8

Conclusions and Recommendations

OVERALL ASSESSMENT

Value of Data as Collected

Although other federal surveys collect data on pay and demographic characteristics, those data are based on individual employees. EEOC surveys differ by collecting data from employers. Component 2 data can be used to identify between-workplace/firm and within-workplace occupational-segregation components of pay gaps. The ability to decompose total pay gaps into workplace, occupation, and within-job components is a fundamental advance in EEOC’s ability to identify whether pay disparities are a function of hiring, segregation, or direct pay disparities. Examples of intuitive and relatively simple decompositions are available in the scientific literature (e.g., Smith-Doerr et al., 2019). Such estimates can support EEOC’s allocation of resources in targeting particular firms, industries, or localities for hiring or job-placement discrimination under Title VII of the Civil Rights Act of 1964 versus pay disparities under the Equal Pay Act of 1963.

During charge investigations, Component 2 data could be suitable for calculating raw sex and race/ethnicity pay gaps across and within job categories for establishments that are charged. Specific pay-gap comparisons should be guided by the charge bases (sex, race, color, national origin) and job/employment issues named in the charge(s). Component 2 data are also suitable for establishing local labor-market/industry average pay gaps and making comparisons between targeted establishments and local labor-market averages using tests of statistical significance. If exact pay data are available from an establishment that generates a pay discrimination charge, it is also possible to make comparisons to peer establishments in the same industry and locality. Such comparisons will be less statistically uncertain than comparisons relying only on Component 2 data, for both the focal establishment and comparison statistics.

Component 2 data could be suitable for calculating raw local labor-market and/or industry-wide pay gaps by sex, race, and/or ethnicity, and for identifying outlier establishments. Such analyses could be used to flag high pay-disparity contexts to inform national or regional enforcement plans, and/or to identify outlier establishments and firms (i.e., those that show exceptionally large raw pay gaps) for further consideration.

Component 2 data could be suitable for employer self-assessment of raw pay gaps but not for finer-grained pay-equity or job-segregation analyses. Establishment profiles, such as those presented in Table 7-4, could be utilized by employers to identify sex, race/ethnicity, and occupation groups with large pay gaps for further assessment. Analysts could then consult individual-level human resource information system data on hours/weeks

worked, tenure, education, job duties, and other factors that affect pay, to adjust raw pay gaps identified by Component 2 profiles.

CONCLUSION 1-1: The 2017–2018 Component 2 data are a potentially valuable resource. They are the only federal data source for pay data and demographic characteristics collected at the employer level, which is helpful for enforcement efforts, for employer self-assessment, and for providing a broad description of pay practices.

However, to be useful, data must be complete and robust. To assist EEOC’s initial investigation of employers facing charges, the employer’s complete data must be included in the Component 2 data file. The panel found numerous and varied data issues in the 2017–2018 Component 2 data collection. Figure 8-1 provides a high-level summary of the panel’s estimates of various key dimensions of data quality for the 2018 Component 2 data: coverage, missing data, and extreme values. Findings for 2017 Component 2 data were similar.

The panel found Component 2 data to be incomplete in three ways. Only two-thirds (65%) of eligible firms were asked to complete the surveys. Almost all firms asked to complete the Component 2 instrument responded, along with volunteers, totaling 58 percent of eligible firms and covering 82 percent of establishments. But, some responding firms chose the option not to provide pay data for their establishments with fewer than 50 employees (i.e., by submitting Type 6 forms), resulting in pay data for only about 68 percent of responding establishments. For these reasons, the coverage rate for pay data is only 58 percent for firms and 55 percent for establishments. In other words, only about half of eligible firms and establishments provided any data for pay analysis (see Chapter 4).

The panel also found concerns with Component 2 data reliability. Although most of the reported numbers of employees and hours worked appeared to be reliable, extreme errors in some of the reported numbers of employees and/or hours worked could lead to highly misleading results if not addressed prior to analysis. For example, one employer reported having more than 250 million employees (the largest U.S. employer has roughly 1.4 million employees), and many employee counts were more than nine times larger in Component 2 data than in Component 1 data, despite both instruments covering a time period within the same three months of the year. Data issues are numerous and varied in their likely causes. These concerns led the panel to exclude 35 percent of the provided pay data (at the establishment level) for the purposes of the exemplar investigations presented in this report, though the most appropriate rules for editing and excluding data will vary depending on the purposes of the analysis (see Chapter 5).

Moreover, there must be sufficient within-job-category differentiation in pay (i.e., at least two pay bands per sex-race/ethnicity-occupation [SRO] group) and sufficient representation of sex and race/ethnicity groups (i.e., at least two groups). If these criteria are met, analyses of targeted establishments’ pay gaps could serve as a first step in charge investigations, help assign investigative priority to charges, and/or inform requests for additional data, such as individual-level information on pay, hours worked, tenure, and education, to calculate adjusted gaps (see Chapter 7).

As described in Chapter 2, the panel recognizes that the 2017–2018 Component 2 data were collected after a court order to immediately proceed with data collection. Although response rates were high in this mandatory data collection, coverage was insufficient (especially for smaller size firms) (see Chapter 4).

CONCLUSION 4-1: As collected, 2017–2018 Component 2 data are limited by significant data coverage issues related primarily to EEOC’s

master list of potential respondents. In addition to respondent coverage, the panel identified issues with both nonresponse and measurement, which should be recognized when using these data.

Identifying Eligible Filers

As discussed in Chapter 4, the Component 2 data contained 63,623 firms with 100 or more employees in 2017, compared with 113,556 in the Business Dynamics Statistics (BDS) database, and contained 139,585 establishments with 100 or more employees, compared with 182,484 in the BDS and 181,111 in the Quarterly Census of Employment and Wages. Thus, EEOC data are highly incomplete, affecting EEOC’s ability to investigate individual establishments, make comparisons among peer establishments, and its ability to produce accurate national statistics. Additionally, the panel found that firm and establishment identifiers were neither consistent nor unique. For example, establishments are renumbered each year, without attempt to keep numbering consistent. As described in Chapters 4 and 5, this inconsistency impedes the trend analysis and data quality checking possible when merging by identifiers.

CONCLUSION 4-2: The 2017–2018 Component 2 data have inconsistent and non-unique firm and establishment identifiers, which impede the maintenance of the master list, trend analysis over time, and data quality checking possible when merging by identifiers.

