate... this strategy has been used to undermine class claims of discrimination. Statistics are cold and must be supported by anecdotal evidence Interviews Personnel files/records (i.e., cohort analysis) 29
QUESTIONS AND COMMENts I want to hear from you! Let me know if I can help you in any way. Jim Higgins Jhiggins@biddle.com (916) 204-1749 www.bcginstitute.org 30