The Pay Gap Hides Three Different Questions
This series examines the U.S. nonprofit labor market and its pay relationship with for-profit and public-sector employers. The immediate question is whether nonprofit workers earn less because different kinds of workers choose nonprofit jobs, or because nonprofit employment itself is associated with a different level of pay. That distinction matters for anyone trying to understand nonprofit work, compare compensation, evaluate public-service capacity, or interpret a claim about a sector that employed millions of people. By the end of this opening, the useful starting point should be clear: a pay-gap headline is an observation that needs to be unpacked, not an explanation in its own right.
The number is real, but it is not yet an explanation
Nonprofit work is too large to treat as a niche curiosity. The Census working paper describes the sector as employing roughly 10 percent of the American workforce. A separate Bureau of Labor Statistics analysis, using Current Population Survey data and a wage-and-salary definition, puts the nonprofit share at 7.2 percent in 2022. Those figures are not a contradiction to be averaged away. They describe different statistical populations and different ways of identifying employment. The shared point is that the sector is large enough for its pay structure to affect workers, organizations, and the provision of services at national scale.
The comparison also has three sides. Most public discussion contrasts nonprofit and for-profit jobs, but the BLS analysis places public-sector employment beside both. That third benchmark matters because the BLS analysis finds nonprofit workers more similar to public-sector workers than to for-profit workers in education and occupation, while also showing that nonprofit work is concentrated in particular service industries. A nonprofit wage can therefore look high next to a private-sector average and low next to a public-sector average, before anyone has asked whether the jobs or workers are comparable.
The first temptation is to read the raw average as a verdict. The BLS article reports that nonprofit workers earned 5 to 13 percent more than for-profit workers in average hourly wages over 1994–2022 before worker characteristics were taken into account. After adjustment for observed characteristics, its estimate becomes a 4 to 7 percent nonprofit penalty relative to for-profit workers. The Census paper reports a different raw comparison: an unconditional nonprofit earnings penalty of 12 percent relative to for-profit workers, with most of that penalty interpreted as causal rather than as selection. The directions do not line up, and that is important evidence about the comparison itself.
It is also a reason for care. The Census analysis uses administrative tax records, annual earnings, and a defined sample of full-time, main-employer worker-years. The BLS analysis uses hourly wages for wage-and-salary workers in the CPS and applies its own exclusions, controls, and corrections. “Unconditional,” “adjusted,” and “selection-corrected” are not interchangeable labels. Each belongs to a particular outcome, population, period, and method. The disagreement does not erase either result; it tells us that the question “Do nonprofits pay less?” is incomplete until the comparison is specified.
There are at least three different questions hiding inside that incomplete sentence. The first is descriptive: what is the average earnings level among people classified as nonprofit workers, and how does it compare with the average in another sector? This is useful for describing the labor market, but it carries every difference in who works there and what work they do. The second asks about comparable observed workers: after accounting for education, experience, occupation, industry, hours, geography, and other measured characteristics, is the nonprofit worker paid differently? This is close to the question answered by the BLS adjusted estimates. The third asks about a worker–firm relationship that persists after using movements between employers to separate a worker component from a firm component. That is the question the Census job-ladder model is built to address.
The questions are related, but they do not share one denominator or one meaning of “same.” A full-time worker-year with annual tax earnings is not the same observation as a CPS worker whose hourly wage is calculated from weekly earnings and usual hours. A model that compares people with similar measured characteristics still differs from one that uses the same worker across connected employers. Even the size of the sector depends on the population: the BLS figure is a share of wage-and-salary employment, while the Census paper describes a broader administrative-tax-data picture. A careful reader does not ask which number is the “real” one before asking what each number counts.
That discipline is especially important when the result will be used outside a research paper. A nonprofit board considering pay cannot infer its own salary scale from a national average; an analyst cannot infer worker quality from a sector label; and a policymaker cannot infer service value from payroll share alone. The evidence can show that the comparison changes when workers, jobs, and employers are treated differently. It cannot turn those changes into a universal ranking of nonprofit and for-profit work.
Three layers inside one pay-gap headline
A raw sector average mixes at least three layers. The first is who works there. In the BLS comparison, nonprofit workers are more educated and older than for-profit workers, more likely to be women and to work in professional occupations, and more concentrated in service industries. In 2018–22, nonprofit workers averaged 15.5 years of schooling and 43.7 years of age, compared with 13.8 years and 40.7 years for for-profit workers. Nonprofit workers also worked fewer usual weekly hours on average. These differences can push the nonprofit average up or down without telling us what the nonprofit employer pays a comparable worker.