Measurement Concerns

The panel also identified a number of measurement concerns in the 2017–2018 Component 2 data collection. As described in Chapter 3, these concerns include the measurement of pay, occupation, protected groups, and sources of legitimate pay differences.

Regarding measurement of pay, the panel concluded that W-2 Box 1 does not reflect total compensation, as it omits income that is not federally taxable, such as employee contributions to 401(k) or 403(b) plans, and therefore may mask pay disparities.

CONCLUSION 3-1: The 2017–2018 Component 2 collection measure of pay (W-2 Box 1) only partially reflects total compensation and may therefore mask compensation differences.

EEOC decided to collect pay data using a categorical measure, with 12 pay bands. These bands are overly wide, leading to a lack of variation in pay and compromising the enforcement utility of Component 2 data, especially with respect to calculating pay gaps at targeted establishments.

This problem is more acute for small establishments and for job categories that lack differentiation in pay (i.e., the highest- and lowest-paying job categories). In the Silicon Valley technology sector, lack of pay differentiation was most problematic for computing pay gaps for executives, with the pay for all or most employees falling in the top pay band and no means of differentiating pay rates. However, in other industries such as retail, this issue will impact low-wage occupations, such as service workers. Furthermore, asking employers to collapse data into pay bands that do not match their actual work practices or routine reporting to other state and federal agencies reduces precision in measurement, increases respondent burden, and reduces regulatory utility. Use of pay bands is not necessary to protect confidentiality.

CONCLUSION 3-2: Use of pay bands in the 2017–2018 Component 2 data collection provides information that is less useful than that provided by individual-level pay data. Using established, improved methods, other federal agencies have demonstrated that individual-level pay data can substantially reduce respondent burden, increase precision in estimating pay gaps, and protect confidentiality. The Bureau of Labor Statistics’ Occupational Employment and Wage Statistics collection is an example.

The 10 job categories used in the data-collection instrument to measure occupation are extremely broad and tend to encompass a wide range of job responsibilities and pay rates. Investigators and researchers need more precise information to determine whether some workers experience pay disparities. For example, about 70 percent of employees in the executive category fall under the three highest pay bands, and about 75 percent of service workers fall under the three lowest pay bands. The concentration of workers in a small number of pay bands may produce establishment reports with insufficient pay-band variation within occupations, as well as overly wide pay bands that make it difficult to detect large pay disparities.

The job categories are also outdated and do not adequately represent the modern workforce. The appropriateness of EEOC’s job categories was questioned by respondents in the business community, and these categories have not been updated to conform to changes in work and federal statistical conventions.¹

CONCLUSION 3-3: The job categories in the 2017–2018 Component 2 data collection are insufficient for describing the modern workforce.

___________________

¹ This text was changed after release of the pre-publication version of the report to correct an error regarding the last update of EEO-1 job categories.

The Component 2 data collection did not fully measure gender identity or sexual orientation (protected under sex) or race/ethnicity, which could impede comparisons of pay for similarly situated employees. Employers could volunteer such information in a section for text comments, but the data were never explicitly collected.

CONCLUSION 3-4: The 2017–2018 Component 2 data collection does not fully measure EEOC-protected groups under the sex and race/ethnicity concepts. The instrument does not provide a way to identify more than one specific race for an individual, nor do the data support distinguishing Hispanic persons by race. The instrument does not collect data on LGBTQIA+ status.

As described in Chapters 1 and 3, measuring pay differences experienced by persons age 40 and older, persons with disabilities, and veterans is within the scope of EEOC’s authority and equities. However, these pay differences cannot be examined with current Component 2 data.

CONCLUSION 3-5: The 2017–2018 Component 2 data collection does not allow measurement of pay differences experienced by other groups protected by EEOC’s authority. Measuring pay differences experienced by persons age 40 and older, persons with disabilities, and veterans is within the scope of EEOC’s authority and policy equities.

Aside from job categories, other important and legitimate causes of pay differences, such as education and tenure, are not included in the Component 2 data collection. As described in Chapter 3, without this information, Component 2 data will not contain important information for EEOC enforcement actions and employers’ self-assessments. Employer self-assessments could contribute to improved employment equity by self-monitoring and adjustment.

CONCLUSION 3-6: The 2017–2018 Component 2 data collection does not include measures of legitimate causes of pay differences, such as educational attainment and tenure. Such information would assist both EEOC enforcement efforts and employers’ self-assessments. Employer self-assessments could contribute to improved employment equity through self-monitoring and adjustment.

APPROPRIATE USE

The panel found numerous measurement issues in the Component 2 data that preclude certain types of analyses, such as examining specific firms

or establishments that have missing or problematic data. The data checking protocols were not well developed (Chapters 2 and 5), and data cleaning after collection was limited. All data-collection efforts require multiple data checks to insure quality analysis files; Component 2 data are no different in this regard. In the panel’s opinion, these data quality issues should be addressed before use.

To the extent that EEOC can either fix certain data issues or filter out problematic data, the remaining data can be useful. Specifically, after cleaning and weighting, 2017–2018 Component 2 data can also be used to obtain estimates of pay differences at the national level, by sex, race/ethnicity, and occupation (see Chapter 6). In addition, these data may be used to calculate raw annual pay gaps at the local labor-market level for establishments targeted by individual charges, and for systemic enforcement in larger establishments (see Chapter 7).

CONCLUSION 6-1: After cleaning, 2017–2018 Component 2 data could be used to obtain estimates of raw pay gaps at the national level by sex, race/ethnicity, and occupation.

CONCLUSION 7-1: After cleaning, 2017–2018 Component 2 data could be used, with limitations, as an initial step to prioritize investigations and the allocation of resources: (1) to calculate raw annual pay gaps for establishments under investigation by individual charges; (2) to make comparisons between investigated establishments and peer establishments in the same industry and metropolitan area, county, and/or core-based statistical area; and (3) for systemic investigations.

There are additional considerations when calculating hourly wages. First, some data appeared highly problematic. The panel applied elementary filters to address these deficiencies, but filtering was not as extensive for work hours as for employee counts; and the lack of comparable data from the Component 1 data collection limited the extent of filtering used even for employee counts. Second, full-time, part-time, and part-year employees are all grouped together in Component 2 data based on annual wages, but annual wages reflect different hourly wages for individual employees, and it is unclear how hours worked should be apportioned to calculate hourly pay rates. The straightforward strategy of assuming equal hours worked for employees in the same sex-race/ethnicity-occupation-pay band results in implausible hourly wage values for some sex-race/ethnicity-occupation (SRO) groups. Thus, these data are not suitable for calculating hourly wages (see Chapter 7). Extensive cleaning of existing data may eliminate implausible values, but the panel suggests caution when converting annual pay to hourly wage rates. Critical information on hours worked is lost

when examining raw annual pay gaps without adjusting for work hours. Regression-based adjustment of yearly pay is preferable to attempting to produce hourly wage estimates, although even here caution is warranted. In addition, hours-worked data as collected are not suitable for determining bias or reasonable cause, nor for employer self-assessments regarding the allocation of work hours (see Chapters 3 and 7).

