Nonprofit earnings and sectoral employment in the United States since 1994

In this article, we present and analyze data on nonprofit employment and earnings by using records of individual wage and salary workers in the monthly Current Population Surveys (CPS), a joint product of the U.S. Bureau of Labor Statistics (BLS) and the U.S. Census Bureau. The monthly CPS files provide annual samples whose sizes range from approximately 130,000 to 184,000 wage and salary workers between 1994 and 2022.¹ The U.S. Census Bureau posts each monthly CPS file online as it becomes available throughout the year.

As we document below, compared with workers in the for-profit sector, nonprofit workers are more highly educated and more likely to be female, White, and in professional occupations. They also are more likely to work in service industries such as hospitals, educational or social services, and other professional services. In terms of education and occupation, nonprofit workers are more similar to workers in the public sector than those in the for-profit sector.

Over the past three decades, the average hourly earnings of nonprofit workers were slightly below those of public sector workers but above those of for-profit workers. In the next section, we discuss competing economic theories suggesting that, in an analysis holding worker characteristics constant, nonprofit workers could earn either more or less than for-profit workers. We then examine the data to test whether employment in the nonprofit sector leads to an earnings penalty or premium, holding worker skills constant. Although nonprofit workers earned 5 to 13 percent more than for-profit workers over the 1994–2022 period, this finding does not account for their individual skills and attributes. After controlling for worker characteristics, we find that, over the same period, workers in the nonprofit sector earned 4 to 7 percent less than for-profit workers, a wage differential implying a wage penalty for working in the nonprofit sector. Controlling for worker characteristics, we find that nonprofit workers earned 3 to 6 percent less than public sector workers.

Economic theory and nonprofit wages

An organization qualifies for nonprofit status if it fits into one of the categories detailed in section 501(a) of the Internal Revenue Code. The code describes a host of such nonprofits, mostly including charitable organizations but also diverse organizations such as labor unions, farm cooperatives, cemetery companies, and mutual insurance companies, to name a few. These organizations are expected to serve some type of common good, but the extent and character of the social benefits they provide vary. Only entities organized for charitable purposes, as described in section 501(c)(3), enjoy benefits beyond the corporate income tax exemption and, notably, the right to solicit tax-deductible donations and exemptions from sales and property taxes.²

The establishment of a firm as a nonprofit can be driven by several considerations.³ First, it may be driven by a desire to attract donations to support the provision of socially beneficial services (e.g., education, healthcare, and legal services). Nonprofit status can make it possible for donors to make tax-deductible contributions. Second, for a firm with nonprofit status, the forfeiture of a claim on profits alters the firm’s objectives in ways that may be desirable to potential donors. For example, without a claim on profits, managers may devote attention to providing services to a wider group or at a lower cost. Finally, nonprofit status may mitigate issues with asymmetric information about the quality of a service (e.g., healthcare or education). If quality is difficult for outsiders to observe, profit-seeking firms may increase profits by reducing quality, whereas nonprofit firms would have less incentive to lower quality. Theory suggests that removing the profit incentive leads nonprofits to provide higher quality service, which requires more skilled workers. Evidence of such behavior is found in long-term care facilities.⁴

Despite differences between the managerial objectives of nonprofit and for-profit firms, wages and benefits for workers in those firms should be influenced largely by the same labor market demand and supply forces that heavily determine wages and benefits in the private for-profit sector. Two prevailing theories exist for why identical workers may be paid differently at the two types of firms: the “labor donation” theory and the “property rights” theory.

The labor donation theory illustrates the theory of compensating wage differentials, suggesting that some workers would be willing to work for a lower wage at a nonprofit than a for-profit firm because they receive some utility from the social benefits provided only in the nonprofit sector.⁵ However, a wage penalty in the nonprofit sector would arise only if the supply of donative workers (i.e., those willing to accept a lower wage for nonprofit work) is sufficiently large. Wage penalties (i.e., labor donations) arise only if “marginal” workers on labor supply curves (where labor demand roughly equals labor supply) have donative preferences. In short, there must be a large share of qualified workers with donative preferences for there to be a nonprofit wage penalty.⁶

The property rights theory is based on an older literature emphasizing that employees at nonprofit organizations might receive earnings that exceed those in competitive for-profit markets.⁷ This literature points out that managers who cannot lay a claim to profits have less incentive to minimize costs and may be more inclined to raise employee pay. Also, the lack of a profit incentive may lead managers to emphasize service quality and employ a more skilled workforce that commands higher pay. As noted earlier, this mechanism could be especially important in industries in which quality is difficult for consumers to observe (e.g., healthcare).

