What the Pay Gap Really Tells Us
The U.S. nonprofit labor market supports a precise answer to the question at the center of this series: the nonprofit pay differential is not explained only by the kinds of workers who choose nonprofit jobs. The stronger conclusion is not that every nonprofit pays less, or that mission never affects a worker's choices. It is that worker selection, employer-side pay, industry overlap, and time must be separated before the evidence can be interpreted. The practical result is a way to evaluate a pay-gap claim without turning a sector average into a judgment about an individual worker or organization.
The answer is “not just selection”—but that is not the same as one universal penalty
Start with the difference between a raw average and a comparable-worker estimate. In the BLS analysis, nonprofit workers earned more than for-profit workers in the unadjusted hourly-wage comparison throughout 1994–2022, with an unadjusted premium in the broad range of 5 to 13 percent. Once the analysis controlled for education, experience, occupation, industry, hours, geography, and other observed characteristics, the nonprofit differential became a penalty of about 4 to 7 percent. The raw premium and adjusted penalty are not rival reports of one fact. They are answers to different questions.
The Census study reaches its central conclusion through a different design. It reports an unconditional nonprofit earnings penalty of 12 percent relative to for-profit workers and uses a worker–firm job-ladder model to account for worker effects and employer effects. In that framework, the authors find that most of the average penalty is causal rather than driven by selection, and that worker selection plays very little role on average. In plain language, the result is difficult to explain as a story in which nonprofits simply attract less capable workers who accept lower pay.
That is the strongest supported answer to the title's question. A selection-only explanation is inadequate. The conclusion survives a demanding comparison more convincingly than it would survive a raw average alone, because the Census model uses workers' movements across employers to distinguish a component that travels with the worker from a component associated with the firm. It is still a conclusion within an additive model and an exogenous-mobility assumption, not the result of randomly assigning workers to nonprofit and for-profit jobs.
The qualification is part of the finding, not a retreat from it. The Census study does not say selection is absent. Its industry results show that nonprofit workers in publishing are, on average, negatively selected relative to for-profit counterparts, while some health-care categories show positive selection. In the arts, a nonprofit penalty remains even though worker selection does not explain it. Health and education workers are often similar in quality across sectors. The average therefore contains several local relationships: in one industry selection contributes to the gap, in another it may work in the opposite direction, and in a third employer-side or market features matter more.
The BLS evidence reinforces the same caution from another angle. Its adjusted estimate remains negative after a substantial set of observed differences is held constant, but that does not make the estimate a direct worker-to-worker experiment. The BLS analysis is based on CPS wage-and-salary workers, hourly wages, and observed controls. The Census analysis uses annual administrative earnings for full-time, main-employer worker-years and a worker–firm mobility design. Their estimates should be reconciled as related evidence about the importance of composition and the persistence of a sector component, not averaged into a new consensus number.
What remains after selection is not automatically “the nonprofit effect” in a complete sense. A firm-associated component can reflect funding, bargaining conditions, local labor supply, mission-related utility, the value placed on service quality, schedules, employment continuity, or other features not fully represented in the earnings measure. The Census paper identifies career trajectories, tax treatment, and family structure as future areas of inquiry. The BLS article presents labor-donation and property-rights theories that point to mission value and employer objectives, but it does not claim that one mechanism dominates everywhere.
This distinction also prevents a moral mistake. Saying that selection explains little of the average penalty is not saying that nonprofit workers are underpaid in every setting, that a lower wage is necessarily unfair, or that workers who value a mission are donating labor in a fully informed and voluntary bargain.