The panel noted that the 2017–2018 Component 2 data have limited utility for analyzing pay differences within small establishments, depending upon the amount of variation present in employee characteristics of pay, sex, race/ethnicity, and occupation. This is not due to problems in the data as such but to the practical reality that, for example, if only two professionals were employed by a firm and they were both in the same pay band, the database would be insufficient for examining pay differences in a classical statistical framework. Raw difference may still be meaningful, to the extent that there is sufficient variation in pay bands within establishments.

For a substantial share of charges, calculating sex and/or race/ethnicity pay gaps and making peer comparisons will be impossible due to few or no workers in relevant SRO cells. In occupations with low sex and race/ethnicity diversity, such as executives in the Silicon Valley technology sector, it may be impossible to calculate pay gaps for a targeted establishment due to the absence of workers in specific SRO cells. This potentially reflects underlying bias in hiring, job assignments, and promotion opportunities for historically disadvantaged groups, which may be the focus of discrimination charges on those bases. This problem will be especially pronounced for charges filed against small establishments, against establishments in (race/ethnicity) homogenous labor markets, for charges involving high-level jobs, and/or for intersectional charges (e.g., race/ethnicity and sex). Similar issues will complicate employer self-assessments of pay gaps, especially for small-size establishments and those in homogenous labor markets. Comparisons of individual-level pay with aggregate pay gaps at the establishment level are still possible but cannot be contextualized relative to sex, race/ethnicity, and occupation peers when few or no peers exist.

CONCLUSION 7-2: The 2017–2018 Component 2 data have limited utility in analyzing pay differences within establishments that lack variation in employee characteristics of pay, sex, race/ethnicity, and occupation.

Component 2 data produced results similar to those that would be obtained using individual-level wage data. After cleaning, Component 2 data are thus suitable for estimating raw pay gaps and could be used for initial investigative work. However, these data lack sufficient detail and accuracy to support enforcement actions. The key difficulties are the lack

of individual-level wage data, the broadness of job and pay categories, and the inability to accurately compute hourly wages.

CONCLUSION 7-3: Without extensive cleaning, 2017–2018 Component 2 hours-worked data are unsuitable for calculating hourly wages.

CONCLUSION 7-4: The 2017–2018 Component 2 data are unsuitable for direct determinations of bias or reasonable cause for enforcement purposes.

IMPROVEMENTS NECESSARY IN THE SHORT TERM

Drawing upon the conclusions above, the panel identified a number of recommendations to improve future collections of Component 2 data. These are divided into necessary and relatively modest improvements in the short term, and additional improvements for EEOC to implement as part of a broader effort to strengthen Component 2 data and thereby more fully advance the Commission’s mission. Although short-term recommendations are presented first, the panel encourages EEOC to act more broadly to improve the Commission’s pay-data collection in the future. Adoption of broader data-modernization recommendations would lessen the need for shorter-term recommendations.

Address Likely Sources of Error

The quality of data collection is a function of identifying the target group of interest (population coverage or completeness), receiving responses from that group (response cooperation), the completeness and accuracy of those responses (measurement error), and conceptual coverage of the topic domain. In organizational surveys, unit and item nonresponse are typically conceptualized as a product of the authority, motive, and capacity of designated respondents to comply with survey requests (Tomaskovic-Devey et al., 1994, 1995; Biemer and Lyberg, 2003). Errors in item response are typically driven by respondent capacity to answer, survey design errors, and mode of administration.

Since EEO-1 reporting is federally mandated, firms should respond regardless of their desire to participate. The Component 2 data collection was court mandated to remain in the field until an acceptable response rate was reached; accordingly, it had a response rate of more than 90 percent (see Chapter 4).

However, the panel finds error associated with mode of response (data-upload mode versus online-entry mode; see Chapters 4 and 5), as well

as error that appears to stem from respondent confusion about how to respond, possibly due to survey design (see Chapters 3–5). Additional errors are suspected due to concepts measured. Below, the panel enumerates strategies that could address these sources of measurement error.

At least two types of errors can potentially be attributed to the design of EEOC’s survey instrument. First, respondents sometimes entered nonzero data for one field (employee counts or hours worked) while entering zero or no data in the other. Second, some employee counts appeared to be responses on hours worked (e.g., an employee count of 2,080). Placing employee-count and hours-worked fields side-by-side, as in the data-upload mode, may help to lessen such discrepancies. Other aspects of instrument design, instructions, and data storage by respondents may also be important for these and other errors. Some errors could be caught and fixed by adding more online edits that will warn respondents about problematic data.

When an EEO-1 report is submitted, an automated data check is advised to confirm that blanks, rather than zeros, are appropriate; and that row and column cells sum to row and column totals. The panel recommends that improbable response flags be reported to the respondent prior to final submission. Identifying improbable responses might be based on prior-year establishment EEO-1 reports or prior-year size-industry-geography mean values. The panel advises that the ratio of hours worked to employee counts be examined, flagging data showing ratios that are either too high (e.g., above 5,160 hours worked per employee) or too low (e.g., below 100 hours worked per employee). If EEOC moves toward collecting individual-level wage data as recommended, then data might be flagged based on computed hourly or weekly wages that are exceptionally high or low as in conventional individual survey analyses.

The panel advises that all automated data checking be implemented prior to data certification. The panel’s analyses suggest that only a small fraction of respondents will need to make these corrections, but the improvement in the quality of EEO-1 reports and their trustworthiness for making regulatory decisions will be dramatically improved.

CONCLUSION 5-1: Important data-quality issues exist in the 2017–2018 Component 2 data, including missing data, response inconsistencies, implausible extreme values, and measurement unreliability. These errors are large and, if not addressed, could generate misleading results. Filtering the data on number of employees by removing a small amount of data can address some, but not all, issues.

RECOMMENDATION 5-1: Before 2017–2018 Component 2 data are used to assist initial investigations of charges, for employer self-assessment, or for research on pay differences more generally, the data

should be carefully reviewed and cleaned. Filtering on employee counts and on hours worked would be beneficial, but some issues would be best addressed by modifying the basic data-collection methodology.