The reversal in the BLS results makes the arithmetic visible. In the late period, the mean hourly wage was $35.10 for nonprofit workers and $34.12 for for-profit workers, so the unadjusted comparison makes nonprofit employment look better paid. But the nonprofit workforce also had more schooling and experience and was older, while its typical hours and occupational mix differed. When the comparison asks what happens among workers with similar measured characteristics, the direction changes: the article estimates a nonprofit penalty. That does not mean the raw wage was false or that the adjusted estimate has uncovered every hidden difference. It means the two estimates answer different questions about the same broad labor market.
This is why “selection” should not be used as a synonym for low ability. Workers sort into sectors for many reasons, and a dataset cannot observe every relevant reason equally well. A credential, occupation, or age measure may capture part of a worker's expected productivity, while leaving preferences, career plans, family circumstances, or unmeasured skills outside the comparison. The point of adjustment is not to label one group as better; it is to ask how much of the initial gap travels with the measured composition and how much remains associated with the sector after that composition is held more nearly constant.
The second layer is what work is being compared. Nonprofit employment is heavily concentrated in hospitals, educational services, social services, other professional services, and religious organizations. A hospital, a school, a publishing organization, and an arts institution do not face the same product markets, funding conditions, or requirements for service quality. If one sector is concentrated in occupations with higher credentials or in industries where nonprofit and for-profit employers rarely meet, a sector-wide average is partly a map of employment composition.
The third layer is what the employer or sector pays for the work. After accounting for worker characteristics, a difference may remain because of pay-setting, benefits not captured in the chosen measure, funding constraints, bargaining conditions, mission-related utility, or the way an organization values service quality. The supplied studies do not reduce this layer to one cause. The Census result says that worker selection explains little of the average penalty within its job-ladder framework. It does not say that every nonprofit has the same employer-side pay, or that the remaining difference has already been fully explained.
These layers should not be mistaken for a literal formula that every study estimates in the same way. They are a reading model. “Worker selection” means that workers with different observed or unobserved characteristics are not distributed randomly across employers. “Employer-side pay” means the component associated with the firm or sector after the worker component is taken into account. “Industry context” describes whether a meaningful for-profit comparison exists and which institutional pressures surround the job. Time matters as a fourth dimension because the composition of the sector and the size of its pay differential can change.
This model changes the practical question. Instead of asking only whether nonprofit workers earn more or less, we can ask: compared with whom, doing what, under which sector definition, and after which differences have been removed? Those questions are not technical decoration. They determine whether a raw premium is mostly a reflection of education and occupation, whether a penalty remains after accounting for worker characteristics, and whether the comparison says anything about an employer rather than about the people who enter the sector.
What the comparison can—and cannot—tell us
The title's claim is deliberately narrower than “nonprofit workers are underpaid” or “mission justifies lower wages.” The strongest supplied evidence says that selection alone is not enough to explain the average nonprofit penalty in the Census analysis. The paper's selection-corrected results find very little role for worker selection on average: nonprofit workers are not generally worse than for-profit workers. That is a substantial correction to a simple story in which nonprofits merely attract less productive workers who accept less pay.
But “not just selection” leaves several questions open. The BLS article describes two theories that predict opposite pay effects for equally skilled workers. Under labor-donation theory, some workers receive utility from an organization's social mission and may accept lower wages as a compensating trade-off. A sector-wide penalty requires enough qualified workers with such preferences at the margin; it cannot be inferred from the existence of a mission alone. Under property-rights theory, nonprofit managers have no residual claim on profits and may place greater weight on service quality, broad access, or employee pay. When quality is hard for outsiders to observe, the organization may employ more skilled workers or pay more in order to deliver the service.
Those mechanisms can coexist across industries. The Census paper finds that health and education workers are often similar across nonprofit and for-profit sectors, with positive selection in some health-care categories. It finds negative selection in publishing, while the arts show a penalty that is not explained by worker selection. The comparison is strongest where nonprofit and for-profit employers actually overlap. In sectors with little overlap, a numerical difference may be telling us more about different activities than about the effect of nonprofit status.
The first judgment supported by this series is therefore both clear and bounded. Nonprofit pay differences should not be dismissed as the inevitable result of who chooses mission-driven work. Nor should a sector average be treated as a universal salary rule, a proof that every worker is donating labor, or a complete account of organizational value. To interpret the evidence, we have to keep the worker, the employer, the industry, and the time period in view at once. The next question is the most basic of those distinctions: which observable differences between nonprofit and for-profit workers make the raw average look different, and what remains when the comparison accounts for them?