Over the past 50 years, increases in women’s participation in the labor force and immigration have resulted in more women, Hispanics, and Asian Americans in the workforce. New data below from a forthcoming EPI report show that the working class is more diverse than ever before—no matter how you define it.

When defined by education, the working class is nearly half workers of color

The traditional definition of the working class comprises workers with less than a bachelor’s degree. Under this definition, in 1979, the working class was over 80% white. Today, workers of color collectively account for nearly half of the working class.

Working class by race and ethnicity

• 52% white
• 26.8% Hispanic
• 15.1% Black
• 4.7% AAPI
• 1.4% Native American

Working class by gender

• 56.4% men
• 43.6% women

When defined by income, the working class is similarly diverse

The working class could reasonably include more college graduates and traditional “white-collar workers” than were previously considered. That’s because the wage boost from getting a college degree is lower than it used to be, and underemployment—along with racial and gender pay disparities—can mean the economic status of graduates of color and women is more like some nongraduates.

By this broader measure, the working class can be considered the bottom 75% of U.S. households, regardless of education level or job type. This works out to a family income of below $150,000 per year.

Under this definition, the working class remains similarly diverse in terms of race and ethnicity. While men are still a majority of the working class, a larger share of women fit within this definition of working class, given lower pay in jobs predominantly held by college-educated women.

Working class by race and ethnicity

• 53.1% white
• 24% Hispanic
• 15.2% Black
• 6.5% AAPI
• 1.2% Native American

Working class by gender

• 52.5% men
• 47.5% women

However you define it, today’s working class is diverse by race, ethnicity, and gender. Improving economic conditions for today’s working class requires policies that meet the needs of all these workers.  

Source: Author’s analysis of pooled 2024–2025 Current Population Survey basic monthly microdata from EPI Microdata Extracts, Version 2026.8.13, https://microdata.epi.org.

This is the first brief in a series exploring comprehensive federal policy recommendations around AI and workers.

Introduction

Amid the seemingly endless predictions about AI, one point is essential to remember: How this plays out is not predetermined. Policy choices determine whether workers are helped or harmed by technological change. Even without AI, many workers already face harmful disruptions. Every year, 20 million workers are laid off in the U.S. economy1—due to employer mismanagement, once-profitable products becoming less so, offshoring, automation, and other factors. For individual workers, losing a job for any reason and for any period of time is traumatic and damaging in large part because we lack adequate social insurance programs to smooth those transitions.

As lawmakers debate how to best regulate AI deployment in our workplaces, today’s priority should be to strengthen the safety net for workers in any scenario and to acknowledge how AI is already affecting workers’ rights, protections, and livelihoods. Decades of underinvestment, coupled with attacks on unions and other labor supports, have left workers without an adequate safety net.

We can improve our current systems to better provide livable benefits for workers who lose their jobs because of large-scale economic shocks that are not their fault and outside their control. We can also learn from other periods of workforce disruption—the COVID-19 pandemic, trade-related shocks, and the 2008 recession—about which policy choices were effective and which fell short.

When workers lose their jobs due to factors outside their control, they are considered “displaced” or “dislocated.” If they meet certain requirements, these workers are eligible for unemployment insurance (UI). To strengthen outcomes for workers, this brief will provide recommendations to:

• increase UI benefit amount and duration
• expand and standardize eligibility to receive benefits
• align employer incentives with worker access to benefits
• center worker needs and power
• strengthen data infrastructure to support workers

These recommendations should be considered amid conversations focused on creating a stand-alone unemployment program geared toward AI-driven job loss. These discussions should weigh considerations and potential pitfalls, including:

• a blank slate for program design
• relevant roles for key stakeholders
• eligibility criteria challenges
• the allocation of resources, time, and awareness to be effective
• reauthorization vulnerability

If policymakers are serious about easing the impact of AI-related job transitions for workers, the priority solution should be strengthening already existing social insurance programs to reduce the economic consequences of losing any job and to provide adequate support to workers while they seek other employment.

Changes to the current system or establishing a new program will require robust resources, including time, money, and expertise. This investment can help all workers, regardless of their risk of AI-based displacement. With a strong safety net, workers who lose their jobs can get the assistance they need, including the opportunity to find work that is well suited to their skills and experiences, and businesses are more inclined to improve their practices because they know workers aren’t trapped in their current jobs by economic necessity.2 This moment presents an opportunity to strengthen the labor market for everyone and to ensure workers have the safety net they deserve.

How to strengthen our current system

Any effective approach to fixing the UI system must be comprehensive and should apply regardless of whether a worker is displaced due to AI. Numerous experts have weighed in on the best approach—including the National Academy of Social Insurance (NASI) Unemployment Insurance Taskforce,3 the National Employment Law Project (NELP),4,5 and The Century Foundation,6 among others. This brief draws from those recommendations to strengthen outcomes for workers. Other important issues, such as modernizing UI system technology and infrastructure, fraud detection and prevention, and the use of AI in benefit determinations and delivery, also shape worker outcomes but fall outside this brief’s scope.

Increase benefit amount and duration

First and foremost, UI should be designed to ensure that workers can afford housing, food, and other essentials during their period of joblessness. The percentage of wages that UI replaces for workers has declined over time and has not kept pace with inflation.7 In 2025, national average UI benefits replaced well under half of national average wages.8 This gap leaves workers across the income spectrum unable to maintain their standard of living and can push those whose earnings were close to the minimum wage into poverty.

Ensure benefit amounts are consistent across states

Benefit amounts vary across states. In Arizona and Florida, UI benefits replace less than 15% of average weekly wages.9 In some states, the weekly benefit amount is just $222 per week.10 While state variation in benefit amounts hurts all workers in stingy states, it weighs especially heavily on Black and Hispanic workers.

One recent study finds that differences in state rules cause the wage replacement rate for Black workers to be meaningfully lower than the rate for white workers, and that the discrepancies harm overall well-being.11 These disparities compound. Even among those likely eligible, Black workers are less likely than white workers to receive UI benefits—a gap explained primarily by structural factors, not a difference in eligibility, that has held steady for three decades.12

Wage replacement rates should be consistent across states. A recent federal bill has proposed changes to increase the replacement rate to 75% and provide additional weekly funding for dependents.13 Given growing wage inequality, the rate also should vary by income level. In one proposal, the lowest earners would receive benefits that represent 85% of their wages.14

Extend the duration of benefits across states

The duration of benefits also varies greatly from state to state. Until 15 years ago, eligible workers across the country had access to 26 weeks of benefits, a generally agreed-upon standard.15 After the Great Recession however, federal law barred states from cutting benefit amounts, so many states instead cut the duration of unemployment benefits to stabilize their trust funds and lower tax rates for employers.16 Some states now only provide 12 weeks of benefits, a short bridge to find a new job well-matched to workers’ skills and experiences.17 A minimum of at least 26 weeks of benefits should be a uniform requirement across states.

Workers cannot currently depend on the Extended Benefits program, which was intended to provide support during economic downturns by providing additional weeks of benefits during periods of high and rising unemployment. The triggers for Extended Benefits have not been effective, leaving workers to wonder if they are available. In their absence, workers must rely on temporary programs funded by Congress, which depend on political priorities.18 More reliable triggers, higher wage replacement levels, and a longer duration of benefits would make the Extended Benefits program more effective.19

Expand and standardize eligibility for benefits

Only certain workers are eligible for UI, leaving others stranded without benefits if they lose their jobs. Eligibility criteria and the base period used to calculate benefits are set by state law, rather than federal statute. This state-by-state variation can make it difficult for workers to know whether they qualify for assistance and plan accordingly.

Generally, the group of workers ineligible for benefits includes independent contractors, job seekers who have not been recently employed, and those who have exhausted their benefits. Some employers have incorrectly classified many workers as independent contractors, and this misclassification prevents those workers from requesting access to the UI benefits to which they would otherwise be legally entitled.20

UI should cover a broader range of workers, including part-time workers, independent contractors, job seekers who have not been recently employed, and seasonal and temporary workers. During the COVID-19 pandemic, Congress tasked states with implementing new UI programs for gig and platform workers who fell outside of the usual verification system, creating significant challenges.21 Barring changes to account for the unique situation of these workers, these same issues will arise again.22 By incorporating these workers into the UI system now, instead of waiting for a crisis, future challenges can be mitigated.

Policymakers should also consider stronger presumptive eligibility standards, under which workers are presumed eligible for benefits and paid promptly, with the burden placed on employers or the state to prove otherwise. Lowering the barriers for workers to receive benefits could meaningfully increase uptake among workers who currently go without benefits.

Align employer incentives with worker access to benefits

All eligible workers should be able to access UI benefits if they want them. State UI programs are generally financed through employer experience rating: Employers pay more taxes when they lay off more workers who are eligible for UI.23 Therefore, the system’s setup incentivizes employers to limit their tax liability by restricting eligibility for the program in different ways. This can include not hiring certain workers as employees, not proactively informing workers about their eligibility, attempting to reclassify workers as independent contractors, and other forms of suppressing UI program access.

A Bureau of Labor Statistics survey found that only about a quarter of workers who were unemployed in 2022 applied for benefits.24 Further, a recent survey found that 19% of recently unemployed workers encountered actions from an employer that were meant to deter them from applying, in addition to a quarter of all UI filers who said that their employer contested their benefits.25 Low-income workers more frequently reported employer contestations , and workers who reported contestations of their claims by employers were less likely to receive benefits and to experience greater material and psychological burdens during the filing process.26 There is a real opportunity to better align employer incentives to support workers in accessing UI benefits.

Presumptive eligibility, discussed above, would address some of these access barriers by shifting the burden of proof away from workers, but it would not resolve the underlying incentive employers have to minimize their tax liability. There have been many proposals for how to address this issue: broadening the tax base, payroll contributions from employees to UI, or increasing state taxes to fund programs.27 Most basically, employers could pay taxes based on the total number of workers they lay off, not just those eligible for UI.28 These questions are tied to broader considerations about how to finance a more effective program.

Center on worker needs and power

The UI system should be structured around ensuring that workers have the support they need as they seek reemployment. This requires a real investment in active labor market policy, including communicating with workers and unions about their needs in order to align training and placement. Congress should allocate funds to revitalize the Employment Service in the U.S. Department of Labor, which once played a key role in supporting reemployment.29 For example, in a 1986 New Jersey pilot program, job search assistance measurably helped dislocated workers return to work sooner and increased their earnings.30 A well-resourced and staffed U.S. Employment Service can leverage state labor market information—frequently used by employers—to help job seekers find opportunities aligned with their skill sets.31

Worker organizations can also play an important role in the delivery of benefits, including facilitating training and connections to job search assistance.32 In 2022, unemployed people who were members of a union were twice as likely to apply for UI benefits, and among applicants, 30 percentage points more likely to receive those benefits.33 This sizable gap suggests that when workers have organizational support navigating the UI system, they are far more likely to receive the benefits they are entitled to, reinforcing the case for involving worker organizations directly in UI outreach, filing assistance, and perhaps even delivery. Furthermore, unions routinely negotiate with employers, both in collective bargaining agreements and when bargaining before layoffs, to secure agreements that employers will not contest workers’ applications for UI benefits.

In some Northern European countries, unions are responsible for administering unemployment benefits, and scholars have found that this system leads to stronger unions and worker voice in the economy.34 Policymakers in the United States should consider how unions and other worker organizations could deliver unemployment benefits to workers.

Strengthen data infrastructure to improve outcomes

State workforce agencies collect quarterly wage records from employers to facilitate UI administration to gauge eligibility for benefits as well as benefit levels. States share wage records with the federal Census Bureau, and these data are included in longitudinal data sets that can follow workers over time. These data are used for both research and the administration of other policies (such as facilitating child support payments or tracking the performance of job training programs). Wage records provide an important but limited set of information that varies by state. A growing number of states have expanded their wage records to include additional information such as job titles, total hours worked, and job location.35

State-level data are a valuable source of information about what is happening to workers, but the lack of standardization across states makes it difficult to gain a clearer picture of employment. To remedy this, lawmakers should create standardized requirements for enhanced wage records to analyze who is losing their job and where workers are finding new opportunities.

The enhanced information coming from states would be especially useful with the advent of AI since no single federal data source can reliably track how AI adoption is reshaping demand for specific occupations in real time or identify when firms are reducing jobs through attrition. Further, workers would benefit from better information about opportunities that align with their skill set.36 Better integration with job posting data—including the National Labor Exchange—and reemployment systems would help workers navigate the job-search process.

Considerations for a stand-alone program

The UI system should be improved regardless of how AI affects employment and job opportunities. Additionally, some policymakers have suggested creating a stand-alone program geared toward AI-driven job loss. There is precedent for this approach. Congress has acted nine times since 1958 to create temporary, additional UI benefits, most recently during the COVID-19 pandemic.37 There is much to be learned from successes and challenges of these previous programs.

Some of these programs also included benefits or services beyond unemployment insurance for workers. For example, Trade Adjustment Assistance has, at various points, included relocation assistance, training programs, a health coverage tax credit, and wage insurance for older workers who took a lower-paying job.38 A new stand-alone program could similarly go beyond cash benefits, offering workers support to develop new skills, cover the costs of a transition, or find new opportunities.

Blank slate for program design

A new program can address the key structural issues with the current system discussed in the previous section. Any new program should be uniform across states so that benefit levels, duration, and eligibility are consistent for every worker regardless of where they live. A new program also presents the opportunity to build in the other improvements outlined above, including benefit levels set to a livable amount, expanded eligibility, and stronger data infrastructure. A new program could also adopt more fundamental changes to the claims process, such as the presumptive eligibility standards that Michigan moved toward during the COVID-19 pandemic.39

As discussed in the worker power section, a new program should be explicitly structured around building worker power, for example, by involving worker organizations in program administration or benefit delivery. Further, a new program can focus on effective job placement by leveraging relationships between employers, unions, and other worker organizations.

Finally, a new program should also be designed to avoid UI’s employer-financing structure, which—as discussed—in most states, disincentivizes employers from making sure workers get the benefits they are due. Left unaddressed, that same structure would allow employers to game or underreport AI-related job loss in a new program.

Relevant roles for key stakeholders

A successful program requires effective coordination among stakeholders. As with UI and previous stand-alone programs, states would likely retain a central role in delivering benefits and services, given their existing infrastructure and more direct knowledge of local labor market conditions. Local agencies can provide labor market information to workers, worker organizations, and employers to facilitate job placement. Any program should carefully consider how to balance flexibility for states to tailor the program to meet local needs, while maintaining strong, uniform standards across states.

As discussed, worker organizations and unions can play a central role in program administration. Worker organizations can support workers in accessing relevant benefits and training opportunities, ensuring that worker needs and outcomes are central to programmatic strategy and implementation.

Employers bear significant responsibility for the program’s success. They are often the first to know a layoff is coming and should proactively connect workers to available services, for example, by supporting on-site services when requested by state or local Rapid Response teams. Employers can proactively share specific information about their talent and skills needs with labor agencies and worker organizations, and hire workers during or after relevant training.

