In light of the new OFCCP compensation Directive released on Friday August 24, 2018, DCI has done a deep dive into the specifics outlined in the Directive as well as the accompanying FAQs. Although this Directive appears to be more transparent than 2013’s Directive 307, there are components that are still leaving experts on pay equity scratching their heads. It will be interesting to track OFCCP practices during compliance reviews going forward to see how the Directive is interpreted in the field.
Below is our summary of the aspects of the Directive that are helpful to the contractor community and pay equity enforcement, the “bad” aspects for contractors, and other “ugly” questions in the aftermath of a new Directive.
The “Good”
The “good” attributes are marked by OFCCP transparency in how its statistical models are being approached in pay equity enforcement and the application of tried-and-true Title VII standards related to pay equity analyses. They include the following:
- OFCCP clarified that they will analyze base pay, total compensation, and each component of pay separately in a desk audit. Although DCI believes that in many situations a ”total compensation” analysis is NOT useful for regression analytics (how do you accurately regress the same factors on different components of pay?), conducting an analysis on each component of pay separately at the desk audit stage is a practical way to start mirroring how pay works.
- OFCCP clarified that they would allow for market data to be a variable in a regression model if job title is not being controlled for in the regression. This is good news for contractors that have clear market-based information defining how they pay their employees in a manner that is consistent with their compensation systems.
- The Directive allows for age as a proxy for prior experience during the desk audit stage of the review. However, OFCCP specifically stated that they will seek actual prior related experience to include in the model after the desk audit phase. Considering most contractors do not have that data readily available in the HRIS or ATS, this is going to be a tall order.
- The agency will allow for squared terms in a regression model when appropriate. If you are not familiar with squared terms, this is a way to account for situations where time variables are not linear (e.g. salary compression may occur with time and company). Time in an organization is not always linearly rated with pay outcomes, but the squared term in a regression model may be appropriate in some situations.
- The Directive and FAQs give contractors the flexibility to submit certain types of compensation data if there is concern over confidentiality in providing data. Specifically, OFCCP states that if the contractor has a concern with information such as lists of employee names, reasons for termination data, or pay data, then alphabetic or numeric coding or the use of an index of pay and pay ranges are acceptable for the purposes of a compliance evaluation. (Author note: Did OFCCP just say that contractors do not have to submit actual compensation data during a desk audit? If so, how could OFCCP even do a desk audit review of compensation?). Regardless, contractors must provide access to compensation data onsite.
- The Directive clarifies that in disparate treatment pattern and practice cases of compensation discrimination, anecdotal evidence is required in conjunction with statistical significance (with the exception of cases where statistical indicators are overwhelming). Contractors will welcome this interpretation of Title VII standards.
- The Directive clarifies that in a disparate impact case, OFCCP must identify the individual compensation policy or practice that caused the disparity and provide the contractor with an opportunity to justify that practice by demonstrating that it is job related and consistent with business necessity. The key here is that if the allegation is disparate impact, OFCCP is obligated to identify the individual compensation component. (Author note: It is of our opinion that OFCCP cannot use the disparate impact claim on “base pay” due to the fact that it isn’t an individual compensation decision but a culmination of many pay decisions over the course of an employee’s tenure).
- The Directive also clarifies that OFCCP will analyze race/ethnicity in the regression model by creating separate dummy (or vector) codes for each group. Meaning, OFCCP will not conduct a minority to non-minority analysis. In addition, the reference group for racial or ethnic comparisons will be the group with the highest average salary (as long as that group has at least five employees). This allows for transparency for contractors that are conducting proactive analytics on the approach that should be taken when examining racial/ethnic subgroups in their regression analysis.
- The Directive clarifies that pay analysis groups need to be at least 30 employees in size. This is good to know considering DCI has experienced audits and NOVs where OFCCP conducted regression analyses on pay analysis groups with fewer than 30 employees.
- The Directive clarifies that OFCCP recommends a 10 to 1 employee-to-variable ratio for regression analytics. If it exceeds that ratio, OFCCP may argue that the model is saturated. While we appreciate the clarification on the employee-to-variable ratio, context matters. If OFCCP conducts a PAG analysis and needs to control for structural variables to control for similarly situated status via regression (and it exceeds the 10 to 1 ratio) all bets are off.