The two prevailing theories yield opposite predictions for the pay level of equally skilled workers in the nonprofit and for-profit sectors. The labor donation theory predicts lower wages in the nonprofit sector, whereas the property rights theory predicts higher wages. The existing empirical evidence on this wage differential is mixed, suggesting that both theories may play a role in explaining it, with the dominant factor depending on the specifics of the particular labor and product markets.

The existing evidence on the wage differential between the nonprofit and for-profit sectors is wide reaching. Studies show that this wage differential varies by gender, race or ethnicity, occupation, and industry. Controlling for worker characteristics (age, education, occupation, and geographic location), one study estimates a 3-percent nonprofit penalty for all workers.⁸ In estimating the differential by gender, the same study finds an 11-percent nonprofit penalty for men and a 1-percent nonprofit premium for women. Although the study finds modest nonprofit penalties for non-Hispanic White workers, it reports nonprofit wage advantages for non-White and Hispanic workers. Findings presented in early studies suggest wage differentials by occupation and industry categories before 2000,⁹ as does the aforementioned study, which uses more recent data (2011–15).¹⁰ The latter study finds substantial nonprofit wage penalties for managers, professionals, and blue-collar workers. In contrast, nonprofit workers in service occupations are found to have slightly higher wages than for-profit service workers. Looking at broad industry categories, the study finds evidence of substantial nonprofit wage advantages for workers in hospitals and service industries, tiny (1 percent) nonprofit wage advantages for workers in nursing care facilities and colleges and universities, and minor nonprofit penalties for employees in elementary and secondary schools. In short, the wage differentials between the nonprofit and for-profit sectors vary across occupations and industries.

Although these differences across demographic and industry groups are of interest, they do not by themselves provide much insight into the relative importance of the labor donation and property rights theories. There are, however, several studies that do shed some light on this issue.

A study on wages at daycare firms finds that nonprofit daycare firms receiving government subsidies pay higher wages than their for-profit counterparts.¹¹ However, the study notes that this effect could be reduced in more competitive product markets. Another study examines a wide range of industries and occupations, seeking evidence on whether nonprofits pay higher or lower wages.¹² The study shows that, within an industry, nonprofit firms frequently pay higher wages than for-profits, although there are some industries in which nonprofits pay less (e.g., legal services). A nonprofit premium within an industry is consistent with the property rights theory. Still, the fact that some nonprofits pay less than for-profits suggests that the labor donation theory may provide a better explanation of the wage differential if the supply of workers who receive utility from working in the nonprofit sector is high or if the added utility from working for a nonprofit is especially large. The study’s finding that highly skilled workers are more likely to incur a nonprofit penalty is consistent with the labor donation theory if higher wage workers are more willing to donate wages to support charitable causes. Also, the fact that healthcare industries are the most likely to pay a nonprofit premium is consistent with the property rights theory’s prediction that nonprofit status could result in greater attention to service quality and the employment of more skilled workers when quality is difficult for outsiders to observe (as is the case, for example, in healthcare industries).¹³

To remove any bias in the estimated wage differential between the nonprofit and for-profit sectors—bias arising from sorting of workers by unobserved skills—some studies use panel data to examine wage changes of workers switching between sectors.¹⁴ The advantage of such longitudinal analysis is that it controls for otherwise unmeasured worker-specific “fixed effects”—in particular, individual worker skills and productivity not measured precisely by years of schooling, age, etc. The results of these longitudinal studies generally point to very small wage differentials between the nonprofit and for-profit sectors. A study using administrative data from Florida finds evidence of a small wage differential.¹⁵ However, it finds that nonprofit penalties are higher in “classic charities” (e.g., churches and civic organizations) than in commercial nonprofits (e.g., banks and insurance providers).

Descriptive evidence from the Current Population Survey

Data tables in this article provide descriptive evidence on annual nonprofit employment, work hours, and earnings, covering the period from 1994 to 2022. Parallel descriptive evidence is reported for wage and salary workers employed in the private for-profit sector and the public sector (federal, state, and local). Our analyses and calculations are based on data from the CPS outgoing-rotation-group (CPS-ORG) microdata files, which are provided monthly by the U.S. Census Bureau. We use the monthly CPS-ORG files from January 1994 (when the CPS first included nonprofit status) to December 2022. We provide evidence on (1) the employment levels of nonprofit workers over time, as compared with those of workers in the for-profit and public sectors; (2) worker characteristics in the nonprofit, for-profit, and public sectors; (3) the industries and occupations with relatively large shares of nonprofit workers; and (4) descriptive evidence and regression-based estimates of wage differentials between “similar” nonprofit and for-profit workers.¹⁶