Eligibility criteria challenges

Creating clear and consistent eligibility criteria will be a significant challenge in establishing a stand-alone program for workers affected by AI-related job displacement. For instance, a program to compensate Department of Energy workers harmed by nuclear materials has faced significant challenges determining eligibility, including the complexity of reconstructing historical exposure to nuclear materials to establish causation.40 Challenges in establishing causation have also been an enormous barrier to worker relief through the Trade Adjustment Assistance program.41

In another example, some pandemic unemployment assistance benefits are still hung up over eligibility disputes years later.42 And from the start of the pandemic, there have been hotly contested debates about who should receive benefits.43 Moreover, concerns about fraud can also complicate eligibility determinations. In the wake of fraud during the COVID-19 pandemic, many states adopted stricter verification requirements to confirm eligibility.44 A new program focused on AI-driven job loss would need to balance similar integrity concerns against the risk of preventing eligible workers from receiving the benefits they are due.45

It is virtually impossible to systematically link job loss with AI adoption. In many instances, AI’s effects on a worker or industry may be indirect. For example, workers may lose their jobs in part due to AI adoption months or years earlier or because of broader economic disruption caused by AI’s impact on different industries. Even setting aside these implementation challenges, restricting eligibility to AI-driven job loss raises a separate concern: fairness. AI is not the only source of economic disruption. Global trade (including offshoring and outsourcing) and the green energy transition continue to displace workers across industries.

Other workers should be entitled to the same benefits as those directly displaced by AI. Drawing arbitrary boundaries around a single cause of job loss will create inequities, a risk compounded if states must independently interpret what counts as AI-driven job loss, therefore creating inconsistent standards nationwide.

Resources, time, and awareness to be effective

Standing up a new program will take significant time and resources. In addition to providing benefits, this includes making meaningful investment in technological systems and program administration, a particular challenge in light of years of underinvestment in modernizing UI systems.46 Federal and state governments need adequate resources for user experience testing, efficient application determinations, coordination with relevant partners, data collection and analysis, and programmatic evaluation. The challenges of implementing relief programs during COVID-19 are instructive. Slow initial payments, overwhelming caseloads that strained public servants, technological breakdowns, and significant confusion about eligibility requirements demonstrated the need to proactively develop systems that can respond more nimbly in times of crisis.47

Starting a new program also requires sustained public outreach, as new and stand-alone programs have historically struggled because of a lack of public familiarity. For example, over half of workers did not know they had access to emergency sick leave during COVID-19.48 In another example, when the Reagan administration made changes to the Trade Adjustment Assistance program, applications plummeted, perhaps because many workers believed the program had been abolished.49 Further, most workers did not know that other services including training, job counseling, and relocation assistance were available through the program.50

A stand-alone program could also come at the expense of the existing UI system, which the vast majority of states say is underfunded.51 From 1999 to 2019, federal funding for UI administration declined by 30% on an inflation-adjusted basis.52 If policymakers decide to divert resources that would otherwise go to the existing UI system, the system’s further deterioration could harm workers served by that system in both the short and long term.

Reauthorization vulnerability in Congress

If an AI-specific UI program requires congressional reauthorization, it will be vulnerable to the difficulty of coordinating bipartisan action in Congress. Several previous programs have been subject to a sunset date, requiring periodic reauthorization to continue accepting new claims. Sunsets and other reauthorization requirements leave workers vulnerable to congressional inaction, especially if the impacts of AI on the workforce play out over an extended time period.

For an example of a program caught up in this type of reauthorization limbo, Trade Adjustment Assistance lapsed multiple times, including most recently in 2022, leaving additional workers without access to the program. Another example is The Emergency Unemployment Compensation program, which was created in 2008 and lapsed five times between 2010 and 2013, before expiring for good at the end of that year.53 This immediately cut off benefits for 1.3 million workers and for more than 3.6 million workers by the following September.54

Financing and the path forward

The current worker safety net has failed to expand enough to cover the basic realities of today’s labor market, and inadequate benefit levels, incomplete coverage, and diminished worker representation are persistent problems. Deployment of AI and related technologies in the workplace adds urgency to addressing these issues.

Implementing the recommendations in this issue brief requires a significant investment that should be considered, along with program parameters and design. As AI drives up stock prices and generates wealth for a select few, policymakers are considering how the public should share in these gains. Ideas include increasing taxes for wealthy individuals, closing the gap between labor and capital tax rates, taxing AI use, and taxing data center development, with the proceeds from these new taxes financing programs that redistribute income more broadly across the economy. Considerations for potential taxation approaches will be discussed in a forthcoming brief.

This moment requires big thinking and corresponding action. While AI drives changes to the economy, improvements to the worker safety net are long overdue. We can learn from the past to improve the labor market for everyone.

Acknowledgments

Angela Hanks, Alex Hertel-Fernandez, Ben Zipperer, Gillian Gutierrez, Jacob Leibenluft, Janelle Jones, Josh Bivens, Michele Evermore, Patrick Oakford, Raj Nayak, and Stephen Nuñez.

Notes

1. Bureau of Labor Statistics, “Layoffs and Discharges: Total Nonfarm” [JTSLDL]. Accessed from FRED, Federal Reserve Bank of St. Louis, October 2026.

2. Patrick Oakford, Josh Bivens, and Celine McNicholas, Federal AI Legislation: An Evaluation of Existing Proposals and a Road Map Forward, Economic Policy Institute, September 2024.

3. National Academy of Social Insurance, Unemployment Insurance Task Force, Unemployment Insurance Task Force Final Report, July 2024.

4. Amy Traub, Boosting Economic Resilience: Seven Ways States Can Support More Workers with Unemployment Insurance, National Employment Law Project, July 2025.

5. Michele Evermore, AI Is New. The Challenge of Mass Unemployment Isn’t. National Employment Law Project, September 2026.

6. Michele Evermore and David Balducchi, A Roadmap for Unemployment Insurance Reform, The Century Foundation, October 2023.

7. National Employment Law Project, Policy Advocacy Brief: Benefit Amounts, November 2023.

8. U.S. Department of Labor, Employment and Training Administration, “UI Replacement Rates Report,” Benefit Accuracy Measurement data. Public data prepared by the OUI Division of Performance Management, Accessed August 7, 2026.

9. Megan Rivera, Why Unemployment Insurance Alone Is Not Enough in the AI Era, Washington Center for Equitable Growth, July 2026.

10. Julie Su, Andrew Stettner, and Michele Evermore, Our Unemployment System Needs Modernizing. Trump Is Doing the Opposite. The Century Foundation, September 2025.

11. Daphné Skandalis, Ioana Marinescu, and Maxim Massenkoff, Racial Inequality in the U.S. Unemployment Insurance System, Washington Center for Equitable Growth Working Paper, July 2023.

12. Elira Kuka and Bryan A. Stuart, Racial Inequality in Unemployment Insurance Receipt and Take-Up, National Bureau of Economic Research Working Paper no. 29595, December 2021.

13. Office of U.S. Senator Michael Bennet, “Bennet, Wyden, Beyer Reintroduce Bicameral Bill to Overhaul Unemployment Insurance” (press release), July 16, 2025.

14. Josh Bivens, Melissa Boteach, Rachel Deutsch, Francisco Díez, Rebecca Dixon, Brian Galle, Alix Gould-Werth, Nicole Marquez, Lily Roberts, Heidi Shierholz, William Spriggs, and Andrew Stettner, Reforming Unemployment Insurance: Stabilizing a System in Crisis and Laying the Foundation for Equity, a joint report of the Center for American Progress, Center for Popular Democracy, Economic Policy Institute, Groundwork Collaborative, National Employment Law Project, National Women’s Law Center, and Washington Center for Equitable Growth, June 2021.

15. National Academy of Social Insurance, Unemployment Insurance Task Force, Unemployment Insurance Task Force Final Report, July 2024.

16. National Academy of Social Insurance, Unemployment Insurance Task Force, Unemployment Insurance Task Force Final Report, July 2024.

17. Julie Su, Andrew Stettner, and Michele Evermore, Our Unemployment System Needs Modernizing. Trump Is Doing the Opposite. The Century Foundation, September 2025.

18. Michele Evermore, “There’s a Growing Bipartisan Consensus to Fix Unemployment Insurance” (commentary), The Century Foundation website, July 17, 2024.

19. Michele Evermore, “There’s a Growing Bipartisan Consensus to Fix Unemployment Insurance” (commentary), The Century Foundation website, July 17, 2024.

20. Ismael Cid-Martinez, Nina Mast, Margaret Poydock, and Valerie Wilson, Misclassifying Workers as Independent Contractors Is Costly for Workers and Social Insurance Systems, Economic Policy Institute, April 2026.

21. Julie Su, Andrew Stettner, and Michele Evermore, Our Unemployment System Needs Modernizing. Trump Is Doing the Opposite. The Century Foundation, September 2025.

22. Adam Grundy and Lucie Alleyne, “Nonemployer Businesses Are Key Contributors to Gig Activities and Economic Growth Nationally” (blog post), U.S. Census Bureau website, July 1, 2025.

23. Michele Evermore, “There’s a Growing Bipartisan Consensus to Fix Unemployment Insurance” (commentary), The Century Foundation website, July 17, 2024

24. Bureau of Labor Statistics, U.S. Department of Labor, “Most Unemployed People in 2022 Did Not Apply for Unemployment Insurance Benefits” (blog post), The Economics Daily (TED), April 14, 2023.

25. Amy Traub, Alexander Hertel-Fernandez, and Sanjay Pinto, The Unemployed Worker Study, National Employment Law Project, April 2025.

26. Alexander Hertel-Fernandez and Alix Gould-Werth, “Employer Contestations and Access to Unemployment Insurance: Implications for Benefit Receipt, Material Hardship, and Administrative Burden,” Social Service Review 100, no. 1 (March 2026): 38–68.

27. Michele Evermore and David Balducchi, A Roadmap for Unemployment Insurance Reform, The Century Foundation, October 2023.

28. Josh Bivens, Melissa Boteach, Rachel Deutsch, Francisco Díez, Rebecca Dixon, Brian Galle, Alix Gould-Werth, Nicole Marquez, Lily Roberts, Heidi Shierholz, William Spriggs, and Andrew Stettner, Reforming Unemployment Insurance: Stabilizing a System in Crisis and Laying the Foundation for Equity, a joint report of the Center for American Progress, Center for Popular Democracy, Economic Policy Institute, Groundwork Collaborative, National Employment Law Project, National Women’s Law Center, and Washington Center for Equitable Growth, June 2021.

29. Michele Evermore, AI Is New. The Challenge of Mass Unemployment Isn’t. National Employment Law Project, September 2026.

30. Stephen A. Wandner, Solving the Reemployment Puzzle: From Research to Policy, W.E. Upjohn Institute for Employment Research, January 2010.

31. Michele Evermore, AI Is New. The Challenge of Mass Unemployment Isn’t. National Employment Law Project, September 2026.

32. Alexander Hertel-Fernandez, “Solidarity Insurance: Reimagining Unemployment Insurance as a Platform for Worker Power.” (policy brief), Political Economy Research Institute, University of Massachusetts Amherst, July 2026.

33. Bureau of Labor Statistics, U.S. Department of Labor, “Characteristics of Unemployment Insurance Applicants and Benefit Recipients—2022” (economic news release), March 29, 2023.

34. Bruce Western, Between Class and Market: Postwar Unionization in the Capitalist Democracies (Princeton, N.J.: Princeton Univ. Press, 1997). 

35. Michele Evermore and Will Raderman, The Case for Enhanced Wage Records: Lessons from States, National Academy of Social Insurance and The Niskanen Center, January 2026.

36. Michele Evermore, “There’s a Growing Bipartisan Consensus to Fix Unemployment Insurance” (commentary), The Century Foundation website, July 17, 2024.

37. Katelin P. Isaacs and Julie M. Whittaker, “The Fundamentals of Unemployment Compensation,” Congressional Research Service, IF10336, updated January 12, 2026.

38. Jacob Leibenluft, Lessons from Past Trade Adjustment Policies to Support Displaced Workers in the Era of Artificial Intelligence, Washington Center for Equitable Growth, March 2026.

39. Michele Evermore, “Michigan Unemployment Claims Processing,” testimony on behalf of National Employment Law Project to a hearing before the Michigan Legislature, Joint Select Committee on the COVID-19 Pandemic, May 21, 2020.

40. U.S. Government Accountability Office, Energy Employees Compensation: DOL Generally Followed Its Procedures to Process Claims but Could Strengthen Some Internal Controls, GAO-16-74, March 2016.

41. Jacob Leibenluft, Lessons from Past Trade Adjustment Policies to Support Displaced Workers in the Era of Artificial Intelligence, Washington Center for Equitable Growth, March 2026.

42. Julie Kay, “‘No Accountability’: Pandemic-Era Unemployment Benefit Delays Persist, 5 Years Later,” WCNC Charlotte, April 3, 2025.

43. Annelies Goger, Tracy Hadden Loh, and Nicole Bateman, Debunking Myths About COVID-19 Relief’s ‘Unemployment Insurance on Steroids,’ The Brookings Institution, May 2020.

44. National Employment Law Project, ID Verification, November 14, 2023.

45. National Academy of Social Insurance, “Testimony of Michele Evermore, Senior Fellow, National Academy of Social Insurance,” testimony before the U.S. House of Representatives, Committee on Ways and Means, Work and Welfare Subcommittee, March 5, 2026. 

46. Julie Su, Andrew Stettner, and Michele Evermore, Our Unemployment System Needs Modernizing. Trump Is Doing the Opposite. The Century Foundation, September 2025.

47. Michele Evermore, AI Is New. The Challenge of Mass Unemployment Isn’t. National Employment Law Project, September 2026.

48. Emma Jelliffe, Paul Pangburn, Stefan Pichler, and Nicolas R. Ziebarth, “Awareness and Use of (Emergency) Sick Leave: US Employees’ Unaddressed Sick Leave Needs in a Global Pandemic,” Proceedings of the National Academy of Sciences 118, no. 29 (July 20, 2021): e2107670118.

49. U.S. Congress, Office of Technology Assessment, Trade Adjustment Assistance: New Ideas for an Old Program, 1987.

50. U.S. Congress, Office of Technology Assessment, Trade Adjustment Assistance: New Ideas for an Old Program, 1987.

51. Bipartisan Policy Center, “Administrative Failures Plague State Unemployment Insurance Programs,” July 1, 2020.

52. Bipartisan Policy Center, “Administrative Failures Plague State Unemployment Insurance Programs,” July 1, 2020.

53. Katelin P. Isaacs and Julie M. Whittaker, Emergency Unemployment Compensation (EUC08): Status of Benefits Prior to Expiration, Library of Congress, Congressional Research Service, CRS Product Number R42444, August 11, 2014.

54. U.S. House of Representatives, Ways and Means Committee Democrats, Unemployment Insurance (fact sheet), 2014.

Federal statistical agencies (FSAs) produce the gold standard economic data that employers, investors, job seekers, workers, and policymakers rely on to assess the health of the U.S. economy. Today, FSAs face historically unprecedented threats to their capacity and even their independence. This raises the specter of a future where FSA data cannot be relied upon to honestly report whether the U.S. economy is experiencing dysfunction.

This dashboard was initially created to display a range of data not collected or disseminated by FSAs to shed some light on the economy during the pause in federal data collection during the 2025 government shutdown. Now—perhaps even more importantly—it serves to provide an accountability check against efforts to manipulate FSA data in the future.