CONCLUSION 3-7: Improvements could be made to the current Component 2 instrument to substantially reduce, and possibly eliminate, many of the errors and weaknesses observed in the 2017–2018 Component 2 data.

The 2013 National Research Council report recommended that EEOC field a pilot study prior to pay-data collection. A pilot study is often recommended to adjust survey procedures; to improve data-collection procedures and measurement choices; and to anticipate data-processing needs after survey submission. Since a pilot study was not done, the Component 2 data collection is the source of much valuable information for increasing the quality of future pay-data collections. In addition to the current panel’s recommendations for improved data quality, the panel further advises that the two methodological reports prepared by the National Opinion Research Center at the University of Chicago should be consulted closely for future data collections, especially if future collections are based on the legacy EEO-1 reporting format.

The panel also recommends that future data-collection efforts be accompanied by pilot studies to assess data quality and fit-for-use issues. More generally, EEOC should routinely perform or delegate data quality assessments of respondent frames, response rates, conceptual appropriateness, item nonresponse, and item quality for its survey programs.

RECOMMENDATION 5-2: Before future collection of Component 2 data, EEOC should conduct a field test to investigate issues of burden, data availability, and instrument design. The field test should examine the sources of errors in the hours-worked and employee count data, and should assess the functioning of new survey questions. Solutions to be tested may include placing employee-count and hours-worked data side-by-side, as in the data-upload mode. Cognitive interviews may inform EEOC of employers’ interpretations of survey questions, difficulties faced in answering, and strategies used to obtain the reported data.

The panel advises that the following recommendations be considered part of a comprehensive plan to ensure the continued relevance, accuracy, and (appropriate) accessibility of EEOC data. Periodic review and steady progress toward continuous improvement signals commitment to data quality for all federal agencies conducting data collection, and may be especially

important to agencies that use their data for enforcement purposes (OMB, 2014).

EEOC originally designed the EEO data collection as a single survey, and the data collection was only divided into two components as the result of the Office of Management and Budget (OMB) rescinding its approval, stopping data collection for what became Component 2. Dividing the data collection into two parts had several consequences. Filers were required to report twice, thus duplicating their efforts; filers could (and often did) pick two separate reporting periods (though still generally within the same three-month time frame), creating the potential for conflicting responses; and data errors became more difficult to identify and correct because of the possibility of real change between the two reporting periods.

RECOMMENDATION 2-1: EEOC should combine the Component 1 and 2 instruments into a single data-collection instrument, thus lessening respondent burden and reducing the chances for inconsistencies or reporting errors.

When reporting for establishments with fewer than 50 employees, filers could submit either Type 6 reports (containing only broad, summary data) or Type 8 reports (containing the same level of detail as for larger establishments). About half of the greater than 40 percent noncoverage of pay data in the Component 2 data collection resulted from firms electing to file Type 6 rather than Type 8 reports. Submission of Type 6 reports results in gaps in the data, reducing the value of the data for enforcement actions and research, and complicating efforts to inspect and clean the data. Type 6 reports were allowed in place of more complete reports to reduce burden, but the burden reduction is likely to be non-existent—multi-establishment firms are still required to file a consolidated firm-level report (Type 2), which includes the complete sex by race/ethnicity by occupation by pay band counts for all establishments, regardless of size. So, even though they are not reported in a Type 6 report, the data for smaller establishments must still be compiled, combined with all other establishment reports, and entered in the Type 2 consolidated report. Furthermore, by eliminating Type 6 reports, consolidated reports could be generated by automatically summing across all submitted establishments in a reporting firm. This would reduce respondent burden and increase data quality, since current respondents must prepare consolidated reports whenever they submit Type 6 reports.

RECOMMENDATION 2-2: EEOC should eliminate Type 6 reports and mandate Type 8 reports for all establishments in multi-establishment firms of 100 or more employees. Consolidated reports (Type 2) could then be eliminated, and firm-level data created by summing across

establishment reports. These actions would increase coverage, simplify reporting, and reduce respondent burden.

When picking the pay period for identifying employees, respondents almost always selected the October-through-December time period for both the Component 1 and Component 2 instruments, as requested, but often picked different pay periods. This meant (1) respondents ended up compiling the data twice; and (2) if responses given for Component 1 and 2 instruments differed, it was unclear whether there was a data error or a true change over time.

RECOMMENDATION 3-1: EEOC should implement a standard reporting period to improve comparability of data and reduce respondent burden.

In 2017 and 2018, 7.1 percent and 7.5 percent of firms, respectively, subcontracted their submissions to external vendors known as professional employer organizations (PEOs) (see Chapter 4). The response rates and data quality for PEOs was generally superior to that of self-filing, providing a clear advantage to continued PEO filings. However, PEOs sometimes reported their own data as belonging to the firms they were filing for (i.e., PEOs reported their own employer identification numbers or North American Industry Classification System categories for client firms for which they were filing).

Under current EEOC regulations, the firm that certifies that the responses are accurate should be the responsible firm, not the PEO. In the panel’s judgment, either EEOC’s certification rule should change or there should be a second step after the PEO has uploaded the data, so that a representative of the employing firm can review and certify the data submission. Since EEOC also intends that firms will use their EEO-1 submissions for self-assessment, it seems most reasonable that the legally responsible submitting firm certify the data. This step may also improve data accuracy.

RECOMMENDATION 3-2: EEOC should require professional employer organizations (PEOs) to submit data separately for each firm they represent, use the client firm’s industry code, and require employing firms to certify PEO submissions before filing.

The Component 1 instrument provided filers with a method for downloading and reviewing their responses, but the Component 2 instrument did not.

RECOMMENDATION 8-1: EEOC should provide filers with a method to download and review responses before submission. This will support data quality and assist with employer self-assessment. Such a method is currently provided for the Component 1 instrument but not for the Component 2 instrument.

The panel found substantial unit undercoverage in the Component 2 data, and recommends improvement in both the respondent frame and outreach to newly eligible firms. Chapter 4 describes that when compared to external Bureau of Labor Statistics and Census Bureau firm and establishment benchmarks, the EEO-1 response frame and resulting data file have substantial undercoverage of eligible entities, particularly for establishments with fewer than 50 employees. About half of the establishment noncoverage for pay data appears to be a function of firms missing from the respondent frame. Firm and establishment coverage differ somewhat across industries and states as well, leading to the potential for bias in calculating national and sub-national statistics.