Summary evidence

As seen in table 1, nonprofit employment in 1994 was 5.4 million, accounting for 5.0 percent of the total U.S. wage and salary employment of 108.0 million in that year. The shares of for-profit and public employment in 1994 were 78.0 and 17.0 percent, respectively. The nonprofit share rose gradually, reaching 7.1 percent in 2009, after which it changed little. Total wage and salary employment peaked in 2019, at 141.7 million. As of 2022, the respective shares of nonprofit, for-profit, and public employment were 7.2, 77.8, and 15.0 percent. From 1994 to 2022, the employment share of the public sector declined from 17.0 to 15.0 percent, while the nonprofit share rose from 5.0 to 7.2 percent. One interesting observation about the time-series data is that, during the Great Recession (2007–09), the nonprofit and public employment shares rose while the for-profit share declined. The same was true during both the 2001 dot-com recession and the most recent pandemic recession, which started with the spread of COVID-19 in 2020. These patterns are consistent with the notion that employment in the nonprofit and public sectors is less procyclical than that in the private sector.

Table 2 provides information on the industries in which nonprofit work is most common. In the table’s top panel, industries are ranked according to the average number of nonprofit workers employed in them between 2003 and 2022.¹⁷ The rightmost column of the table lists the share of total nonprofit workers employed in a given industry. The four industries employing the most nonprofit workers account for slightly over one-half (52.2 percent) of all nonprofit employment. Over the 2003–22 period, the four industries with the largest number of nonprofit employees were hospitals (2.16 million), religious organizations (1.04 million), elementary and secondary schools (0.80 million), and colleges and universities (0.76 million).

The bottom panel of table 2 ranks industries on the basis of the share of their workers who are employed by nonprofits. Four industries have 100.0 percent of their workers employed in the nonprofit sector: business, professional, political, and similar organizations; labor unions; civic, social, advocacy organizations, and grantmaking and giving services; and religious organizations.

Table 3 is structured identically to table 2, although this time the rankings are for occupations instead of industries.¹⁸ The table’s top panel lists the occupations with the largest average annual number of nonprofit employees. Over the 2003–22 period, registered nurses were the occupation with the highest number of nonprofit employees (0.79 million), followed by clergy (0.40 million). The bottom panel of the table shows the occupations with the highest concentration of workers in the nonprofit sector. The occupations that top this ranking include workers engaged in religious activities (directors, religious activities and education; religious workers, not elsewhere classified; and clergy); social and community service managers; and musicians, singers, and related workers.

Table 4 provides the share of a given sector’s workers in each broad industry and occupation for the early (1994–98) and late (2018–22) periods of our sample. The three industries accounting for the largest share of nonprofit workers in the late period were hospitals (24.1 percent), other professional services (23.1 percent), and educational services (19.4 percent). Not surprisingly, nonprofit employment was uncommon in industries involved in producing, transporting, or distributing goods. The three occupations with the largest share of nonprofit workers in the late period were professional and related occupations (50.5 percent);¹⁹ management, business, and financial occupations (19.4 percent); and service occupations (13.4 percent).

Table 5 presents union density for the nonprofit, for-profit, and public sectors over the 1994–2022 period. In sharp contrast to overall private sector unionization, which has exhibited a long-run decline, union membership among nonprofits has been relatively flat. In the for-profit sector, union membership fell steadily over the study period, from 11.0 percent in 1994 to only 5.8 percent in 2022. In contrast, union membership for nonprofit workers rose slightly over time, from 7.1 percent in 1994 to 7.8 percent in 2022. Union membership in the public sector is far higher than that in the private sector, but it has exhibited a downward drift over time. Union membership in the public sector was 38.7 percent in 1994 and 33.2 percent in 2022.

Topcodes, nonresponse, and imputation

As discussed earlier, many studies have examined earnings differentials between the nonprofit, for-profit, and public sectors. The CPS data we use make it possible to examine trends in the earnings differentials over time. We make several adjustments to the CPS earnings data to make the earnings comparisons.

Topcodes: Pareto estimates of mean earnings above CPS earnings caps

The CPS measure of weekly earnings is topcoded, with the value of the topcode exhibiting changes over time. For example, the topcode was $999 from 1973 to 1988, $1,923 from 1989 to 1997, and $2,885 from 1998 to 2022. Workers reporting weekly earnings at or above the relevant topcode in the survey have their earnings reported as the topcoded value. For example, people reporting weekly earnings of $5,000 in 2022 would have their earnings recoded in the CPS public-release data as the topcode for 2022 ($2,885). Because this topcoding understates the measure of earnings, we estimate the mean earnings of workers with earnings at or above the topcode by assuming that the right tail of the earnings distributions follows a Pareto distribution.