This set of “next-best” data sources is clearly inferior to the datasets that have historically been collected and analyzed by the nonpartisan, expert professionals who staff FSAs. Among many other relative weaknesses, these next-best data offer no insights on how the economy is affecting U.S. households differently by race, gender, or ethnicity.

Indeed job postings vs. BLS JOLTS

       

Indeed job postings correlate strongly with total job openings from the Bureau of Labor Statistics’ Job Openings and Labor Turnover Survey (JOLTS). JOLTS job openings tend to fall in recessions and rise in expansionary periods.

The Indeed Hiring Lab, the research arm of job search website Indeed.com, publishes a job postings data index that aggregates a complete count of all job postings on the Indeed website. This includes a measure of new job postings, which only counts job postings the first time they are visible. The data—available starting in February 2020—are collected daily and reported as a seven-day average. For this comparison, we take monthly averages of the seasonally adjusted daily data.

In the figure, we show that Indeed new job postings track closely with job openings from JOLTS. The correlation between these measures between February 2020 and August 2025 is quite high (0.95).

While Indeed data are only available for the last five years, JOLTS data go back to 2000 and show a clear relationship between job openings and business cycles. Job openings tend to fall in recessions and rise in expansionary periods. That relationship is clearest in the 2001 and 2007 recessions and expansions, where a fall in job openings preceded the start of the recession.

In the wake of the government shutdown, the Indeed new job postings can give use some information on the state of new positions needed by employers. Moving forward, a strong and sustained divergence of trends between the Indeed and BLS measures could provide a worrying signal of degraded data quality or integrity in BLS reports.

ADP employment vs. BLS private-sector payrolls

       

Monthly changes in ADP employment levels tend to track changes in private-sector payroll employment from the BLS Current Employment Statistics, and both measures show that private-sector employment growth falls in recessions and rises in expansionary periods.

The ADP National Employment Report is a monthly measure of the private-sector labor market based on aggregated payroll data of more than 26 million U.S. workers. ADP Research, a research arm of the ADP payroll processing firm, releases the report monthly. While the report provides much detail of employment by firm characteristics, the topline number in the report is private-sector employment changes, available since January 2010.

ADP employment changes track closely with the Bureau of Labor Statistics (BLS) measure of private-sector payroll employment in the Current Employment Statistics survey, data that is published every month as part of the Employment Situation Summary, also known as jobs day. In the figure, we compare seasonally adjusted monthly changes in employment for each measure smoothed to three-month moving averages. These smoothed changes are highly correlated (0.76).

While ADP data are only available as of 2010, BLS private-sector employment data goes back to 1938 and shows a clear relationship between the number of jobs and business cycles. Employment falls in recessions and rises in expansionary periods. The close relationship between these measures allows us to track any notable and sustained divergence, which would indicate a concern about the quality of government data.

Revelio vs. BLS employment data

       

Revelio Labs generates monthly U.S. employment estimates by counting social networking profiles on sites like LinkedIn. Total nonfarm employment changes from Revelio generally track BLS-based estimates.

Seasonally adjusted and nonseasonally adjusted estimates from Revelio are released monthly by industry, occupation, and geographic region. National data are available as of 2021. The data are normally published the day before the BLS employment report is released. Historically, Revelio’s monthly change usually falls below the BLS estimate, but over relatively short periods of time, the data series do tend to rise and fall together.

Challenger, Gray & Christmas vs. BLS unemployment rate

       

The Challenger job cut report data are somewhat noisy on a month-to-month basis, but spike noticeably during recessions.

Challenger, Gray & Christmas collects its job cut data by tracking public announcements made by U.S. companies in both the private and public sectors. The job cuts can include cuts in multinational plants (i.e., outside the United States). The report has been published monthly since 1994 and is typically released at the end of every month or the first week of the following month. The figure shows that large increases in announced job cuts occur in recessions (and when the unemployment rate spikes).

Of note, the spike in layoffs in spring 2025 was partially driven by announced layoffs in the federal government.

State-level UI claims vs. BLS unemployment rate

       

Monthly changes in unemployment insurance claims track the unemployment rate and provide a useful indication of recessionary periods.

While the U.S. Department of Labor (DOL) aggregates unemployment insurance (UI) claims data and is the source of the national data, it relies on administrative data processed at the state level that are generally posted on state government websites, potentially allowing a real-time accuracy check if concerns are raised about the accuracy of data released through the DOL portal. During the shutdown, the number of initial and continued claims has been updated weekly on the 539 report, a DOL-compiled dataset based on information submitted by state unemployment insurance offices. Right now, the data are only available on a nonseasonally adjusted basis because of the shutdown. The figure displays continued (or insured) UI claims and the unemployment rate, both as 12-month moving averages to remove some volatility and seasonality.

Continued UI claims closely track the unemployment rate: They spike during recessions and fall during economic recoveries. The close relationship between UI claims and the unemployment rate allows us to track any notable and sustained divergence, which could indicate a concern about the quality of government data. More information on the UI claims data, including initial claims, federal claims, and state-specific data, is updated weekly on EPI’s UI claims page.

Google Trends vs. BLS unemployment measures

       

Google Trends provides real-time data on the relative frequency of search terms for a given keyword, category of keywords, or a topic. With the right search terms, we can construct an index that tracks the overall unemployment rate

Google Trends is a publicly available database maintained by Google that provides Search Volume Indices. Search Volume Indices are a measure from 0–100 of the relative search intensity (number of searches for a given keyword divided by total searches) by geographic location and period. The data are updated in real time and are available from 2004 onward.

Google Trends data can be benchmarked to several different economic indicators. In fact, academics have used Google Trends data to “nowcast” GDP, private consumption, unemployment, recessions, and health outcomes. The figure shows the monthly average of select labor market-related search terms and the unemployment rate. The Google Trends relative search frequency for these terms tracks the unemployment rate and spikes during recessions. The close relationship between these measures allows us to track any notable and sustained divergence. Any prolonged increase in the search frequency of these select search terms could indicate a recessionary period.

Economic policy uncertainty vs. recession indicators

       

The economic policy uncertainty (EPU) index clearly responds to business cycle downturns—spiking sharply during the recessions of the early 1990s, early 2000s, late 2000s, and the COVID-19 pandemic.

The EPU index is calculated by Baker, Bloom, and Davis, and their full methodology can be found here. The index aggregates information from three basic components: search results from 10 large national newspapers measuring the volume of news articles discussing economic policy uncertainty; the number of federal tax code provisions set to expire over the next year; and the degree of disagreement among economic forecasters about future levels of key economic variables.

While the EPU is clearly cyclical, it does rise during some non-recessionary periods, most notably in the early 2000s and from 2011–-2013. The early 2000s increase is almost surely driven by the Iraq War. From 2011–2013, a rolling series of economic crises in the Eurozone, as well as short-term extensions of expiring provisions from the tax cuts passed in the first George W. Bush administration explain the spikes. The very large spike in early 2025 was related to the “Liberation Day” tariff announcements.

University of Michigan consumer sentiment data vs. recession indicators

       

Sharp falls in consumer sentiment are often followed shortly by recessions.

The University of Michigan consumer sentiment index has been calculated since 1960 (and has been calculated monthly since 1978). The index is derived from answers to five survey questions that have categorical answers (that is, their questions provide survey respondents with two to three possible responses to choose from, for example, “Would you say that your (and your family) are better off, or worse off financially than you were a year ago?”).

For each question, the University of Michigan researchers calculate a relative score for answers that subtracts the unfavorable responses from favorable responses. The sum of responses is then compared with a base value from 1966.

The index reached lows in 2022 that were comparable with most recessionary periods, despite a very strong economy. The clear explanation for that was the sharp inflation spike in late 2021 through mid-2022. Early 2025 saw similarly low measures, likely driven by concerns over the potential effects of tariffs.

Because the 1966 base year was a long time ago, it seems fair to ask if there are structural changes that might reliably change the level of consumer sentiment over time. For example, if rising inequality (or anything else) made U.S. households consistently less happy about their relative economic situation over time, they might generally have a lower “baseline” level of favorable sentiment.

To assess this, and to evaluate whether the University of Michigan consumer sentiment index might be useful for assessing the broader health of the economy, the figure below shows changes in the level of consumer sentiment and changes in inflation-adjusted personal consumption expenditures (the broadest level of household spending). The theory is that consumers’ sentiment about the economy should correlate positively with their actual spending patterns. The pattern is clear: Changes in consumer sentiment do indeed coincide tightly with changes in consumer spending.

Data Accountability Dashboard

Chicago FED's flow consistent unemployment rate (FCR) vs. BLS unemployment

       

The Chicago Federal Reserve flow-consistent unemployment rate (FCR) combines several sources of labor market data to predict monthly unemployment rates. The Chicago Fed FCR closely predicts the official unemployment rate published by the Bureau of Labor Statistics (BLS).

Twice a month, the Chicago Federal Reserve publishes the Real-Time Unemployment Rate Forecast, which predicts the BLS unemployment rate in advance of its official release. The key ingredient in this prediction is the Chicago Fed’s flow-consistent unemployment rate, which is in turn a prediction of Current Population Survey job finding and separation rates using real-time data like UI claims; Google Trends index for unemployment; Bloomberg consensus unemployment rate forecasts; Indeed and Lightcast job openings; ADP employment levels; weekly Morning Consult unemployment and job search activity; Conference Board labor market differentials; and JOLTS layoffs, discharges, and hiring rates. As a result, the Chicago Fed labor market indicators are partially based on U.S. government data, but they can be extended even in the absence of government data releases.

The numbers bear this out. According to new Bureau of Labor Statistics data:

• Union workers have greater access to employer retirement plans. 94% of unionized workers have access to employer-sponsored retirement plans compared with just 72% of nonunion workers.
  • 85% of unionized workers participate in employer-sponsored retirement plans compared with 52% of nonunionized workers.
• Union workers are more likely to have pensions. 62% of union workers have access to pensions compared with only 10% of nonunion workers.

Not only do unions provide workers with secure and dignified retirements, they also create healthier and safer workplaces and communities.

• Union workers have greater access to health insurance. 93% of unionized workers have access to employer-sponsored health benefits compared with 71% of nonunion workers. Union employers also contribute more to their employees’ health care benefits.
• Union workers have greater access to paid sick leave. 93% of unionized workers have access to paid sick leave compared with 81% of nonunion workers.
• Unions create safer workplaces. Unions empower workers to bring up unsafe working conditions without fear of retaliation.

When unions are strong, they help raise standards for all workers. States with higher union density are more likely to have higher incomes, greater access to retirement and health insurance, and enacted paid sick leave legislation.

Read more on the benefits of unions and the roadmap to rebuilding union power.

The cost of this wage suppression is real. Because wages make up a large share of income for most U.S. working families, lost wage growth meant sluggish income growth. However, there was one group where wage growth, and corresponding income growth, was substantial: the top 1%, whose wages skyrocketed by 182% since 1979. This uneven growth implies that today’s labor market, and the income derived from it, is deeply unbalanced. Rather than workers reaping the rewards of their rising productivity, a growing share of their labor has gone to the top 1%, leading to rising wage and income inequality.  

Decades of sluggish wage growth for typical workers—paired with the message that it was workers’ own fault for not obtaining the necessary skills for today’s economy—have left many pessimistic about whether policy changes can help them earn more. But if we want to achieve broad-based economic security in the U.S. economy, there is no alternative to raising pay for workers up and down the wage distribution. The U.S. economy generates enough income in aggregate to provide decent rates of pay growth for all workers—the challenge is a political one.

1

1

The cost of inequality to middle-class households

The stakes of rising inequality for middle-income households in the United States are enormous. By 2022, steadily rising inequality was costing these households roughly $30,000 per year.

Figure 1 compares actual income growth of the middle fifth of nonelderly households since 1979 with what this income growth could have been had inequality not increased. The graph highlights market-based incomes because rising inequality was driven entirely by this income category—earnings from the labor market and business income like dividends and interest payments. We focus on incomes of nonelderly households because inequality largely stemmed from unbalanced labor market power, and nonelderly income is dominated by labor earnings.

In 2022, the average market income of the middle fifth of households was $96,335. But their income would have been $127,011—roughly 32% (over $30,000) higher—had inequality not increased after 1979. Put another way, if the middle fifth of households had seen market income grow at the overall average rate—an overall average pulled up by stratospheric growth at the very top—their income could have been roughly $30,000 higher by 2022.

Share this chart:

2

2

Income inequality is driven by unbalanced power in labor markets

The inequality in market-based incomes highlighted in Figure 1 was overwhelmingly driven by unbalanced power in labor markets that prevented typical workers from seeing their paychecks keep up with overall economic growth. Figure 2 shows hourly pay (including benefits) for the roughly 80% of the private-sector workforce who are not managers or supervisors and a measure of economy-wide productivity—defined as the income generated in an average hour of work in the U.S. economy. The rising gap between these lines is how we define wage suppression—keeping typical workers’ wage growth slower than growth in incomes in the rest of the economy.

In the three decades following World War II, hourly compensation for the vast majority of workers rose 2.1% on an annualized basis, roughly in line with productivity growth of 2.5%. But for most of the last five decades (except for a brief period in the late 1990s), pay for the vast majority lagged further behind overall productivity. Over the entire 1979 to 2025 period, hourly pay for a typical worker rose just 0.6% while productivity increased 1.4% on an annualized basis. This means that workers have been producing far more than they receive in their paychecks and benefit packages from their employers, and this wedge has grown substantially over time. Some of those gains in productivity went to greater profits for corporations but an even larger portion went to highly paid managers and professionals, resulting in a vast increase in wage inequality over the last five decades.

Had we pursued policies to support broad-based wage growth—instead of wage suppression—typical workers would have much higher pay today. In fact, if annual pay for production and nonsupervisory workers had instead tracked productivity growth since 1979, their pay would be 45% higher, or $30,000 more than it is today.

Share this chart:

3

3

Wage suppression boosted profits

The declining labor share of corporate-sector income—the share of corporate income received by workers in the form of wages and benefits—explains a portion of the wedge between productivity and pay. The labor share of corporate-sector income has not recovered to its pre-pandemic level, and remains well below its 1979 to 2008 average. Figure 3 illustrates the labor share of corporate-sector income since 1979. Typically, the labor share rises during recessions because profits fall much faster than wages during downturns. Then, in the early stages of recovery, the labor share falls significantly as profits increase much more rapidly than wages. Usually, the labor share then rises again late in an expansion as labor markets tighten and workers regain the bargaining power necessary to secure wage increases. Since 2000, no business cycle expansion has gone on long enough with unemployment low enough to put serious upward pressure on the labor share of income. Small upticks were seen sporadically over the last 15 years, but this progress, but this progress has stalled out.

Two things about the shift from labor compensation to profits in recent decades deserve some context. First, even as measured in official data, this shift explains well under half of the overall wedge between productivity and pay—the biggest piece of this gap remains inequality within labor compensation (we return to this point further below). Second, some of the changes in the official data might be more the outcome of accounting decisions than of fundamental changes in the economy.

Since the early 2000s, capital income (dividends, capital gains, and business income) has been taxed at lower rates than labor income. Highly privileged economic actors who have the ability to engage in tax evasion by reclassifying their income have likely done so in this time—relabeling income once classified as wages and salaries as capital income instead to pay lower tax rates. Recent research has estimated that roughly a third of the decline in the labor share of income might be driven by this kind of tax evasion rather than by a genuine change in which factors of production are seeing income gains.