Although the panel’s estimates are not definitive, it appears that about 20 percent of establishments legally required to file Component 2 are missing from the final data collection, and more than 40 percent lack pay data. This suggests that when EEOC uses these data for enforcement purposes, many workplaces will be missing from the database.

Since response rates are high in this mandatory data collection, improvements in potential respondent frame coverage along with elimination of the Type 6 response option would significantly improve coverage of the missing establishments/firms. This would help with EEOC’s routine regulatory functions and could reduce population bias with respect to industry and local labor-market benchmarks, both of which are central to EEOC case evaluation. Furthermore, firms and establishments are formed, sold, or go out of business continuously, and the contacts at these workplaces change frequently (EEOC, 2020h), continuous outreach efforts to firms and establishments would be beneficial. Prescreening calls or emails to update this information prior to subsequent field work could be considered as well.

There are multiple ways to potentially improve respondent frame coverage. The panel notes that an interagency agreement between EEOC and BLS could be leveraged to assist assessment of coverage and to help with statistical weighting to adjust for nonresponse and undercoverage. EEOC might also investigate potential cooperation with human resource firms and payroll software firms to automatically notify clients when it is time to submit EEO-1 reports. In the interim, EEOC could consider developing data-sharing agreements with states, such as Illinois and California, where parallel pay-data collection initiatives and local firm respondent lists are being generated.

RECOMMENDATION 4-1: EEOC should improve the coverage of its master list, perhaps using an interagency agreement between EEOC and the Bureau of Labor Statistics to appropriately maintain business registers.

Survey statisticians commonly use weighting to adjust for the probability of selection into a sample and to adjust for nonresponse. Weighting helps to produce national totals that correspond to known benchmarks (and that are not underestimates), and to adjust for possible bias if some types of firms respond at a higher rate than others. Since EEOC’s data collection is intended to be a census of all firms rather than a probability sample, in principle it should not be necessary to adjust for the probability of selection; however, since EEOC’s master list appears to miss many of the firms counted by the Census Bureau and BLS, weighting would be appropriate to adjust for undercoverage until EEOC’s frame is comprehensive. Statistical weighting to adjust for nonresponse will always be useful.

RECOMMENDATION 4-2: When preparing national and sub-national statistics for the public, EEOC should adopt statistical weighting to adjust for possible undercoverage and nonresponse biases.

As described in Chapters 4 and 5, decisions by EEOC regarding the assignment of identification numbers made it difficult to match Component 2 data with Component 1 data, and to match Component 2 data from 2017 with Component 2 data from 2018. The panel understands this was done to protect the confidentiality of EEO-1 respondents. However, there are several preferred ways to maintain confidentiality that do not impair the ability of authorized users to match and thereby compare and assess data quality and employment and pay trends over time (Federal Committee on Statistical Methdology, 2005).

RECOMMENDATION 4-3: EEOC should use consistent and unique firm and establishment identifiers, facilitating data merges and data checking.

Address Measurement Gaps

As stated in Conclusion 3-1, the use of W-2 Box 1 does not capture employees’ total compensation. W-2 Box 5 has a more inclusive definition of earnings, capturing earnings that are not taxable under federal taxation rules, such as earnings that contribute to medical insurance and retirement accounts. Since Box 5 is already computed by firms, requesting it does not impose new burden on employers. California’s pay-data collection program

has already made the switch to using Box 5. It may be useful to collect both Box 1 and Box 5 for one or two transition years to assess the impact of changing the measures. As for all changes to the instrument, the panel recommends field testing of the changes before implementation.

RECOMMENDATION 3-3: EEOC should collect W-2 Box 5 data to measure total compensation, instead of W-2 Box 1 data.

Pay rate refers to the contractual rate of hourly pay or yearly earnings. Both employer pay-equity stakeholders and recent court hearings have focused on rates of pay (Sempowich vs. Tactile Systems Technology, Inc., 2021b, 2021c). Others have noted that pay differences can arise in performance pay (Castilla, 2008) or bonuses (Elivira and Graham, 2002). Data on the components of earnings (e.g., base pay, overtime, or bonuses) are all potentially of interest to EEOC but could be more reasonably explored after EEOC decides to pursue a charge in more depth. If EEOC decides to convert future pay-data collection to individual-level wage data, as outlined in Recommendation 3-7, collecting data on components of pay might be a viable, low-respondent-burden option.

The panel also noted that W-2 reports are limited to regular employees of the firm, thus omitting the expanded use of independent contractors, who are increasingly used by firms to do crucial work but are not treated as employees under current labor law. Currently, independent contractors are outside of EEOC’s jurisdiction, so this limitation is not a present concern. However, the panel notes a need for EEOC to adapt to the changing economic environment, and if Congress later enlarges EEOC’s jurisdiction, then the corresponding data could be obtained from the Internal Revenue Service form 1099-NEC, Box 1. In presentations to the panel, stakeholders repeatedly noted that EEOC’s definition of establishments as “places of work” no longer matches the actual organization of the workforce in many firms.

For pay bands to be useful in detecting pay disparities, an occupation must appear in at least two pay bands within an establishment. Some job categories appear overwhelmingly, but not exclusively, in one pay band: 47 percent of executives are in the top pay band, 51 percent of sales personnel are in the bottom pay band, 45 percent of laborers/helpers are in the bottom pay band, and 58 percent of service workers are in the bottom pay band (Appendix 6-5). The lack of differentiation is more extreme when the distribution of pay bands within individual establishments is examined. There is a tradeoff here. If an occupation is widely distributed across multiple pay bands, one might raise questions about the predictive value of the occupation. (Higher-paid job categories are expected to appear in even the lowest pay bands, not only through natural variation but because of the classification of part-time and part-year employees into pay bands; thus, it

is better to focus on those pay bands with large percentages of employees.) On the other hand, if everyone within a given occupation is in the same Component 2 pay band, then pay disparities cannot be detected using the data, and EEOC would have to seek detailed data from employers to pursue initial investigations of pay differences. As discussed in Chapter 3, the width of a pay band is approximately 24 log points or about 27 percent, while the raw pay gap by sex is about 22 log points overall, and raw pay gaps for Hispanic and Black workers tend to be somewhat smaller. If EEOC is only focused on the largest pay disparities, then wide pay bands may be less of an issue, but current pay bands will not be useful in many situations.