- The Directive states that after the desk audit stage of the review, if OFCCP needs additional information, the agency must identify an indicator and tell the contractor what the indicator is. This is extremely helpful for contractors attempting to mitigate voluminous data requests without indication of the group of interest. We hope this will narrow OFCCP’s focus in follow-up communications with contractors.
- The Directive suggests that if the OFCCP creates pay analysis groups, these groups should be created in a way that is consistent with the contractor’s compensation system.
- The Directive requires the OFCCP field office to attach a copy of the underlying database and regression results (our preference would have been the actual regression analysis versus results) to any Pre-Determination Notice (PDN) so that the contractor can fully replicate what OFCCP did.
- Finally, the Directive suggests that if the contractor develops pay analysis groups and submits them during the desk audit, OFCCP will use them (if deemed reasonable) at the desk audit stage for initial regression analyses. This is a ‘win’ for contractors who have thoughtful and structured pay systems, and helps curb OFCCP’s creation of inappropriate pay groupings.
The “Bad”
The ”bad” attributes of this Directive are from the perspective of how OFCCP approaches pay groupings and regression modeling during a compliance evaluation.
- This Directive essentially codifies that pooled regression is the norm and that OFCCP will control for similarly situated via structural variables. It is our opinion that consistent with Title VII standards, the analysis should be conducted by similarly situated employee groupings (SSEG) as the norm and in situations where appropriate, the agency could conduct pooled regression when appropriate (See OFCCP Interpretive Standards from 2006). Said another way, pay analysis group (PAG) analyses should be the exception, not the norm. In addition, we believe that the PAG approach controlling for structural variables will likely conflict often with the advice noted above related to the 10 to 1 employee-to-variable ratio in a regression model. Although a pooled, or aggregated, workforce regression analysis may offer better coverage of the workforce than a disaggregated approach, the increase of statistical coverage is meaningless if the model is mis-specified. Unless all levels of each structural variable, including interactions, are accounted for, it may be that OFCCP will still not fully account for differences in compensation that are due to legitimate non-discriminatory factors . For more information and research on this topic, review this paper written by DCI staff.
- The Directive also did not clarify that if OFCCP utilizes pooled regression (by definition a PAG analysis is a pooled regression), it would be theoretically appropriate to test the regression model for interactions. Procedures for evaluating the generalizability of main effects across different SSEGs, such as conducting a Chow Test or modeling interactions, are available but require a fairly sophisticated understanding of multiple regression procedures. In practice, we have observed that most practitioners have difficulty, in most cases, devoting the time and expertise necessary to appropriately investigate the equivalence of regression equations or regression weights across groups of dissimilarly situated employees. Such an observation further underscores our point that the most appropriate way to conduct compensation equity analyses is to establish groups of similarly situated employees and apply the pay model for each similarly situated group of employees separately.
- The Directive suggests and supports OFCCP analyzing multiple years of compensation data and the repeated measures model. For more information on why this may not be as applicable for contractors in an OFCCP context, DCI clients can review our webinar on the topic.
- If a contractor doesn’t submit pay analysis groups, the OFCCP will run the desk audit compensation analysis by either EEO-1 or AAP job group. As most contractors know, these are very broad groupings that are not consistent with the contractor’s compensation system.
The “Ugly”
- The Directive suggests that OFCCP will in most instances run a separate regression for sex and race but reserves the right to interact race and sex in the model. This could make the analysis extremely complicated. The authors find this fact hypocritical in the context that OFCCP acknowledges an interaction term here but not with how the merit variables may interact with SSEGs in the PAG analyses as noted above.
- OFCCP states that they will control for job title and other structural variables in the model but only those job titles that have 5 or more employees. They will then arbitrarily take those titles representing fewer than 5 people and assign them to other titles. We will never know how they did it, nor is this even remotely theoretically appropriate from a compensation standard. How will OFCCP determine where to include these ‘small’ job titles? Will they conduct a full-blown job analysis to determine similar knowledge, skills, and abilities of ‘small’ job titles to ‘larger’ job titles?
Please stay tuned on further insight from DCI on these matters as enforcement of pay equity continues to unfold in light of this new Directive.
By David Cohen, President, and Joanna Colosimo, Director of EEO Compliance, at DCI Consulting Group