Share this chart:

4

4

Highly paid corporate managers and professionals benefited the most from wage suppression

The largest portion of the growing wedge between productivity and pay can be explained by the astronomical wage growth for those at the highest end of the wage distribution. The ability of those at the very top to claim an ever-larger share of overall wages is evident in Figure 4. Annual earnings for the top 1% have grown 182% since 1979, while wages of the bottom 90% have grown just 44%. If the wages of the bottom 90% had grown at the average pace over this period—meaning that wages grew equally across the board—then wages for the bottom 90% would have grown by 65%, far higher and roughly 50% faster than the actual growth experienced. Earnings are so concentrated at the top that average growth is well above the 90th percentile of the earnings distribution. That means workers need to be among the highest 10% of wage earners to even experience average earnings growth. And everyone else—more than 90% of the workforce—saw less than average growth over the last four and a half decades.

Share this chart:

5

5

Since 1979, middle-wage workers only saw short-lived real wage growth during tightening labor markets

Between 1979 and 2025, the hourly wages of middle-wage workers (workers earning the median wage) cumulatively grew just 30%—about 0.6% per year. Yet even this disappointing cumulative wage growth occurred only because wages grew in the late 1990s and the late 2010s–2020s. Apart from these two periods, the wages of middle-wage workers were totally flat or in decline since 1979. Figure 5 shows real median wages since 1979. The red portions of the line highlight periods when median wage growth was genuinely stagnant, while the green portions highlight when growth occurred.

In the absence of strong institutions and labor standards like robust unions and high minimum wages, strong wage growth occurs only when labor markets are tight. In tight labor markets, the relative scarcity of workers gives them more power in the workplace to effectively demand decent hourly wage growth across the board.

Share this chart:

6

6

Wage suppression was not driven by increased benefits

The suppressed “pay” portion of the pay–productivity chart includes wages and benefits, suggesting that benefits have not offset slow wage growth. In fact, by many measures benefit generosity and availability have deteriorated alongside wage suppression. We explore trends in two key employment benefits in Figure 6: employer-sponsored health insurance and retirement coverage. We define employer-provided health insurance as insurance for which the employer pays some or all of the premium. We define retiree coverage as coverage a worker receives at their firm that offers it. We isolate workers who are strongly attached to their job, working at least 20 hours per week and 26 weeks per year.

For workers, sluggish pay has been compounded by the fact that health insurance and retirement coverage have declined significantly over time. Health insurance fell from 69.0% to 52.2% between 1979 and 2024. At the same time, retiree coverage nearly fell by half from 50.6% down to 29.0%. Furthermore, plan generosity has suffered with the trend away from defined benefit to defined contribution retirement benefits, as well as the shift to high deductible health plans.

Share this chart:

7

7

CEO pay grabs a larger share of wages

In a nation of increasing inequality, the most extreme disparities are between the heads of large American corporations and typical workers. Figure 7 tracks the ratio of pay of CEOs at the 350 largest public U.S. firms to the pay of typical workers. In 1965, these CEOs made 21 times what typical workers made. As of 2025, they make 325 times typical workers’ pay. This higher pay for CEOs does not reflect any increased contribution to corporate output or growth in the skills or productivity of CEOs. CEO pay gains help explain the growing divergence between pay and productivity. Further, the pay of CEOs and other highly placed corporate managers sets norms and market benchmarks across a range of elite occupations. The pay of specialty physicians, lawyers, consultants, and even sectors like university and nonprofit administration is likely pulled upward by rising pay for corporate executives.

Share this chart:

8

8

Tolerating excess unemployment has been deeply damaging for wage growth

As a rough rule of thumb, real wages for middle-wage workers outright fall when unemployment is greater than 5%.

Prioritizing high-pressure labor markets—those identified by extended periods of low unemployment—is key to providing lower- and middle-wage workers the leverage to bid up their wages. Figure 8 shows that the threshold for achieving any real wage growth at all is around 5% unemployment. In general, real wages grow when unemployment is lower than 5%, but they fall when unemployment is higher. Extended periods of low unemployment can also reduce wage inequality and racial disparities.

Share this chart:

9

9

The federal minimum wage no longer lifts wages

Figure 9 shows the decline in the real (inflation-adjusted) value of the federal minimum wage since its high in 1968. The last increase to the minimum wage was enacted in 2009, 17 years ago, which is the longest period without a minimum wage increase since its inception in 1938.

The minimum wage is essential for establishing the wage levels for the bottom fifth of wage earners. Today’s low-wage workers are far more educated, older, and experienced than low-wage workers in 1968. Yet, despite being more skilled and productive, their wages are 43% lower than wages earned in 1968, and 32% lower than the most recent increase in the minimum wage 17 years ago. The failure to raise the federal minimum wage has especially adverse effects on women and Black and Hispanic workers.

Luckily, many states have refused to follow the federal government’s inaction and have passed higher minimum wages at an increasing rate over the past two decades. Seventeen states and the District of Columbia now have minimum wages of at least $15 per hour, and more workers live in states with minimum wages over $15 than in states only bound by the too low federal minimum wage.

Share this chart:

10

10

The assault on collective bargaining hurts all workers

One of the main causes of wage suppression and rising wage inequality is the decline of collective bargaining, which lowers the wages of both union and nonunion workers. Collective bargaining not only raises wages for organized workers but also leads other employers to raise the wages and benefits of nonunion workers to come closer to union wage standards.

The erosion of collective bargaining can explain from one-fifth to one-third of the growth of wage inequality between 1973 and 2007; it had a greater impact on men than women. This erosion of unionization occurred despite large numbers of workers indicating they would prefer collective bargaining if they had a choice. But the political power of those with the most income, wealth, and power has prevented the adoption of laws to modernize our labor–management system and enable workers to pursue collective bargaining.

Figure 10 shows how wages could have grown if unionization hadn’t declined since 1973. If union density in 2025 had been the same level as union density levels of 1973 (14 percentage points higher than it is today), median wages would rise by 10.1%.

Share this chart:

Click here for the 2015 archive of this page.

The Gallup polling also showed that unions are favorable across generations and party lines.

• 77% of adults ages 18–34, 74% of adults ages 35–54, and 67% of adults ages 55 and older approve of unions. 

• 89% of Democrats, 70% of Independents, and 52% of Republicans approve of unions. 

This is just one of many recent surveys demonstrating the broad popularity of unions:  

• Unions are more popular than big business. People in the U.S. favor unions over big business by more than 16 percentage points—the greatest gap ever recorded.
• 56 million U.S. workers would join a union tomorrow if they could. 43% of nonunion workers would vote to unionize their workplace if given the opportunity. 

• Majority views union decline as bad for U.S. Most people in the U.S. say the decline in union density is bad for the country (60%) and bad for working people (62%). 

It should be no surprise that the public views unions favorably: Unions help workers win higher wages, advocate for safer and healthier workplaces, support a thriving middle class, and strengthen democracy by boosting civic participation.  

Get the fact sheet

Now is the time for policymakers to pass comprehensive legislation making it easier for workers to unionize. A new EPI report shows that tripling U.S. union membership would deliver: 

• 14.5% wage increases for the typical worker, equal to more than $7,700 per year or nearly $270,000 over a 35-year career. 

• $1.2 trillion shifted into workers’ pockets every year, reversing a third of the growth in inequality since 1979.  

• Health insurance for 6.8 million more people.  

• Greater equality and smaller racial wage gaps. Unions raise wages for everyone and help prevent pay disparities, especially for Black and Hispanic workers.  

Read more about the benefits of rebuilding union power—and the roadmap to get there.

Key takeaways:

• So-called right-to-work (RTW) laws undermine workers’ ability to form unions. Unionization rates are clearly lower in RTW states—and since unions raise wages, that suppression lowers pay.   

• States with RTW laws are also more likely to have other anti-worker laws and generally weaker labor standards. This is likely because both RTW and other anti-worker policies stem from the same anti-worker political roots, and because RTW’s suppression of unions deprives these states of strong potential political champions (unions) for other pro-worker policies.  

• Non-RTW states are more likely to have pro-worker policies like higher minimum wages and prevailing wage laws, restrictions on noncompete agreements, pay transparency policies, and more protective unemployment insurance benefits. 

• As a result, workers in RTW states are paid 6.7% less on average than workers in non-RTW states—more than double the 3.2% wage penalty we found in 2015. That means the RTW wage penalty has increased in recent years and now translates to over $4,000 less per year for a median full-time worker in a RTW state.

• The RTW wage penalty is steeper for women (7.3%) than for men (6.3%), and for Black (9.7%) and Hispanic workers (10.0%) than for white workers (5.4%).   

• Lawmakers already have the policy tools they need to boost wages and improve other outcomes for workers: 

• • At the state level, rolling back existing RTW laws is the most powerful lever.  

• • At the federal level, the Protecting the Right to Organize Act (which includes a ban on RTW) would be transformational for efforts to organize workers.  

• • Policymakers could also make progress on numerous other fronts, such as making UI systems more protective and raising minimum wages. 

In the United States, so-called right-to-work (RTW) laws are anti-union state policies promoted by big business interests to weaken unions and limit workers’ bargaining power. Despite the name, RTW laws do not provide any sort of job protection or right to a job. Rather, the deceptively named policy starves unions of resources, making it harder for them to negotiate collectively for better wages, benefits, and working conditions.

RTW laws emerged in the 1940s as part of anti-union campaigns to suppress worker organizing and maintain Jim Crow labor relations in Southern states; these campaigns worsened economic inequality and racial disparities (Childers 2024). Today, 26 states have active “right-to-work” statutes.1

As intended, RTW laws reduce union density (the share of workers who are either members of a union or covered by a union contract). These policies prohibit unions and employers from negotiating over union security—the contract terms under which workers covered by a union contract either join the union or pay an agency fee covering their share of representation costs. This deprives unions of critical resources and suppresses union membership (Sherer et al. 2026). RTW states—specifically those with RTW laws for at least 25 years—have an average union density rate of 6.2%, far less than half the union density rate in non-RTW states (16.0%) (see Appendix Table 1).

RTW laws not only reduce the number of workers with collective bargaining agreements but also limit unions’ political power to advocate for stronger labor standards for workers more broadly. This is, in part, why states without anti-union restrictions are more likely to pass protections such as prevailing wage mandates, restrictions on noncompete agreements, higher minimum wages, and pay transparency laws.

To assess the effects of RTW laws on outcomes for workers, we analyze whether wages are lower in RTW states after controlling for key demographic-, job-, and state-level characteristics (see Appendix Table 1 for full demographic controls as well as median and average wages by RTW status). We find that wages in “right-to-work” states are, on average, 6.7% lower than in non-RTW states. This penalty is worse for Black and Hispanic workers, as well as young workers and those without college degrees. Notably, the penalty is also larger for nonunion workers. This is because strong unions benefit workers across the state, not just those in unionized workplaces (Shierholz et al. 2026). In RTW states with weaker unions, nonunionized workers do not experience these positive “spillover” effects.

Pro-worker policies are more common in states without RTW laws

The strong correlation between wages and states’ RTW status is not surprising. Not only do RTW states have lower unionization rates, leaving workers with less leverage to bid up wages, but they often maintain other anti-worker policies. Combined, these anti-worker policies tilt bargaining power toward employers, weakening workers’ ability to garner higher wages. States without RTW laws are more likely to have a wide array of stronger labor standards in place that raise pay, protect workers’ ability to switch jobs, and support those who lose one. These same states—where unionization rates have not been suppressed by RTW laws—have a better safety net and stronger labor standards in part because stronger unions give workers more political power and the ability to better lobby for pro-worker policies writ large (Shierholz et al. 2026).

Figure A displays the incidence of these labor standards for states with and without anti-union RTW laws.2 This is far from a complete list of relevant policies that may correlate with RTW status and wage levels, but they are illustrative.

Figure A

Prevailing wage

Prevailing wage laws apply to certain contracts for public works and require that contractors and subcontractors pay the employees performing such work a prevailing wage rate. The prevailing wage rate is defined as the average wage paid to similarly employed workers in a specific occupation in the area of intended employment. Federal law requires this of federal contracts while state and local laws may apply to state and local contracts.

Prevailing wage laws are more common in non-RTW states. Figure A shows that nearly all (96.0%) of non-RTW states have prevailing wage laws, while only 23.1% of RTW states have these protections for workers on state contracts.3 Without prevailing wage requirements, contractors may reduce their workers’ wages to win bids on government contracts—putting contractors who pay their workers fair wages at a disadvantage.

Restrictions on noncompetes

Noncompete agreements are employment provisions that ban workers at a company from joining a competing business or starting their own for a set period after leaving a job. Noncompetes undermine economic dynamism and innovation; they depress business formation and labor mobility, hurt productivity and growth, raise prices, shrink workers’ wages, and restrict workers’ freedom (Shierholz 2024).

One of the most effective ways workers achieve higher wages is by securing another offer, and then either renegotiating their current salary or leaving for the new higher-paying job. It is not surprising then that workers in states that enforce noncompetes are paid less than similar workers in states that do not enforce noncompetes (Starr 2019). States without anti-worker RTW laws are nearly four times as likely to restrict noncompetes as RTW states (44.0% vs. 11.5%).4 In this analysis, we include only full bans on noncompetes and states with income-based restrictions. States with only small carve-outs for certain occupations provide a much weaker standard and are, by definition, less broad, so those aren’t included.

Minimum wage

Raising the minimum wage is particularly important for lower-wage workers. While the federal minimum wage has sat at $7.25 an hour since 2009, losing over 30% of its purchasing power, workers in 31 states enjoy a higher wage floor (Zipperer 2026). Non-RTW states are three times as likely to have minimum wages higher than the federal minimum compared with RTW states (92.0% vs. 30.8%). This is not a small difference. As of 2026, the average value of the minimum wage in non-RTW states is 60% higher—$5.36 more per hour—than in RTW states.5

Pay transparency

Pay transparency laws prevent employers from lowballing wage offers by requiring them to include wage information in job postings. While they vary, all laws include some requirement that employers provide salary information in job postings or if directly requested by applicants. Lack of knowledge about wages and benefits keeps potential workers in the dark and limits their ability to learn about and apply for better paid opportunities. A study on Colorado’s newly passed wage transparency law found that wages increased 4.2% faster than those in neighboring states without such laws (Shedge 2025). Further, wage transparency has the potential to reduce gender- and race-based discrimination by arming jobseekers with more information and limiting employers’ ability to pay different amounts to similarly qualified candidates. Only two RTW states have a pay transparency law, compared with 64% of non-RTW states.6

Unemployment insurance

Access to unemployment insurance (UI) also shapes workers’ ability to seek better wages. When a worker is laid off and their household income falls, unemployment insurance supports them and their family until they find another job. UI is a joint federal-state program that relies on state UI systems to effectively deliver benefits to unemployed workers. States differ in their rules and approaches under the federal UI framework, leading to wide variation in the share of unemployed workers receiving UI benefits (the UI recipiency rate).

