Q&A: A Labor Economist on EEOC’s Pay Data Proposal


The Equal Employment Opportunity Commission hopes to collect compensation data from employers to determine possible bad actors, but pay data gathered at the initial stages of an investigation can be misleading, Paul F. White, Ph.D., a specialist in statistical analysis of employment practices, told Bloomberg BNA.

On Feb. 1, the EEOC published in the Federal Register a notice inviting public input on a proposal to revise the Employer Information Report (EEO-1) to include aggregate pay data annually by gender, race and ethnicity. The comment period ends on April 1. Currently, employers, including federal contractors, are required to file an annual EEO-1 Report, which provides a workforce breakdown by job category and by race, ethnicity and sex.

In next week’s blog Q&A, White, a partner at Resolution Economics, LLC, will discuss using "big data” analytics in the talent acquisition stage and compensation reviews focusing on the “outliers” within the company.

Bloomberg BNA: From a labor economist’s perspective, what are the implications for employers in terms of the pay proposal?

White: While the full effect won’t be known until the regulations are final, under the current proposal some of the primary implications for employers from a labor economics and statistical standpoint involve measurement issues and the interpretation of any statistical analyses that are conducted on this new information.

While the stated intention of the new data collection effort is to help the EEOC determine which employers warrant further investigation, the information gathered at this aggregated level still may create false positives and false negatives.  

As an example of a measurement issue, it appears that the EEOC intends to analyze the hours worked for employees in the same job classification and salary range.  

However, many employers do not collect data on the hours worked for salaried employees. Employers are then faced with the option of starting to collect this information (which is contrary to the EEOC’s desire not to create additional burdens for employers) or they have to use another method for recording hours for salaried employees.  

Given that the hours worked for salaried employees often vary widely, and are sometimes not uniform between race and gender groups, simply assuming that all salaried employees work 40 hours per week is an oversimplification that can lead to false positives or false negatives. The EEOC has invited public commentary on this topic and it will be important that the regulations strike a balance between not being a burden on employers, yet the collected information is not misleading.   

Reading Statistical Tea Leaves

 As an example of the interpretation of the statistical tests, some of the analyses mentioned by the EEOC in the proposal address whether there is a different distribution between males and females in their representation at each salary band level. In other words, do males tend to be over-represented in the higher salary bands within a job classification and females tend to be over-represented in the lower salary bands?  

Some of their suggested analyses show whether there is a different distribution between the protected and non-protected groups, but they do not indicate the direction of the effect. For example, are males favored or are females?

In some instances, this can easily be determined by visually inspecting the counts at each salary band level.  However, the interpretation gets more complicated when the analysis compares multiple protected groups, for example various race and ethnic groups. Is the significant finding due to only one of the multiple groups?  What if one protected group looks favored but another appears to be disadvantaged?    

Other Factors to Consider

Furthermore, employers should be aware that because the data is collected on an aggregated (not individual-level) basis, and because additional information about experience and qualifications is not collected at this stage, statistical tests conducted with data at this level will not sufficiently compare employees who are similarly-situated to one another.  

Even if the EEOC finds that females are significantly disadvantaged with respect to their representation across the various salary bands, it still may be the case that there are no statistically significant differences between male and female compensation when one properly compares employees who are similarly-situated by accounting for the factors that determine compensation levels.  This analysis would need to be conducted using data not being collected by the EEOC at this initial stage of their investigation.

Proposed Pay Bands Problematic

It should also be noted that the combination of aggregated EEO-1 job categories and the wide pay bands suggested by the EEOC could also result in comparisons of employees who are not similarly-situated simply because of different job levels within the same EEO-1 categories.  

If there are significant demographic differences between the workers in these jobs, then an analysis at the EEO-1 category level may yield a false positive due to demographic differences from one job to the other (even if there are no pay disparities within the same job).   

If the EEOC were to investigate every instance of such a disparity, they would likely be looking into an unwieldy number of companies.  Either the EEO-1 categories would need to be further disaggregated, or the pay bands would need to be modified.

Take advantage of an authoritative resource to address the ever-evolving issues in labor and employment law with a free trial to the Labor & Employment Law Resource Center .