RECOMMENDATION 3-4: If EEOC continues to collect pay data in bands, narrower pay bands should be adopted, and the number of bands should be expanded for top earners to better capture variation in pay.

As stated in Conclusion 3-4, EEOC’s data collection does not measure race/ethnicity in conformance with federal policy and does not explicitly collect data on employees’ LGBTQIA+ status, leaving employers to report such information as responses in a remarks section of the form. However, the 2000 Equal Opportunity Survey conducted by the Office of Federal Contract Compliance Programs collected pay data using a combined measurement approach that was consistent with the federal race/ethnicity standard (see Chapter 3). In addition, the panel understands that EEOC contracted with NORC in 2018 to evaluate, among other aspects, EEOC’s compliance with the federal standard on measuring race/ethnicity. The panel understands that EEOC’s work is ongoing, but preliminary results suggest the need to further study how to best count persons of more than one race (EEOC, 2021g, 2021h). Federal statistical agencies have developed methods to bridge data collected using prior measures of race/ethnicity to correspond to current federal standards (Ingram et al., 2003). Furthermore, recent guidance from the National Academies of Sciences, Engineering, and Medicine on the measurement of sex, gender identity, and sexual orientation could also be taken into account to improve measurement in the Component 2 instrument (the National Academies of Sciences, Engineering, and Medicine, 2022).

The panel recognizes that collecting new data may impose additional burdens on employers, and that particularly in the case of sexual orientation and gender identity, there are difficult data-collection issues involved—both in terms of collecting accurate data and protecting the privacy of the employees. In the panel’s judgment, no changes should be adopted without first conducting field testing to assess the burden and appropriateness of the new data collection. Nothing in this report should be interpreted as saying that employees should be required to report their sexual orientation or gender identity to their employers.

RECOMMENDATION 3-5: EEOC should update instructions to filers to conform to the federal standard on measuring race/ethnicity. This standard offers solutions for reporting race/ethnicity data in a combined format.

RECOMMENDATION 3-6: EEOC should work with other federal agencies to develop and test ways to measure employees’ sex, gender identity, and sexual orientation in a manner appropriate for EEOC data collections.

BROADEN AND STRENGTHEN DATA COLLECTION AND ANALYSIS

If EEOC is willing to reconsider its current approach to data collections and implement substantial changes to its measures, the Commission may greatly improve the value and quality of its data. These changes are substantial in that they require more than tinkering with the current data-collection process; but such changes may provide greater utility to EEOC. When making changes, it is always advisable to conduct pilot testing; this is especially true for these more substantial changes.

The impact on respondent burden is difficult to determine and depends on the particular changes being made. There is evidence that collecting individual-level data may be less burdensome than asking firms to aggregate their data into unfamiliar occupation and pay-band categories (see Chapter 3), though adding new items to the form is likely to increase burden. It is also possible that the primary burden would appear in the first year when employers set up their data systems to respond, while in later years the survey might be relatively routine with little burden. To further limit employer burden, the frequency of the survey could be reduced, for example conducting it only every two years. However, this approach may make little difference to burden if the primary burden is in the initial setup and could reduce utility if alternating years of data collection will reduce the timeliness of the data. The panel recommends that these issues be examined further by field testing and cognitive interviews with respondents, as well as through internal discussions at EEOC about the timeliness of data for enforcement purposes.

The panel also notes that the Chamber of Commerce Foundation and the T3 Innovation Network have launched the Jobs and Employment Data Exchange (JEDx) initiative, which is working to “develop a public-private approach for collecting and using standards-based jobs and employment data.”² EEOC may wish to coordinate with this initiative both to express its own needs and to help minimize the burden placed upon firms while

___________________

² https://www.uschamberfoundation.org/JEDx

collecting necessary data. Recommendation 3-7 seems consistent with the JEDx initiative because it suggests the use of a measure that firms are already reporting to other federal agencies rather than requiring firms to reconfigure their data into pay bands.

EEOC’s decision to collect data at the aggregate level using broad pay and job categories rather than at the level of the employee is a prime example. Recall that the Sage Computing (2015) report recommended EEOC to use of pay bands rather than individual-level pay data under the dual assumptions that pay bands would reduce reporting burden and address confidentiality concerns. However, the panel finds that pay bands are not the appropriate solution for managing respondent burden or confidentiality in pay-data collections given the availability of established, improved methods of pay-data collection and disclosure avoidance.

While broad categories might seem easier in that they request less detail, such categories require employers to reconfigure their internal data to match external categories. Requesting individual-level data would better fit the data structure of most employers’ internal systems.

Indeed, the capacity of employers to respond to individual-level data requests is now ubiquitous. Employers use automated systems to submit individual-level data to state unemployment systems, state and national tax systems, and large sample surveys. Bureau of Labor Statistics’ Occupational Employment and Wage Statistics survey transitioned from the use of pay bands to individual-level pay-data collection at the initiation of employers and informed by field testing (see Chapter 3). Existing processes at BLS address data-confidentiality concerns.

Table 8-1 gives an example of a possible data-collection instrument. Since the data fields are already available in standard human resource and payroll software, this reporting format is anticipated to reduce employer burden relative to the existing EEO-1 form. It would also increase the precision of pay and occupation measures and the reliability of hours worked. The panel recommends not collecting hours-worked data for Fair Labor Standards Act-exempt employees, although number of weeks worked and full-time/part-time status are necessary to adjust for part-year employment.

In the panel’s judgment, future pay-data collection design decisions should only be made after robust field testing has been conducted, and survey measurement and feasibility (including respondent burden) have been deemed acceptable.

CONCLUSION 3-8: EEOC’s current approach for aggregate pay and hours-worked data severely limits the utility of the data collected, unnecessarily increases employer burden, and complicates the collection of additional key information. Collecting data from employers at the level of individual workers may be less burdensome than the current approach and would markedly increase the utility of pay data.

TABLE 8-1 Example of Possible EEO-1 Pay-Data Collection Form Obtaining Individual-Level Data (Spreadsheet Version)

RECOMMENDATION 3-7: EEOC should develop, test, and (if found acceptable) implement modifications to the Component 2 instrument to collect individual-level employee pay data, which reflects employers’ current reporting practice to state and federal agencies. EEOC’s transition to individual-level pay data should be informed by the Occupational Employment and Wage Survey instrument and protocol. Field testing should estimate respondent burden relative to alternative methods and assess confidentiality protections to be applied.