Better UI not only cushions workers and their families but also provides workers with additional leverage in the labor market. When UI access and generosity fall, job seekers are forced to settle for lower wages and employers leverage this by offering less (Dahl and Knepper 2026). On the flip side, research shows a positive relationship between more UI generosity and wages (Rinz and Wasser 2026). The UI recipiency rate is significantly higher in non-RTW states compared with RTW states (32.8% vs. 18.7%).7 This is not surprising because non-RTW states have stronger unions, which fight to increase UI eligibility, ease of access, benefit levels, and benefit duration—to the benefit of all workers, union and nonunion (Hertel-Fernandez and Gould-Werth 2020). As with other pro-worker policies, better UI access can contribute to higher wages in non-RTW states.

Other policy levers

Not only do workers in non-RTW states enjoy higher wages, but they also benefit from increased economic security thanks to higher health insurance rates, higher public education spending, paid family and medical leave access, and no preemptions against higher labor standards. Every non-RTW state expanded Medicaid compared with fewer than two-thirds of RTW states (61.5%).8 Unsurprisingly, the uninsured rate is higher in RTW states than in non-RTW states (8.9% vs. 6.2%).9

Investments in public education deliver a more productive workforce and more informed and engaged society. Increased funding for schools not only improves educational attainment, but also increases wages and family incomes (Jackson, Johnson, and Persico 2016). Non-RTW states spend 45% more on public education than RTW states ($24,084 vs. $16,620).10 Paid family and medical leave provides essential benefits that help workers maintain their livelihoods while taking care of themselves and their families. These laws are also shown to increase labor force participation, job retention, and earnings (Glynn 2020). Only one RTW state has comprehensive paid family and medical leave provisions for private-sector workers, compared with 56% of non-RTW states.11

Finally, some cities, counties, and other local governments enact policies that raise standards for working people, but some state legislatures force those standards back down with preemption—the use of state law to void local ordinances. Preemptions apply to policies such as minimum wage, prevailing wages, paid leave, and fair scheduling (EPI 2025). Preemption laws interfere with local governments’ ability to set job quality standards. They suppress wages, exacerbate racial inequities, and reduce worker power (Sherer, Cohn, and Ahdoot 2025). These preemptions are more common in RTW states.

Workers in RTW states are paid less

Our analysis of the relationship between wages and RTW status puts states into three categories: long-term RTW states, always non-RTW states, and “switcher” states that have changed their RTW status since 2011.12 Appendix Table 1 provides descriptive statistics for these three groups of states, and Appendix Figure A provides a map of states by RTW status. Our estimates focus on comparisons between long-term RTW states and always non-RTW states.

A simple comparison of hourly wages, the primary variable of interest, reveals that average wages are 22.9% higher in non-RTW states ($40.28 in non-RTW states vs. $32.78 in RTW states). Median wages are 19.2% higher in non-RTW states ($28.79 vs. $24.16). Because there are differences between worker-, job-, and state-level characteristics in RTW and non-RTW states, and since some of these characteristics will directly impact workers’ wages, it is important to control for these factors in a multivariate regression model. This allows us to more accurately identify the relationship between RTW status and wages.

We estimate log wage equations using Bureau of Labor Statistics Current Population Survey Outgoing Rotation Group (CPS-ORG) data for 2023–2025. We pool three years of data to minimize any spurious year-specific economic relationships, thereby helping us achieve more precise estimates. The total sample consists of 234,155 workers, ages 16+, who earn wages and salaries. About 42% of the sample lives in states with RTW laws (see Appendix Table 1 for the sample’s full demographic breakdown).13

In Table 1, we construct a regression model, starting with an unadjusted framework and building up to a model that controls for the full range of explanatory variables. The dependent variable is the natural log of hourly wages.

The variable of interest is an indicator variable when the worker lives in a long-term RTW state. We also include an indicator in the regression for a switcher state, though that is not our focus of interest (see Appendix Table 2 for a complete set of regression results, including the coefficient for switcher states). There are currently 26 states with RTW laws on the books and 25 without them including Washington, D.C. (see Appendix Figure A). There are five switcher states that have adopted RTW in recent years: Indiana (2012), Michigan (2013), Wisconsin (2015), West Virginia (2016), and Kentucky (2017). In 2023, Michigan became the first state to repeal its RTW statute. We separate the five switcher states from the analysis to better isolate the relationship between wages and the long-run effects of RTW.

The naïve model, without any controls, suggests that workers in RTW states are paid 17.8% less than workers in non-RTW states (Model I in Table 1).14 But many factors contribute to wage differences: demographic characteristics such as age, race/ethnicity, gender, and level of education are relevant as well as job-related factors such as work hours, occupation, and industry. Since these vary across states, they reduce the measured relationship between RTW status and wages. Model II shows that when the analysis compares more similar workers with each other, the RTW penalty drops to 14.0%.15

There are other factors that are different in RTW states, aside from these individual and job-related characteristics. Most importantly, the cost of living. The cost of living in non-RTW states is higher than in RTW states, as shown in Appendix Table 1. In prior research, Gould and Kimball (2015) showed that choice of price indicator makes no material difference on the measured relationship between RTW and wages. Here we include the Bureau of Economic Analysis’ Regional Price Parity index: The regional price parity index is higher in non-RTW states.16 Not surprisingly, this correlates with wage levels and therefore is important to include in the model. As with earlier studies, we also include state unemployment rates to control for differences in economic conditions.

After including the full set of demographic, individual-, and state-level labor market and cost of living controls (Model III), our analysis finds that workers in RTW states are paid 6.7% less than similar workers in non-RTW states. For the average worker, this is a difference of $2.70 per hour, or more than $5,600 for full-time workers at 2,080 hours over the year. However, because the average is skewed upward due to wage inequality, an assessment at the middle of the wage distribution provides a more conservative but perhaps more robust estimate of the RTW wage penalty. Acknowledging that this is a likely understatement of the full relationship, the RTW wage penalty for the median worker at full-time hours is over $4,000 for the year.

Table 1

This does not mean that RTW laws are a causal explanation for this 6.7% wage penalty. As demonstrated earlier in this report, RTW status correlates strongly with a number of other anti-worker policies. Other research has attempted to calculate the causal effects of RTW laws on wages. Using the switcher states in the most recent period, Fortin, Lemieux, and Lloyd (2022) found smaller wage penalties, which is to be expected given the shorter time frame for measurement post-adoption. Our wage penalty for switcher states—shown in the final column of Appendix Table 2—is much smaller (closer to zero) than the coefficient for long-term RTW states, though not statistically significant at conventional levels. Dasgupta and Merchant (2023) also examine states recently adopting RTW laws and find a statistically significant decline in annual wages by almost $1,900. Again, these states are more recent RTW adopters, so the full effects we find may take time to materialize.

We find that this relationship between long-term RTW status and wages remains economically and statistically significant under alternative specifications of our econometric model. For instance, removing major industries and occupations changes the coefficient to -0.071. Restricting the model to RTW status as of 2026—removing the separate indicator for states that switched since 2011—yields a coefficient estimate of -0.060. Doing the same but dropping Michigan (since its status changed in the measurement period) results in a coefficient of -0.064. In our final model, we do not include an indicator for unionization, unlike Gould and Kimball (2015). If we reintroduce that variable, the coefficient is -0.063.

In each of our regressions discussed above, we include a three-year average as it improves data reliability and allows us to run smaller groups of the data (see next section). When we run our analysis using just 2025, our results do not change. In fact, when we run our three-year final model for the periods 2021–2023, 2022–2024, and 2023–2025, our coefficients remain in the -0.070 to -0.068 range.

Wage penalties are steeper for women, Black, Hispanic, and nonunion workers in RTW states

The RTW penalty is not uniform across demographic groups and other labor market characteristics. Table 2 below shows the results of a series of final-model regressions (comparable to Model III in Table 1) for a set of demographic groups.

We find that women’s wages in RTW states are penalized at a higher rate (7.3%) than men’s (6.3%). The wage penalty also persists across all racial and ethnic groups, but it is almost twice as large for Black (9.7%) and Hispanic (10.0%) workers as for similar white workers (5.4%) and more than twice as large as for Asian American and Pacific Islander (AAPI) workers (4.6%). Young workers between the ages of 16 and 24 face a greater penalty (8.4%) than older workers. And workers without a college degree experience a much higher penalty (7.8%) than workers with a college degree (4.7%).

Table 2

It is particularly important to note that RTW laws have a statistically significant negative effect on the wages of nonunion workers, as shown in the last row of Table 2. Our analysis indicates that nonunion workers in RTW states have wages that are 6.3% lower, on average, than their counterparts in non-RTW states. Union members likely experience a smaller penalty (3.6%) because their wages tend to be higher than those in nonunionized workplaces, regardless of the state’s RTW status (McNicholas et al. 2025). However, the wage impact of unions extends beyond those directly covered by a union contract. Unions lift wage standards across entire industries, meaning that nonunion workers in states with higher union density are more likely to receive better pay.

As the gap in wages and labor standards between RTW and non-RTW states has grown, so has the wage penalty

Using our final model (Model III in Table 1) with pooled 2023–2025 data, we find a 6.7% RTW wage penalty—a larger penalty than we found in earlier examinations. For example, we found a 3.2% wage penalty using data for 2009 (Gould and Shierholz 2011) and pooled 2010–2012 data (Gould and Kimball 2015). Some of the increase in the current estimate of the RTW wage penalty reflects small changes (likely improvements) in our methods. But part of the higher estimate persists even when we apply our current methods to earlier data: Applying these methods to the 2010–2012 data yields an RTW wage penalty of 4.7%.

One method change concerns “switchers”—states that changed from non-RTW to RTW status (or back) since 2019. In the 2010–2012 period, there were no switchers to affect the data.17 In our analysis, we isolate switchers from the long-term RTW group, but the RTW penalty remains negative and significant no matter how one deals with the issue of switcher states. For example, recategorizing the five switchers to match their 2012 status (all non-RTW) lowers the measured RTW wage penalty in the 2010–2012 period to 4.1%, but it remains significant.

Other changes relative to our earlier research include controlling for public-sector status; removing union as an independent variable from the regression; using the full working age range of 16 and up (as opposed to 18–64); and correcting an issue with nonrandom missing values to the metropolitan variable. We’ve further clustered standard errors at the state level as suggested by Douglas (2024), which does not affect the size of the coefficients, only the size of their standard errors and resulting statistical significance.18

Setting those minor model differences aside, a consistent comparison over time sees the RTW wage penalty rising from 4.7% in 2010–2012 to 6.7% in 2023–2025. One possible reason why this RTW wage penalty grew could be the increasingly divergent policy trajectories of RTW and non-RTW states over this period.

For example, the gap between minimum wages in RTW states and non-RTW states has grown since 2010, largely because policymakers in most RTW states have refused to increase their minimum wage despite the fact that the federal minimum wage has been stagnant in nominal terms since 2009. In that year, the average minimum wage in RTW states was $7.25 while the average minimum wage in non-RTW states was $7.54. That gap has grown enormously, from just $0.29 per hour in 2009 to $5.36 per hour by 2026 ($14.24 in non-RTW states vs. $8.89 in RTW states).

There were other potentially relevant policy changes between 2012 and 2025. The first state-level pay transparency law was passed in Maryland in 2020. Seventeen states followed suit in subsequent years, 15 of which were non-RTW states. On the flip side, six states have repealed their prevailing wage laws since 2012—all of which were RTW. Only one state, Michigan, reinstated its prevailing wage law. It did so in 2023, the same year it repealed RTW. While this is far from an exhaustive list, it’s not surprising that the measured RTW wage penalty has grown over time.

Repealing anti-union “right-to-work” laws and strengthening labor law is key to raising wages

The wage penalty for workers in RTW states has grown since the last time we conducted this research a decade ago. Long-standing and more recently enacted anti-union RTW laws in 26 states, in tandem with other anti-worker policies, have further eroded workers’ collective bargaining power, suppressed wages, and lowered labor standards. These outcomes are linked to clear policy choices, and lawmakers at every level of government have the tools they need to restore workers’ union rights, strengthen labor standards, and lift wages. Across the U.S., 56 million workers say they want a union in their workplace, yet only 10% of all workers have one (Shierholz et al. 2026). Current federal and state labor laws—including anti-union RTW laws—place too many obstacles in workers’ path to unionizing and give employers too much power to interfere with workers’ free choice.

First and foremost, states with RTW laws should follow Michigan’s recent lead and repeal them. Repealing state RTW laws will enable workers to unionize and raise wages across the country. Indeed, recent EPI research finds that equalizing collective bargaining rights across all states by repealing RTW laws that affect private employees and ensuring collective bargaining rights for public employees would increase national union density by almost 50% (from 9.9 to 14.4%) (Shierholz et al. 2026). States should also consider passing constitutional amendments that affirm collective bargaining rights and explicitly bar RTW-style restrictions, such as the 2022 Workers’ Rights Amendment in Illinois (Sherer 2026).

At the federal level, lawmakers should pass the Protecting the Right to Organize (PRO) Act. The PRO Act is designed to address major weaknesses of the National Labor Relations Act (NLRA), which has been severely eroded by amendments and court decisions. Among many other reforms, the PRO Act would eliminate the option for states to maintain anti-union RTW laws—restoring full bargaining rights to workers in all states as intended when the NLRA was originally passed in 1935 (McNicholas, Poydock, and Rhinehart 2021). Since its first introduction in 2019, the PRO Act has passed the House of Representatives twice with bipartisan support (Shierholz et al. 2026).

In addition to proposed PRO Act reforms that would enable newly unionized workers to pursue arbitration to achieve a first contract (if an employer fails to negotiate in good faith), federal labor law should set a minimum standard that such contracts include a cost-of-living adjustment (COLA). To further expand the benefits of collective bargaining, Congress should amend the NLRA to require default collective bargaining at any firm where CEO-to-worker pay ratios exceed 100 times what a typical worker makes in their industry (as this ratio would be one proxy for whether conditions at an employer are particularly unequal) (Shierholz et al. 2026). 

Legislators at both the state and federal level should also look to the list of anti-worker laws that are typically found in RTW states for more opportunities to rebalance unequal power in the labor market and improve wages and working conditions. These policies include:

• Raising the minimum wage;
• Banning noncompete agreements;
• Strengthening federal unemployment insurance guidelines and state UI systems;
• Removing state restrictions that preempt local governments from raising wages and strengthening labor standards.

These are among just a few of the policy levers states can use to raise wages and standards (EPI 2026b).

Our analysis suggests that repealing anti-union RTW laws and enabling workers to increase unionization levels are key to raising wages, both because RTW laws directly suppress unionization rates (and unions raise wages), and because weakening unions limits workers’ collective ability to shape other public policies.

Acknowledgements

The authors are grateful to Josh Bivens, Jennifer Sherer, Hilary Wething, and Ben Zipperer for their helpful comments and expertise.

Appendix

Appendix Figure A

Appendix Table 1

Appendix Table 2

Notes

1. See Appendix Figure A for RTW status. As of 2026, these states are Alabama, Arizona, Arkansas, Florida, Georgia, Iowa, Idaho, Indiana, Kansas, Kentucky, Louisiana, Mississippi, North Carolina, North Dakota, Nebraska, Nevada, Oklahoma, South Carolina, South Dakota, Tennessee, Texas, Utah, Virginia, Wisconsin, West Virginia, and Wyoming.

2. States are categorized as “RTW” or “non-RTW” based on their status as of 2026.

3. Authors’ analysis of prevailing wage laws from DOL (2023) and state legislative websites.

4. Authors’ analysis of noncompete restriction policies from EIG (2026).

5. Authors’ analysis of minimum wage data from EPI (2026c).

6. Authors’ analysis of pay transparency laws from GovDocs (2026).

7. Authors’ analysis of UI recipiency rate data from DOL-ETA.

8. Authors’ analysis of Medicaid expansion policies from KFF (2026).

9. Authors’ analysis of uninsurance rates by state, 2023 through 2025, from Carter (2025).

10. Authors’ analysis of per-pupil education spending data from NCES-NPEFS (2024).

11. Authors’ analysis of paid family and medical leave policies from Wielk (2026).

12. Prior to 2012, the last state to change its status was Oklahoma in 2001.

13. This analysis builds on earlier work by Gould and Kimball (2015); see the report for a full description of the methodology and list of controls.

14. Interpreting the results from these semilogarithmic functions requires utilizing the exponential function on the coefficient. Specific to the binary variable coefficient (β1) for RTW, the percent change in workers’ wages resulting from a state being RTW can be calculated by the formula: 100*[exp(β1)-1]. Typically, the result of this equation will be very close to the coefficient itself but will differ more as the coefficient becomes larger. We show the coefficient in the first row of Table 1 and the exponentiated result as a percent in the third row.

15. Full regression results for covariates in all three models can be found in Appendix Table 2.

16. Because the 2025 BEA RPP data were not yet available at the time of publication, we assign an average of 2022–2024 data to all years in our sample (2023–2025).

17. Gould and Kimball (2015) classify Indiana as non-RTW as its RTW law had just taken effect.

18. We do not control for state-level fixed effects, as Douglas (2024) suggested, because state-level fixed effects perfectly correlate with RTW status and would deem our results uninterpretable.

References

Bureau of Economic Analysis, Regional Price Parities by State and Metro Area: All items (BEA-RPP). Various years [Release date: February 19, 2026]. Public data series accessed via the Regional Data interactive tool. Accessed August 1, 2026. 

Carter, Caitlin. 2025. “Health Insurance Coverage by State: 2023 and 2024.” U.S. Census Bureau, September 11, 2025. 

Childers, Chandra. 2024. The Evolution of the Southern Economic Development Strategy: Rooted in Racism and Economic Exploitation: Part One. Economic Policy Institute, May 2024. 

Dahl, B. Gordon, and Matthew Knepper. 2026. “Unemployment Insurance, Starting Salaries, and Jobs: Evidence from Multi-state Firms.” The Review of Economic Studies, rdag056. https://doi.org/10.1093/restud/rdag056.

Dasgupta, Kabir, and Zofsha Merchant. 2023. “Understanding Workers’ Financial Wellbeing in States with Right-to-Work Laws.” FEDS Notes (The Federal Reserve), September 8, 2023.

Department of Labor (DOL). 2023. “Dollar Threshold Amount for Contract Coverage Under State Prevailing Wage Laws.” Last modified January 1, 2023. 

Department of Labor, Employment and Training Administration (DOL-ETA). Various years. Unemployment Insurance Chartbook. Public data series accessed via the interactive tool. Accessed August 6, 2026. 

Douglas, Christopher C. 2024. Right-to-Work States Do Not Have Lower Wages. Mackinac Center, December 2024. 

Economic Innovation Group (EIG). 2026. “State Noncompete Law Tracker.” Last modified March 24, 2026. 

Economic Policy Institute (EPI). 2025. “Workers’ Rights Preemption in the U.S.” Last modified February 2025. 

Economic Policy Institute (EPI). 2026a. Current Population Survey Extracts, Version 2026.7.8, https://microdata.epi.org.

Economic Policy Institute (EPI). 2026b. Holding the Line: State solutions to the U.S. Worker Rights Crisis.

Economic Policy Institute (EPI). 2026c. “Minimum Wage Tracker.” Last Modified July 1, 2026. 

Fortin, Nicole, Thomas Lemieux, and Neil Lloyd. 2022. “Right-to-Work Laws, Unionization, and Wage Setting.” National Bureau of Economic Research Working Paper no. 30098, June 2022. http://www.nber.org/papers/w30098. 

Glynn, Sarah Jane. 2020. “The Rising Cost of Inaction on Work-Family Policies.” Center for American Progress, January 21, 2020.

Gould, Elise, and Will Kimball. 2015. “Right-to-Work” States Still Have Lower Wages. Economic Policy Institute, April 2015.

Gould, Elise, and Heidi Shierholz. 2011. The Compensation Penalty of “Right-to-Work” Laws. Economic Policy Institute, February 2011. 

GovDocs. 2026. “Pay Transparency Laws by State and Province.” Last modified July 2026.

Hertel-Fernandez, Alexander, and Alix Gould-Werth. 2020. Labor Organizations and Unemployment Insurance: A Virtuous Circle Supporting U.S. Workers’ Voices and Reducing Disparities in Benefits. Washington Center for Equitable Growth, October 2020.

Jackson, C. Kirabo, Rucker C. Johnson, and Claudia Persico. 2016. “The Effects of School Spending on Educational and Economic Outcomes: Evidence from School Finance Reforms.” The Quarterly Journal of Economics 131, no. 1: 157–218. https://doi.org/10.1093/qje/qjv036.

KFF. 2026. “Status of State Medicaid Expansion Decisions.” Last modified May 21, 2026. 

McNicholas, Celine, Margaret Poydock, and Lynn Rhinehart. 2021. “How the PRO Act restores workers’ right to unionize” (fact sheet). Economic Policy Institute, February 4, 2021. 

McNicholas, Celine, Margaret Poydock, Heidi Shierholz, and Hilary Wething. 2025. Unions Aren’t Just Good for Workers—They Also Benefit Communities and Democracy. Economic Policy Institute, August 2025.

National Center for Education Statistics, National Public Education Financial Survey (NCES-NPEFS). 2024. Public data series accessed via the Common Core of Data database. Accessed July 1, 2026. 

Rinz, Kevin, and David Wasser. 2026. “Unemployment Insurance Generosity and Wage Determination.” Federal Reserve Bank of Cleveland Working Paper no. 26-11, May 2026. https://doi.org/10.26509/frbc-wp-202611.

Shedge, Nirav. 2025. Pay Transparency Laws and Wage Dynamics: Evidence from Colorado. Social Science Research Network, November 2025. 

Sherer, Jennifer. 2026. Rights to Unionize and Collectively Bargain: State Solutions to the U.S. Workers Rights Crisis. Economic Policy Institute, February 2026.

Sherer, Jennifer, Emma Cohn, and Ruby Ahdoot. 2025. “Updated EPI Tracker Shows More States Obstructing Progress on Workers’ Rights: Harmful Preemption Laws are Increasing Inequality and Repressing Democracy.” Working Economics Blog (Economic Policy Institute), March 6, 2025.

Sherer, Jennifer, Nina Mast, Elise Gould, and Emma Cohn. 2026. Everything You Need to Know About Anti-Union “Right-to-Work” Laws. Economic Policy Institute, forthcoming 2026.

Shierholz, Heidi. 2024. “Banning Noncompete Agreements.” Testimony before the U.S. Senate Banking, Housing, and Urban Affairs Subcommittee on Economic Policy, July 30, 2024. 

Shierholz, Heidi, Celine McNicholas, Josh Bivens, Jennifer Sherer, Ben Zipperer, and Margaret Poydock. 2026. The Case for Tripling Union Membership: How Rebuilding Union Power Would Strengthen Workers, the Economy, and Our Democracy. Economic Policy Institute, July 2026. 

Starr, Evan. 2019. “Consider This: Training, Wages and the Enforceability of Covenants Not to Compete.” ILR Review 72, no. 4 (August): 783–817. https://doi.org/10.1177/0019793919826060.

Wielk, Emily. 2026. “State Paid Family Leave Laws Across the U.S.” Bipartisan Policy Center. Last updated April 23, 2026. 

Zipperer, Ben. 2026. Setting High Standards for a Federal Minimum Wage. Economic Policy Institute, May 2026. 

Key takeaways

• The label “right to work” is intentionally misleading. RTW laws provide no job protections. What they do is make it harder for workers to form and sustain strong unions.
• RTW laws are intended to weaken unions. They are linked to lower unionization rates, lower wages and benefits, and worse safety outcomes for all workers.
• RTW laws have racist, anti-union origins dating to the 1940s. For decades, big business and white supremacist groups promoted RTW laws to diminish workers’ collective bargaining rights and block multiracial organizing.
• RTW laws don’t boost job growth but do erode job quality and increase inequality.
• Trends on RTW are reversing. No state has adopted a new RTW law since 2017, and multiple states have repealed or rejected RTW.

This report updates a two-decade body of research tracking the teacher pay penalty—the growing pay gap between public school teachers and their college-educated peers.1 Because public school teachers must attain at least a bachelor’s degree to teach in the U.S., this research compares weekly earnings of public school teachers (elementary, middle, and secondary)2 with those of college graduates who chose other careers. Documenting the widening divergence between the wages of teachers and their college-educated counterparts over time allows for a historical analysis of an issue that is critical to the future of the United States: the quality of our education. If teachers’ compensation doesn’t catch up with that of similarly educated and experienced professionals, schools will struggle to retain and attract qualified workers into the profession.

Data and relevant information

In analyzing differences in pay between public school teachers and other college graduates, I use two sources of data, both from the Bureau of Labor Statistics (BLS).3 First, I use Current Population Survey Outgoing Rotation Groups (CPS-ORG) data for the weekly wage analyses (BLS 2025a). I focus on weekly wages, rather than weekly hours worked or the length of the work year, to account for the “summers off” issue that affects teachers but not other college graduates.4 The sample is restricted to full-time workers (working at least 35 hours per week) aged between 18 and 64, with at least a bachelor’s degree, since teachers today need at least a bachelor’s degree to teach. Note that the 2025 annual data do not include October due to the U.S. federal government shutdown.

The sample is further limited to those who reported their wage information directly (those who didn’t respond and whose wages were estimated by BLS are excluded).5 To preserve data confidentiality, the BLS records weekly wages only up to a defined threshold, so the wage amounts above this threshold aren’t specifically identifiable in the data. This is called top-coding. Historically, the threshold was rarely updated. As a result, a growing share of workers are assigned top-coded wages that are below their actual wages, which has generated a growing understatement of college graduate wages relative to those of teachers. EPI replaces original top-coded values with Pareto-distribution implied means above the original CPS top-code separately for men and women.6 CPS demographic variables (e.g., gender, race/ethnicity, state of residence, marital status, age) are also used for the regression analyses.

The BLS’s National Compensation Survey’s Employer Costs for Employee Compensation program (BLS 2025b) is the second data source. Specifically, I pull data on employer costs per hour worked for detailed categories of compensation for “primary, secondary, and special education school teachers” in the public sector, and the same data for “civilian professionals,” which is the broadest category available that largely corresponds to college graduates. “Benefits,” in this analysis, refer to employer costs for health and life insurance, retirement plans, and payroll taxes (covering Social Security, unemployment insurance, and workers’ compensation).

The remaining components of compensation are “W-2 wages,” a measure that corresponds to the wages captured in the CPS data used above. W-2 wages are the wages reported to employees and to the Internal Revenue Service. They include “direct wages,” defined by the BLS as “regular payments from the employer to the employee as compensation for straight-time hourly work, or for any salaried work performed,” and other wage items, including “supplemental pay.” Supplemental pay includes premium pay for overtime, bonus pay, profit-sharing, and paid leave.

Findings

I present the results of this research in four sections. First, I examine trends in the simple (i.e., not regression-adjusted, but adjusted for inflation) average weekly wages for public school teachers and other college graduates from 1979 through 2025. Second, I report annual estimates of the relative national teacher weekly wage gap. The relative gap is estimated using standard regression techniques to control for systematic differences in age, education, state of residence, and other factors known to affect wage rates. Third, I analyze the regression-adjusted estimates of the teacher wage gap for each state and the District of Columbia. Lastly, I factor in nonwage benefits to estimate a total compensation penalty that accounts for the estimated teacher wage penalty, along with the teacher “benefits advantage”—teachers’ relatively larger share of compensation coming from benefits, like health insurance or retirement plans. I use these figures to estimate a total compensation differential at the national level (which is not possible to calculate for each state).

Simple level differences: Weekly wage trends

The average weekly wages of public school teachers and other college graduates are shown in Figure A. These data are national annual averages adjusted only for inflation (i.e., not regression-adjusted). It is important to keep in mind that real improvements in living standards require wages to outpace inflation, which has not been the case for teachers—but has for other college graduates. 

Figure A

Figure A shows that the inflation-adjusted average weekly wages for teachers were relatively flat from 1996 through 2021, indicating that teacher wages, on average, were just keeping up with the rate of inflation. By 2025, teacher wages were 6.2% less than they were on average in 1996. The average weekly wages of other college graduates also experienced a stretch of stagnation, but for a shorter time span (2002–2014), after which real increases ensued. Since 1996, the wages of other college graduates increased by 28.8%.

Illustrated in Figure A is a noteworthy fall in teacher wages that occurred in 2022 due to high rates of inflation. A similar decline did not occur in the wages of other college graduates as private-sector wages can be more responsive to economic conditions, whereas that is not possible with public-sector teacher pay. At the time, I posited that a new, lower trend in teacher wages would likely occur post-2022. History has shown that teacher pay has barely kept up with moderate rates of inflation—as indicated by the flat trend in Figure A—and pay increases following 2022 seemed unlikely to be large enough to recoup that significant decline. Thus far, they have not.

Addressing the long-term stagnation of teacher wages requires that future increases in pay meet and exceed rates of inflation in the future to recover the decline in wages since 2021, and to drive an increasing trend in teacher wages to close the gap relative to other college graduates.

Relative differences: Regression-adjusted trends

The discussion of weekly wages shown in Figure A is centered around simple averages—meaning, they are not adjusted for fundamental differences that may exist between the samples of teachers and other college graduates. In this section, I move to regression adjusted estimation, which helps to account for ways the two groups may differ fundamentally on factors that typically affect pay—such as age, educational attainment, race/ethnicity, and state of residence. For instance, all else being equal, one would expect experienced workers to earn more than younger workers who are just starting out in their careers. Controlling for age within a regression model accounts for such differences across the two samples. Thus, I employ standard regression techniques to estimate weekly wages of public school teachers relative to other similarly situated college graduates working in other professions, which can provide a more apples-to-apples comparison of earnings.7

Figure B reports regression-based results that show how much less (or more) teachers earn in weekly wages relative to other college graduates. A weekly wage “penalty” for teachers is reported when the regression estimates suggest that teachers, all else equal, are paid less than other college graduates. Teachers’ weekly wage penalties are reported as negative numbers in Figure B. When teachers are paid relatively more, the reported gap is positive and is referred to as a “premium.” The regression analysis provides estimates for all teachers (which includes a gender control), as well as separately for women and men. 

Figure B

The historical trends in relative teacher pay are depicted in Figure B. The overall trend represents three decades of teacher wages faring ever more poorly compared with other similarly qualified professionals. In the pre-1994 period, the teacher wage gap averaged 8.7%, but the relative gap worsened considerably starting in the mid-1990s. The teaching penalty hit a record of 26.9% in 2024, before slightly improving to an estimated 25.2% in 2025. However, this slight improvement is unlikely to be sustained, given that similar improvements throughout this long series have historically proven temporary.