In the panel’s judgment, to improve accuracy and comparability of data, EEOC should stop using legacy EEOC job categories, which are neither directly comparable to other federal occupational data nor sufficiently detailed for analysis of similarly situated employees. Instead, EEOC should employ federal Standard Occupational Classification system (SOC) categories and follow federal practice in periodically updating those categories as U.S. occupational structure evolves.

As described in Chapter 3, the SOC system classifies all occupations for which work is performed for pay or profit (OMB, 2017). It covers all jobs in the national economy, including occupations in the public, private, and military sectors. The SOC is designed to reflect the current occupational composition of the United States. The SOC has a nested design, moving from aggregate to very detailed occupational coding. The more detailed the occupational coding EEOC deploys, the more its data would approach the employer’s job title distinction and the easier it would be for EEOC to make “apples-to-apples” comparisons in its regulatory efforts.

The SOC supports the efficiency and effectiveness of the federal statistical system by providing a standard for occupation-based statistical data classification and thereby ensuring comparability of these data across federal statistical agencies. Accordingly, all federal agencies that publish occupational data for statistical purposes are required to use the SOC; state and local government agencies should use this national system to promote a common language for categorizing and analyzing occupations (OMB, 2017). Consistent with good statistical practice, these classifications are reviewed and revised periodically to ensure relevance and accuracy. The SOC system is refreshed roughly every five years, with the most recent revision issued in 2018. There are crosswalks that allow for the SOC codes to be connected across time, thus allowing time comparisons.

As described in Chapter 3, if EEOC adopts an individual-level data-collection strategy, the panel advises that the Commission collect job titles rather than EEOC job categories. Firm job titles can be reported with minimal respondent burden. When collected with NAICS codes, automated occupational coding programs (such as that used by Occupational Employment and Wage Statistics [OEWS] and others) result in acceptable rates

of agreement with the SOC. This will improve precision and reliability of occupational coding as well as the utility of the occupation data collected, while also reducing respondent burden. The more detailed the occupational coding the more regulatory decisions will be guided by information on similarly situated groups. In principle, if autocoding of job titles were used for EEOC pay data in the future, Bureau of Labor Statistics OEWS staff could apply their autocoding tool to EEOC data (once stripped of identifiers by EEOC, as appropriate) and return the result to EEOC for remerging (Fincher, personal communication, December 6, 2021).

We recognize that such a move would provide a break in EEOC’s categorization. EEOC may wish to consider a transitional year, in which both job titles and the current job categories are collected, supporting a crosswalk between the two.

RECOMMENDATION 3-8: EEOC should adopt the Standard Occupational Classification system for classifying occupations to provide greater precision for comparisons of similarly situated employees. To limit respondent burden, EEOC should explore established, improved data systems for occupational coding of individual-level job titles, such as those used by the Bureau of Labor Statistics’ Occupational Employment Wage Statistics collection.

EEOC anticipated that employers may not track hours worked for employees exempt from the Fair Labor Standards Act. Therefore, EEOC gave employers the option of reporting either actual or estimated hours worked. Still, collecting data on hours worked has less meaning for exempt employees, and mixing data on exempt and non-exempt employees obscures data on exempt employees. What is needed, however, is the ability to weight exempt employees’ income by whether they were part-time or worked less than a full year. Note that collecting data separately on exempt employees will be simplified if EEOC adopts Recommendation 3-7 and switches to individual-level data.

RECOMMENDATION 3-9: EEOC should distinguish between Fair Labor Standards Act-exempt and non-exempt workers, and between part-time and full-time workers. A measure for the number of weeks worked should be included to account for part-year employment. EEOC should only collect hours worked for non-exempt employees.

EEOC is charged with protecting groups beyond those that can be assessed using data on race/ethnicity and sex. The panel understands that practical and legal constraints exist for data collection on broader groups. While it makes historical sense that EEOC has invested primarily in data

collection on biological sex and race/ethnicity, EEOC’s responsibilities have increased over time to include disability, age, sexual orientation, gender identity, and, in some circumstances, veteran status. Collecting data on additional protected groups will be simplified if EEOC begins collecting individual-level data.

RECOMMENDATION 3-10: EEOC should explore the measurement of pay gaps for additional groups protected under its authority or policy equities, including persons age 40 and older, persons with disabilities, and veterans. To do this robustly while minimizing respondent burden, other federal data collections measuring pay of these groups, such as the American Community Survey, may be instructive.

The federal government requires employment-related data collection from employers via multiple agencies, including the Internal Revenue Service, Bureau of Labor Statistics, and the Census Bureau. These agencies have deep expertise in workplace data collection and analysis. The panel encourages EEOC to learn from and share its expertise with these agencies. In the panel’s judgment, any addition of new items, such as education and job experience, should be based both on the expertise of other agencies and on field testing. Furthermore, the panel considers the addition of these new items to be a lower priority than collecting individual-level wage data, and it would be appropriate to switch to collecting individual-level wage data while continuing to explore whether and how to collect additional individual-level data.

The panel also deems it important, as a general rule, to reduce employer burden in responding to federal data inquiries. As employers must report to multiple federal agencies, coordination in the conceptualization and measurement of items; in the definition of earnings, occupations, sex, and race categories; and in modes of administration are important mechanisms for reducing respondent burden. EEOC participation in the Jobs and Employment Data Exchange project (described above) may be valuable in this regard.

RECOMMENDATION 3-11: EEOC should work with employer groups and federal data-collection agencies to explore ways to collect individual-level data, such as education, job experience, and tenure, which will support detailed pay-disparity analyses and employer self-assessments.

EEOC data are potentially of great value to policymakers and researchers, as long as privacy concerns are met. States such as California and Illinois are making their own efforts to collect similar data. Increased coordination with other federal agencies will, in some instances, become

easier if the Office of Enterprise Data and Analytics (OEDA) or some other subunit within EEOC becomes a recognized statistical unit (OMB, 2007).

One role of a recognized statistical unit would be to advise on the sharing of EEO-1 pay data with employers in a way that informs employer self-assessment while appropriately addressing confidentiality concerns. Sharing anonymized data with employers is one method EEOC could use to carry out its mandate to provide compliance assistance to the regulated community. Sending anonymized peer comparisons to survey respondents as a way of thanking them for their cooperation and publishing benchmarks online could help firms achieve greater diversity without putting both the federal government and the firms through costly and lengthy litigation. However, employer groups advised the panel that employers would need more detailed information (such as education and job experience) for such data to be useful, which stresses the importance of Recommendation 3-11. Of course, data sharing would need to include appropriate protections for privacy.