In sum, teachers earned on average 74.8 cents on the dollar in 2025 compared with what similar college graduates earned working in other professions—much less than the relative 93.9 cents on the dollar that teachers earned in 1996.

Separating the analysis by gender shows that in the period from 1979 through 1993, the relative female teacher weekly wage (i.e., comparing female teachers with other female college graduates) was at a premium that averaged 3.3%. But starting in 1996, the female teacher wage gap quickly went from a small premium to an ever-increasing penalty—landing at a 20.0% penalty in 2025.

My previous research (using decennial Census data) confirmed that, over a longer timeframe, the relative wage estimates for female teachers moved from significant premiums to large penalties. For example, I documented that relative female teacher earnings were at a 14.7% premium in 1960, which lessened to 10.4% in 1970 and to near parity in 1980 (pre-1979 years not shown in Figure B). Using the estimates from 2025, the cumulative change has been a 34.7 percentage-point deterioration in the relative wage of female teachers since 1960.8

There is an important story behind the declining relative wages experienced by female teachers. Historically, the teaching profession relied on a somewhat captive labor pool of educated women who had few employment opportunities. This is thankfully no longer the case, but increased opportunity costs are a part of the story, and they are reflected in these results. Expanding opportunities for women enabled them to earn more as they entered occupations and professions from which they were once barred.

The wages of male teachers have long been outpaced by those garnered by male college graduates who chose other professions. My previously documented estimates going back to 1960 have never been even close to parity. But similar to their female counterparts, the relative male teacher penalty grew considerably over time—from 20.5% in 1960 to 34.5% in 2025.9

The historically large and worsening male teacher penalty partly explains why about 3 in 4 teachers today are women—a ratio that has not changed much since 1960. The pay penalty experienced by male teachers is unfortunate given the recent statistics and reporting of boys struggling in school. Performing poorly in school is associated with problems encountered later in life—including addiction, mental and physical health issues, and involvement with the criminal justice system.10 Further, Thomas Dee (2010) found that a teacher’s gender has large effects on student test performance, teacher perceptions of students, and students’ engagement with academic material.

It shouldn’t be surprising, then, that today a much smaller share of college-educated women choose the teaching profession over expanding opportunities with better pay—even as three-quarters of teachers are women. Moreover, the very large, persisting male teaching penalty goes a long way in explaining why men who may want to teach are compelled to choose other career paths, which are on average much more lucrative.

Relative teacher weekly wage penalties by state

To this point, I have reported the relative teacher wage gaps for the country as a whole, but there is a lot of state-by-state variation. To produce regression estimates by state, I pool six years (2020–2025) of CPS data to assure ample sample sizes for each state. Again, I compare public school teachers with nonteacher college graduates within each state and estimate regression-adjusted weekly wage gaps for each state and the District of Columbia.

Figure C

As in previous reports, Figure C shows that in no state does the relative (i.e., regression-adjusted) weekly wage for teachers equal or surpass that of their nonteaching college graduate counterparts. The results in Figure C are sorted from the largest (40.7%) to the smallest (10.4%) penalties across the U.S.

The teaching penalty was at least 25% in 25 states, and at least 30% in 11 states. Colorado’s 40.7% teacher penalty is the first time in this research series that a penalty was estimated at or above 40%; on average, teachers in Colorado earned 59.3 cents on the dollar compared with similar college graduates in 2025.

The map in Figure D illustrates the state penalties reported in Figure C.

Figure D

Adding benefits to the analysis

This section adds a nationwide analysis of benefits to see how they affect total compensation between teachers and other college graduates. Teachers are thought to have a “benefits advantage,” which refers to the understanding that, on average in the U.S., teachers generally receive a larger share of their total compensation in benefits—such as health insurance, other insurance, and retirement plans—compared with employment benefits received by other professionals.

Keep in mind that a larger share of total compensation via benefits means a smaller wage share, given that total compensation is made up of these two components. Here, I calculate how the relatively more generous benefits package for teachers partially offsets the large teacher wage penalty.

The BLS Employer Costs for Employee Compensation (ECEC) series measures the average employer cost per employee hour worked for total compensation, wages and salaries, benefits, and costs as a share of total compensation. I compare benefits packages of primary, secondary, and special education public school teachers with those of comparable workers (specifically, workers in professional occupations).11 Table 1 shows a summary of my calculations.

Table 1

The first two columns in Table 1 under “W-2 wage share of compensation” report the share of W-2 wages that make up total compensation for professionals in all occupations and for state and local K–12 public school teachers. The shares of compensation for W-2 wages and benefits add up to 100. The W-2 shares allow for an examination of how important wages are relative to benefits in the total compensation package.

In 2025, W-2 wages made up 69.4% of teachers’ total compensation, while the share was 79.3% for nonteaching professionals. That means that for every dollar of teachers’ total compensation, 69.4 cents went to wages, and 30.6 cents went to benefits. For professionals, 79.3 cents went to wages, and 20.7 cents went to benefits. Therefore, for every dollar of total compensation, public school teachers receive more in benefits than other professionals, but less in wages. I refer to this as the “benefits advantage.”12

The right-side columns in Table 1 provide the information needed to assess total compensation on average for the United States from the “public school teachers” vantage point. The “wage penalty” column reports the teacher wage penalty estimates from Figure B, followed by the benefits advantage calculation for teachers. Combining the two gives us a measure of how teachers compare with other professionals on total compensation, which is reported in the last column. Per usual, the benefits advantage for teachers partially offsets their estimated relative wage disadvantage, but teachers are still left with a significant total compensation gap of 14.5% in 2025—down somewhat from 16.7% in 2024. This slight change was due to a 0.5 percentage point relative increase in the teacher benefits advantage, and a 1.7 percentage point decrease in the relative teacher wage penalty.

As reported in Table 1, the teacher total compensation gap over the last decade has varied in a rather narrow range from 10.2% in 2019 to a high of 17.0% in 2022. Generally, over the long run, the teacher wage penalty has been worsening as the teacher benefits advantage has grown larger—albeit at a slower rate than the wage penalty. Of course, even if the teacher benefits advantage exceeded the large teacher wage penalty, the standard of living for teachers would likely fall as they would have little in the way of earnings to make ends meet.

How much further can teacher pay fall behind?

Teaching is one of the most consequential professions—not just for kids and their families, but for the country as a whole. Educating each and every child, regardless of means, is among our highest ideals as a country, and the future of the U.S. economy depends on it. The highest standard is still worth fighting for, even as we have always fallen short of the ideal.13

Given the critical role that teachers have on student outcomes (Chetty, Friedman, and Rockoff 2014), I have been updating this series on teacher pay for over 20 years. After Larry Mishel, Sean Corcoran, and I wrote two books on the subject in 2004 and 2008, it was not obvious that the trends we depicted would continue to get worse—but overall, they have.

The nearly 50-year trends documented in this paper have no doubt already taken a serious toll on the teaching profession. That toll shows up in teacher retention, recruitment, and staffing (Fortin and Fawcett 2023; NCES 2023); college students forgoing teaching careers citing pay as a main barrier (Croft, Guffy, and Vitale 2018); parents actively steering their children into professions that pay better than teaching (PDK 2019); fast-tracking credentials in response to permanent teacher shortages (Povich 2023); the heavy use of unqualified teachers (Tamez-Robledo 2023; Lopez and Van Overschelde 2024); and the reliance on unqualified substitute teachers (Franco and Kemper Patrick 2023).

I have long asserted that providing teachers a standard of living commensurate with that of similar nonteacher professionals is not simply a matter of fairness—teacher pay is a central issue in public education, and thus a matter of national consequence. The quality of a public education hinges on our willingness to fairly and sufficiently invest in both schools and teachers, including the full public school workforce, its infrastructure, and essential wrap-around services (Wething and Bivens 2025).

It would be willful blindness to ignore the toll that the decades-long teacher pay penalty has had on the quality of public education. Regrettably, policymakers have failed to deliver sustained and effective interventions capable of mitigating—let alone substantially improving—the trends outlined in this long-running series. This is an unfortunate reality, especially in a country as wealthy as the United States, with more than enough resources for its public schools to be the envy of the world.

Notes

1. See Allegretto, Corcoran, and Mishel 2004, 2008; Allegretto and Tojerow 2014; Allegretto and Mishel 2016, 2018, 2019; and Allegretto 2023 and 2024.

2. The teacher sample does not include kindergarten or pre-kindergarten; if included, the teacher pay penalties would be even larger. 

3. Allegretto and Mishel 2019, Appendix A provides a comprehensive discussion of the data and methodologies that were used to produce our teacher weekly wage and total compensation estimates. 

4. In Allegretto and Mishel 2019, we provide evidence that teachers work weekly hours similar to those of other professionals.

5. Our earlier work documents that BLS’s imputation method overstates teacher earnings, which is not the case for the other college graduate sample (Allegretto, Corcoran, and Mishel 2008, 9).

6. For more about top-code adjustments, see Economic Policy Institute 2026b.

7. The wage model includes controls for both public and private school teachers. The weekly wage penalty estimates are based on the coefficient on the public school teacher indicator. Regression for all teachers includes a gender control. The percentage gap is calculated as (eb -1) x 100. See Allegretto and Mishel 2019, Appendix A, for specification details.

8. See Allegretto, Corcoran, and Mishel 2008 for 1960, 1970, and 1980 estimates using decennial censuses.

9. The 1960 results are not shown in Figure B. They can be found in Allegretto, Corcoran, and Mishel 2008, 7.

10. See Abrams 2023.

11. The ECEC provides compensation data for a narrower category of “primary, secondary, and special education school teachers” and for a broader category of “teachers.” I analyze the narrower category, which closely matches the definition of teachers in the CPS-ORG data, using data limited to state and local public-sector workers. The inclusion of kindergarten and special education teachers in the benefits analysis does not produce any more substantial differences than if they were excluded (as they are in the CPS sample used to estimate the wage penalty). Greater methodological detail is provided in Appendix A of Allegretto and Mishel 2019.

12. My analysis accounts for differences in annual weeks worked, as it is based on the usual weekly wages of teachers and other college graduates, not hourly wages or annual earnings. One reason health and pension costs are higher for teachers is that teacher health benefits are provided for a full year, while teacher salaries are for less than a full year.

13. See Allegretto, Garcia, and Weiss 2022. This paper describes inequities in public education funding. We also argue that the federal government should play a larger role in funding public education.

References

Abrams, Zara. 2023. Boys Are Facing Key Challenges in School. Inside the Effort to Support Their Success. American Psychological Association, April 2023.

Allegretto, Sylvia A., Sean P. Corcoran, and Lawrence Mishel. 2004. How Does Teacher Pay Compare? Methodological Challenges and Answers. Washington, D.C.: Economic Policy Institute.

Allegretto, Sylvia A., Sean P. Corcoran, and Lawrence Mishel. 2008. The Teaching Penalty: Teacher Pay Losing Ground. Washington, D.C.: Economic Policy Institute.

Allegretto, Sylvia A., Emma García, and Elaine Weiss. 2022. Public Education Funding in the U.S. Needs an Overhaul: How a Larger Federal Role Would Boost Equity and Shield Children from Disinvestment During Downturns. Economic Policy Institute, July 2022.

Allegretto, Sylvia A., and Lawrence Mishel. 2019. The Teacher Weekly Wage Penalty Hit 21.4 Percent in 2018, a Record High. Economic Policy Institute, April 2019.

Allegretto, Sylvia A., and Ilan Tojerow. 2014. Teacher Staffing and Pay Differences: Public and Private Schools. Monthly Labor Review. U.S. Department of Labor, Bureau of Labor Statistics, September 2014.

Blad, Evie. 2024. Teachers Report Lower Pay, More Stress Than Workers in Other Fields. Education Week, June 19, 2024, sec. Teaching & Learning, Teaching Profession.

Bureau of Labor Statistics (BLS). 2025a. Current Population Survey.

Bureau of Labor Statistics (BLS). 2025b. Employer Costs for Employee Compensation Historical Listing: National Compensation Survey, data tables. Accessed July 12, 2026.

Chetty, Raj, John N. Friedman, and Jonah E. Rockoff. 2014. Measuring the Impacts of Teachers II: Teacher Value-Added and Student Outcomes in Adulthood. American Economic Review 104, no. 9 (September 2014): 2633–2679.

Croft, Michelle, Gretchen Guffy, and Dan Vitale. 2018. Encouraging More High School Students to Consider Teaching. ACT Research & Policy, June 2018.

Dee, Thomas S. 2010 The Why Chromosome. Education Next, January 26, 2010.

Economic Policy Institute (EPI). 2026a. Current Population Survey Extracts, Version 2026.7.8, https://microdata.epi.org. Accessed July 6, 2026.

Economic Policy Institute (EPI). 2026b. “Methodology: Wage Variables.” EPI Microdata Extracts documentation.

Fortin, Jacey, and Eliza Fawcett. 2023. How Bad Is the Teacher Shortage? Depends Where You Live. New York Times, August 29, 2023.

Franco, Marguerite, and Susan Kemper Patrick. 2023. State Teacher Shortages: Teaching Positions Left Vacant or Filled by Teachers Without Full Certification. Learning Policy Institute, July 2023.

Lopez, Minda, and James P. Van Overschelde. 2024. Unlicensed Teachers Now Dominate New Teacher Hires in Rural Texas Schools. The Dallas Observer, June 20, 2024.

Merod, Anna. 2023. Low Pay, Long Hours Top Reasons Teachers Consider Leaving. K-12 Dive, September 12, 2023.

National Center for Education Statistics (NCES). 2023. Most Public Schools Face Challenges in Hiring Teachers and Other Personnel Entering the 2023-24 Academic Year. October 2023.

Phi Delta Kappan (PDK). 2018. Teaching: Respect but Dwindling Appeal. The 50th Annual PDK Poll of the Public’s Attitudes Toward the Public Schools. Supplement to Kappan magazine.

Povich, Elaine S. 2023. Plagued By Teacher Shortages, Some States Turn to Fast-Track Credentialing. Stateline, July 24, 2023.

Steiner, Elizabeth D., Ashley Woo, and Sy Doan. 2023. All Work and No Pay — Teachers’ Perceptions of Their Pay and Hours Worked: Findings from the 2023 State of the American Teacher Survey. RAND Corporation, September 12, 2023.

Tamez-Robledo, Nadia. 2023. These States Have the Most ‘Underqualified’ Teachers Stepping in to Fill Open Positions. EdSurge, April 4, 2023.

Wething, Hilary and Josh Bivens. 2025. U.S. Investment in Public Education is at Risk: Vouchers, State Budget Austerity, and Federal Attacks on the Department of Education Threaten Children’s Futures. Economic Policy Institute, August 2025. 

Raymond Windmiller, Executive Officer
Executive Secretariat
U.S. Equal Employment Opportunity Commission
131 M Street, NE
Washington, DC 20507

Re: Proposed Rule for Removal of Reporting Requirements (RIN 3046-AB37)

Dear Commissioners,

I write to submit this comment on behalf of the Economic Policy Institute (EPI), responding to the Equal Employment Opportunity Commission’s proposed rule to rescind and remove the requirements for filing the EEO-1, EEO-2, EEO-3, EEO-4, EEO-5, and EEO-6 reports1, and the recordkeeping and record preservation requirements related to these reports, under 29 CFR part 1602.

EPI is a nonprofit, nonpartisan think tank created in 1986 to include the needs of low- and middle-income workers in economic policy discussions. EPI conducts research and analysis on the economic status of working America, proposes public policies that protect and improve the economic conditions of low- and middle-income workers, and assesses policies with respect to how well they further those goals. For 18 years, EPI’s Program on Race, Ethnicity and the Economy (PREE) has been a nationally recognized source for expert analysis of the many ways that race, ethnicity, gender, class and policy decisions intersect to affect economic outcomes.

We strongly oppose the EEOC’s proposal to rescind the EEO reporting, recordkeeping and record preservation requirements. If implemented, this rule would undermine the Commission’s capacity to enforce federal laws against employment discrimination. Furthermore, this rule would deprive employers, researchers, policymakers and advocates of their ability to independently analyze and use this unique data source to develop and advocate for policies that support the EEOC’s mission of ending workplace discrimination.

The U.S. Equal Employment Opportunity Commission (EEOC) is a cornerstone in upholding the civil rights of U.S. workers. Established under the same law that officially prohibited employment discrimination (Title VII of the Civil Rights Act of 1964), the EEOC embodies the fact that effective federal laws against employment discrimination require consistent enforcement. For the last 60 years, EEO reports have been central to the agency’s capacity to enforce the law and to monitor and track the nation’s progress toward ending workplace discrimination.

That goal is still a work in progress, as documented in an extensive body of methodologically rigorous research by economists, sociologists and other social scientists. In what follows, we explain why the Commission’s proposal to rescind all EEO data collection would be counterproductive to reaching that goal and detrimental to the U.S. economy by:

• Summarizing the research on labor market discrimination that justifies the ongoing need for EEO reporting requirements;
• Describing how broad EEO reporting requirements help to rebalance power and information asymmetries between employers and employees; and
• Presenting estimates of how much discrimination costs the U.S. economy in economic output and average living standards.

Research and national data provide compelling evidence of persistent labor market discrimination against “minorities”

In their proposed rule, the Commission claims that EEO Reports “may encourage employers to discriminate against employees who are not considered ‘minorities,’ may promote racial stereotyping, and may encourage employers to engage in discrimination to avoid potential EEOC enforcement actions or to address perceived inequitable outcomes.” In addition to this being highly speculative, it is also inconsistent with what we know from research and national statistics. Large and persistent disparities in employment and pay of Black workers relative to white workers are among the most durable features of the U.S. labor market. Compelling empirical evidence points to discrimination as a significant factor in the persistence of those disparities.

EPI’s detailed expert analysis of Bureau of Labor Statistics (BLS) data shows that the significant racial disparities in unemployment that are observed at each level of education, across age cohorts, and among both men and women strongly suggest that education or skills differentials alone can’t account for the near constant 2-to-1 Black-white unemployment gap2.

These findings are consistent with field experiments revealing that Black job applicants with equivalent, and sometimes superior, credentials to white applicants are less likely to receive job callbacks3. One of the starkest audit study findings in this regard concludes that employers treated white applicants with criminal records more favorably than Black applicants without criminal records4. Researchers also found that when résumés of Black and Asian job applicants were stripped of clues of their racial identity, they received callbacks at a much higher rate than those that clearly indicated an applicant’s racial identity5.

While audit and correspondence studies have been criticized for not adequately capturing unobserved characteristics that might influence hiring decisions, newer studies have shown how robust these findings can be to such considerations6. In fact, subsequent field experiments reveal a pattern of hiring discrimination experienced by Black and Hispanic job seekers that remained remarkably constant over time, with white applicants receiving 36% more callbacks than Black applicants and 24% more callbacks than Hispanic applicants7.

EPI’s analysis has also shown that since 1979, less than half of the observed black-white difference in average hourly wages can be explained by differences in education, experience, or region—the main factors presumed to determine pay – and discrimination has consistently played a major role8. However, changes in this racial wage gap track closely with changes in policy, including civil rights enforcement, and with structural factors that affect wage inequality. According to trend analysis research, the narrowing of the gap from the late 1960s through the 1970s can be attributed to the passage of important civil rights legislation9, combined with the 1960s economic boom, active enforcement of anti-discrimination10 and affirmative action policy11, and the narrowing of the educational attainment gap between Black and white workers12. On the other hand, retrenchment on anti-discrimination policy was found to be a significant factor in the widening of the gap during the 1980s13. Since 2000, research has shown a troubling trend of larger discriminatory differentials among the more highly educated14.

Contrary to the Commission’s proposed justification for ending EEO reporting requirements, the research literature cited above offers no evidence of systemic discrimination “against employees who are not considered minorities”. The research, however, underscores the need for ongoing monitoring of employment and pay discrimination through the collection of EEO data.

Elimination of EEO data would worsen power and information asymmetries that undermine worker protections

In an essay published under EPI’s Unequal Power project, former EEOC Chair Jenny Yang explains that “in a system that places the primary responsibility for enforcing anti-discrimination laws on individual workers, who must file complaints with their employer or a government agency, the government plays a fundamental role in rebalancing the power disparities between workers and employers.”15 She further concludes that “policies that encourage employer transparency and require data collection to support prevention and accountability are essential to addressing the inherent information asymmetry workers face.”16

EEO data helps to provide the necessary transparency and accountability across major segments of the U.S. labor market, including private employers with 100 or more employees and federal contractors with 50 or more employees (EEO-1), apprenticeship programs (EEO-2), labor unions (EEO-3), state and local governments (EEO-4), elementary and secondary schools (EEO-5) and institutions of higher education (EEO-6)17.

While the Commission now argues that reporting requirements are too broad because they include non-discriminating employers along with those accused of discrimination, that feature is actually a strength of EEO data. As the nation’s only federally collected source of firm-level information on employment by race, ethnicity, sex, and job category, the EEO-1 provides EEOC investigators with the information they need to assess each individual claim within the context of a larger group of employers.

In addition to being used in this important first step of the charge intake process, these data can also identify potential patterns of systemic discrimination within an industry, occupation, or worksite location. This information helps to direct EEOC systemic enforcement and commissioner’s charges, enabling the EEOC to investigate and address discriminatory practices in cases where workers either lack information or fear retaliation for filing an individual charge of discrimination. Many state and local government Fair Employment Practices Agencies (FEPAs) also have data sharing agreements with EEOC, allowing them to utilize EEO data for local enforcement activities. Similar data collections are too costly for most FEPAs to undertake on their own. As such, EEO data address information and power asymmetries that make it difficult and costly for workers to identify and prove discrimination.

Elimination of EEO data would limit EEOC’s ability to support economic growth through reduced discrimination

Finally, the Commission claims that elimination of the EEO surveys is consistent with the policy underlying E.O. 14192, “Unleashing Prosperity Through Deregulation,” requiring agencies to be “prudent and financially responsible in the expenditure of funds . . . to alleviate unnecessary regulatory burdens placed on the American people.” However, the Commission is overlooking the cost discrimination imposes on the economy and the long-term consequences ending EEO data collection would have on the EEOC’s ability to reduce discrimination through the enforcement mechanisms described above.

Research shows that since the 1960s, the decline in workplace discrimination has contributed to a stronger economy and more inclusive labor force. A widely cited 2019 report concludes that reduced discrimination alone accounted for almost 8% of GDP per capita growth between 1960 and 201018. Extrapolating this estimate through 2024 and applying it to published national estimates of real GDP per capita, we estimate that reduced discrimination boosted average living standards by $4,932 per person since 196019. While it is difficult to precisely measure the EEOC’s impact apart from other forces, assuming the Commission’s role in reducing discrimination accounts for 10–25% of that growth, we estimate that EEOC enforcement, aided by EEO data, helped to boost average living standards by $493 to $1,233 per person since 1960. This estimated increase in average living standards applies to all people, not just those with proven claims of discrimination. Elimination of EEO data would compromise many of the EEOC’s tools for enforcing laws against employment discrimination, and as a result, undermine economic growth.

Conclusion

For years, the EEOC has been vastly under-resourced relative to the magnitude of its enforcement responsibilities, and EEO data is essential to directing those limited resources. Changes to the categories and definitions used in EEO forms over time represent the ways in which Commissioners have grappled with how to accurately capture individual markers of identity useful in detecting unfair influence over employment opportunities for members of legally protected classes.

Under prior administrations, EEOC has convened expert panels to evaluate EEO data collections. These panels were comprised of statisticians, economists, sociologists, legal scholars, and other practitioners with decades of expertise in statistical analysis, labor and civil rights law and the study of labor market discrimination. I have had the privilege of serving on two of these panels. In both instances, panel members expressed differing opinions with respect to data quality, what conclusions could be reliably drawn from the data, and how to best improve the data collection process. Undoubtedly, there were also likely differences in panel members’ political views, but I don’t ever recall any assertion that the problem discrimination had been sufficiently reduced to justify removal of EEO reporting requirements.

In conclusion, EPI strongly urges the Commission to withdraw its proposal to eliminate all six of the EEO surveys. Failure to do so would renege on the promise of our nation’s anti-discrimination laws for workers, their families and the economy.

Sincerely,

Valerie Wilson, PhD
Director, Program on Race, Ethnicity and the Economy
Economic Policy Institute

1. The EEO-2 remains authorized under Title VII of the Civil Rights Act of 1964 but has not been collected since 1981. In this case, our recommendation to continue the surveys is a recommendation to reinstate it.

 The EEO-6 remains authorized under Title VII but has not been collected since 1993. In this case, our recommendation to continue the surveys is a recommendation to retain this authorization and reinstate the survey to the extent that it is not superseded by the Integrated Postsecondary Education Data System (IPEDS) Staff Survey of the U.S. Department of Education.

2. Valerie Wilson and William Darity Jr., Understanding black-white disparities in labor market outcomes requires models that account for persistent discrimination and unequal bargaining power, Economic Policy Institute, March 2022.

3. Examples include: Michael Fix, George C. Galster, and Raymond J. Struyk, “An Overview of Auditing for Discrimination.” in Michael Fix and Raymond Struyk, eds., Clear and Convincing Evidence: Measurement of Discrimination in America, Urban Institute Press, 1993; Marc Bendick, Jr., Charles W. Jackson, and Victor Reinoso, “Measuring Employment Discrimination Through Controlled Experiments” in James B. Stewart, ed., African-Americans and Post-Industrial Labor Markets, Transaction Publishers, 1994; Michael Fix, George C. Galster, and Raymond J. Struyk, “An Overview of Auditing for Discrimination” in Michael Fix and Raymond Struyk, eds., Clear and Convincing Evidence: Measurement of Discrimination in America. Urban Institute Press, 1993; and Margery Turner, Michael Fix, and Raymond Struyk, Opportunities Denied, Opportunities Diminished: Racial Discrimination in Hiring, Urban Institute Press, 1991.

4. Devah Pager, “The Mark of a Criminal Record,” American Journal of Sociology 108 (March 2003): 937–75.

5. Sonia Kang, Katy DeCellesa, András Tilcsika, and Sora Jun, “Whitened Résumés: Race and Self-Presentation in the Labor Market,” Administrative Science Quarterly 61, no. 3 (March 17, 2016): 469-502.

6. David Neumark, “Detecting Discrimination in Audit and Correspondence Studies,” Journal of Human Resources 47 (Fall 2012): 1128–57. 

7. Lincoln Quillian, Devah Pager, Ole Hexel, and Arnfinn H. Midtbøen, “Meta-Analysis of Field Experiments Shows No Change in Racial Discrimination in Hiring Over Time,” Proceedings of the National Academy of Sciences 114, no. 41 (October 10, 2017): 10870-875.

8. Valerie Wilson and William Darity Jr., Understanding black-white disparities in labor market outcomes requires models that account for persistent discrimination and unequal bargaining power, Economic Policy Institute, March 2022.

9. Examples include: John Bound and Richard Freeman, “Black Economic Progress: Erosion of the Post-1965 Gains in the 1980s?” in Steven Shulman and William Darity, Jr., eds., Question of Discrimination: Racial Inequality in the U.S. Labor Market, Wesleyan University Press, 1989; David Card and Alan Krueger, “School Quality and Black-White Relative Earnings: A Direct Assessment,” Quarterly Journal of Economics 107 (February 1992): 151–200; and John Donohue and James Heckman, “Continuous vs. Episodic Change: The Impact of Civil Rights Policy on the Economic Status of Blacks,” Journal of Economic Literature 29 (December 1991): 1603–43. 

10. Examples include: Augustin K. Fosu, “Occupational Mobility of Black Women, 1958–1981: The Impact of Post-1964 Antidiscrimination Measures,” Industrial & Labor Relations Review 45, no. 2 (1992): 281–94; and James J. Heckman and Brook Payner, “Determining the Impact of Federal Antidiscrimination Policy on the Economic Status of Blacks: A Study of South Carolina,” National Bureau of Economic Research Working Paper no. 2854, 1989. 

11. Examples include: Charles Betsey, “Litigation of Employment Discrimination Under Title VII: The Case of African American Women,” American Economic Review 84, no. 2 (1994): 98–102; and Jonathan S. Leonard, “The Impact of Affirmative Action Regulation and Equal Employment Law on Black Employment,” Journal of Economic Perspectives 4, no. 4 (1990): 47–63. 

12. Examples include: Leonard Carlson and Caroline Swartz, “The Earnings of Women and Ethnic Minorities, 1959–1979,” Industrial & Labor Relations Review 41, no. 4 (1988): 530–46; James S. Cunningham and Nadja Zalokar, “The Economic Progress of Black Women, 1940–1980: Occupational Distribution and Relative Wages,” Industrial & Labor Relations Review 45, no. 3 (1992): 540–55; and Nadja Zalokar, The Economic Status of Black Women: An Exploratory Investigation. U.S. Commission on Civil Rights, 1990.

13. Jonathan S. Leonard, “The Impact of Affirmative Action Regulation and Equal Employment Law on Black Employment,” Journal of Economic Perspectives 4, no. 4 (1990): 47–63. 

14. Examples include: Donald Tomaskovic-Devy, Melvin Thomas, and Kecia Johnson, “Race and the Accumulation of Human Capital Across the Career: A Theoretical Model and Fixed-Effects Application,” American Journal of Sociology 111, no. 1 (2005): 58–89; and Valerie Wilson and William M. Rodgers III, Black-White Wage Gaps Expand with Rising Wage Inequality, Economic Policy Institute, September 2016. 

15. Jenny R. Yang and Jane Liu, Strengthening Accountability for Discrimination: Confronting Fundamental Power Imbalances in the Employment Relationship, Economic Policy Institute, January 2021. 

16. Ibid.

17. See endnote 1.

18. Chang-Tai Hsieh, Erik Hurst, Charles I. Jones and Peter J. Klenow. “The Allocation of Talent and U.S. Economic Growth”, Econometrica, Vol. 87, No. 5 (September 2019), 1439-1474.

19. Adewale A. Maye and Valerie Wilson, Trump is Making it Easier for Employers to Discriminate. This Stifles Equity and Hurts Economic Growth. Economic Policy Institute, May 2025.