In addition, a recognized statistical unit could prepare de-identified data files to be shared with other federal agencies to improve efficiencies in frame management (i.e., completeness of the master list), automated occupational coding, and data quality—similar to partnering with the Census Bureau and Bureau of Labor Statistics (BLS) programs. Maintaining an up-to-date business register is a difficult task because new businesses are constantly created, while existing businesses may close or merge with other businesses. BLS and the Census Bureau each have their own business registers, with BLS’s business register largely based on unemployment insurance records, while the Census Bureau’s business register is largely based on IRS records (Becker et al., 2005). A commercial business register is maintained by Dun & Bradstreet, whose database is based largely on credit reporting.³ By contrast, EEOC primarily depends on self-reports, though it sometimes works to update its list. Unlike these other organizations, it lacks automatic access to a source of updated records. Confidentiality restrictions limit the extent to which federal databases can be shared.

Tension currently exists between maintaining confidentiality of data shared among federal agencies and an ongoing effort to increase opportunities for sharing. As a recognized statistical unit, EEOC’s OEDA potentially could send the EEOC frame (or master list) without other data values to federal statistical agencies, such as BLS or the Census Bureau, to refresh with business-register data. The updated frame then could be used by EEOC to contact filers. Data would be collected for EEOC enforcement purposes. When complete, EEOC would provide the dataset to OEDA, which could prepare a de-identified dataset for public use. This dataset could be used by employers for self-assessments and by other stakeholders for their research. Data.gov is an obvious destination for such data products. Cooperative

___________________

³ https://www.dnb.com/duns-number.html

aggregate data sharing with BLS, the Census Bureau, and other federal statistical agencies might be considered as well.

RECOMMENDATION 8-2: EEOC should strengthen consultation and data sharing with the public, and with federal and state employment data-collection agencies. To do so, EEOC could consider joining the Federal Committee on Statistical Methodology’s Committee on Data Access and Confidentiality to discuss modern methods to improve data access while protecting against disclosure. EEOC could consider designating its Office of Enterprise Data and Analytics as a federal statistical unit to collect, report, and protect data in anonymized format for research purposes (including employer self-assessment), while targeted investigations for enforcement purposes proceed as a separate data activity.

POLICY CASE FOR IMPLEMENTING CHANGE

EEO-1 pay data could be an essential resource to advance EEOC’s mission (Box 8-2). If well-designed, these data could assist in identifying, tracking, investigating, resolving,⁴ and, when necessary, prosecuting allegations of discrimination. Collection and analysis of appropriate data could support EEOC statistical reports, trend identification, and analyses of sex and race/ethnicity disparities at the workplace, firm, industry, and national levels. In addition to supporting EEOC’s regulatory efforts, collection of pay data could encourage firms to increase internal pay-equity analyses.

___________________

⁴ This text was changed after release of the pre-publication version of the report to clarify the ways in which EEO-1 data could be used by EEOC to address allegations of discrimination.

The panel finds that the EEO-1 pay-data collection, as currently designed, is not well-suited to measure pay equity by EEOC or by employers. However, the panel believes these issues could be addressed by adopting well-established, improved data-collection methods currently in use by federal agencies. In addition to the panel’s specific recommendations, the panel encourages EEOC to adopt a business model of continuous assessment of its EEO-1 pay-data collection to take into account emerging data needs and efficiencies.

Tables 8-2 through 8-4 align the panel’s recommendations with EEOC’s mission. Most important are the columns that note a policy mechanism for each suggested recommendation. If the panel’s recommendations are implemented, existing Component 2 data and future data collections could be used with greater confidence and thus further support EEOC’s mission and vision.

Enforce Pay Equity in the Workplace

Many of the panel’s recommendations focus on improving the coverage of employers included in the EEO-1 frame, the design of the survey instruments, and the collection operations (Table 8-2). The panel also provided recommendations regarding ways to protect employer privacy and support employer self-assessment through increased access to (anonymized, aggregated) EEO-1 pay data. The statistical and survey methods needed to accomplish these recommendations are well-established among federal data collections. The panel encourages EEOC’s Office of Field Programs and Office of Enterprise Data and Analytics to engage with federal agency data and statistical science experts to explore specific solutions best suited to EEOC’s context.

Account for a Changing Society

Many of the panel’s recommendations focus on improving measurement to account for the changing workforce.

The nature of work and judicial and legislative understandings of equity have changed since the EEO-1 form was first collected in 1966. The EEO-1 form has not been sufficiently updated to reflect these changes. The measurement of work as captured by EEO-1 job categories is not coordinated with the occupation classifications in use and regularly updated by other federal agencies. The EEO-1 measurement of race/ethnicity is not coordinated with federal standards that allow capture of greater diversity. The importance of protecting pay equity on the basis of sex—to include not only biological sex but also gender identity and sexual orientation—is now established. Yet, the EEO-1 measurement does not reflect this. The

TABLE 8-2 Recommendations to Enforce Pay Equity in the Workplace

panel also noted other groups protected by EEOC authorities that are not currently measured in the EEO-1 form.

The panel made several recommendations for improved measurement (Table 8-3). As important as these recommendations are, they are not intended to be static, one-time fixes. The relationship of work to place of work (such as outsourced contract employment), and location of work (such as the physical establishment not being the actual work location) are a few examples of situations in which new measures and methods may need to be considered in the future. To be useful for policy making, measures need to reflect current society. Therefore, the panel advises EEOC to adopt a forward-looking approach to anticipate, test, and thereby provide the most relevant measurement of pay, work, and protected groups.

Good Government

Several of the panel’s recommendations, if implemented, are anticipated to reduce respondent burden and cost, and add value to stakeholders. (Table 8-4). Improvements such as data-upload, standardization of unique identifiers, automatic data quality checks, autocoding of occupations, and, potentially, individual-level pay-data collection represent efforts to pursue “good government.” With additional refinements to Component 2 data based on the panel’s recommendations, these data will impose a lower administrative burden and provide employees, non-profit organizations, and the public with better-quality data for use in planning and enforcement decisions.

If the panel’s recommendations are implemented, EEOC analysts, researchers, and consultants will be able to conduct higher-quality studies that assess the impact of charges and lawsuits on pay disparities. Most important, the public’s trust in EEOC and the Commission’s ability to ensure fair workplaces will be enhanced.

TABLE 8-3 Recommendations to Account for a Changing Society

TABLE 8-4 Recommendations to Use Good Government and Statistical Practices