Letting Y to represent earnings, with A and α being constants, the Pareto law states that the proportion p of people with incomes greater than or equal to Y is given by AY^-a or, in log-form, ln(p) = ln(A) – αln(Y).²⁰ If the Pareto distribution parameters are known, the expected value of income for those with income above some topcode (Y_T) would be given by E(Y|Y ≥ Y_T) = Y_T × α/(α – 1).²¹ For example, if α = 2 and the topcode for earnings is $2,000, the expected earnings of workers with earnings at or above the topcode would be 2 times the topcode ($4,000).

Our early research used a simple regression approach to estimate the parameters of the Pareto distribution. We restricted the sample to workers with weekly earnings at or above the median. For each of the N distinct earnings (y_i) observed in the restricted sample, we estimated the percentage of workers with earnings at or above that amount of earnings (p_i). We then estimated the regression equation ln(p_i) = θ₀ + θ₁ln(y_i), [i = 1…N). The Pareto parameter α was obtained by reversing the sign of the estimate θ₁. Not surprisingly, the estimate of α was positive because as the level of earnings rises, the percentage of people with earnings at or above that level falls.

More recent research suggests improvements to the estimation process for the Pareto parameters describing the distribution of earnings.²² Two of these improvements include restricting the range over which earnings are assumed to follow the Pareto distribution to higher percentiles than the median and using more information available in the data. Considering the suggested refinements, we now use the following formula for estimating α:²³

where M is the number of workers with earnings between the lower cutoff (Y_C) for the restricted sample (e.g., the 80th percentile of earnings) and the topcode for earnings (Y_T), T is the number of people with earnings at or above the topcode Y_T, and Y_i is the earnings of individual i.

The earnings measures in table 6 provide a historical view of the earnings topcodes, the percentage of workers with weekly earnings at or above the topcodes, and the Pareto estimates of mean earnings for those with earnings at or above the topcodes. Over the 1973–88 period, during which the topcode was frozen at $999, wage growth caused the percentage of men with topcoded earnings to rise from 0.2 to 7.1 percent. Over the same period, the percentage of women at the topcode rose from 0.0 to 1.1 percent. In 1989, the topcode for weekly earnings was increased to $1,923 (about $100,000 per year), and the percentage of workers at the topcode dropped for both men (1.0 percent) and women (0.1 percent). With the topcode frozen at $1,923 until 1997, the percentage of workers at the topcode rose steadily, to 2.6 percent for men and 0.6 percent for women. The share of workers with topcoded earnings fell in 1998, when the topcode was increased to $2,885. With the topcode remaining the same from 1998 to 2022, wage growth pushed the percentage of men and women with topcoded earnings to 10.2 and 4.6 percent, respectively. The Pareto estimates of mean earnings for those with earnings at or above the topcode have ranged from 1.4 to 2.0 times the topcode for men and from 1.3 to 1.8 times the topcode for women.²⁴

To calculate workers’ hourly wages, we divide their usual weekly earnings (adjusted for topcoding) by their usual weekly hours. For workers who do not report usual weekly hours, we use their previous-week hours as a substitute. Estimates of the average hourly wage can be sensitive to unusually high earners entering or leaving the sample. To reduce the impact of these outliers on earnings estimates, BLS frequently reports median rather than mean earnings. Because standard regression analysis provides estimates of how control variables affect the conditional mean of earnings, we focus on mean wages but drop all observations with hourly wages above the top 0.5 percent or below the bottom 0.5 percent of earners, mitigating the effect of outliers.

Match bias: attenuation of earnings differentials due to U.S. Census Bureau imputations

CPS earnings nonresponse, another measurement issue, has increased steadily over time, affecting roughly a third of all current wage and salary workers. Nonresponse can create a substantial “match bias” (attenuation toward zero) of the estimated effect of observable characteristics on wages (e.g., the impact of nonprofit status on wages).²⁵ This bias occurs because the U.S. Census Bureau uses imputed earnings (“allocations,” in the Bureau’s jargon) for individuals who do not report earnings. These imputed earnings are the earnings of respondents who are “similar” to respondents not reporting earnings. Nonprofit status, however, is not a criterion included in matching earnings from a respondent to a similar nonrespondent, and neither are industry, location, and other wage determinants. The U.S. Census Bureau imputation matching includes race, usual hours worked, gender, and broad measures of education, age, and occupation.²⁶ To mitigate the effect of allocated earnings on the estimation of earnings differentials between the nonprofit sector and the for-profit and public sectors, the following analysis drops all observations identified as having allocated earnings.²⁷

Although there were no CPS earnings imputations before 1979, from that year onward all wage measures and estimated wage gaps have been affected by wage imputations (i.e., allocations). Earnings imputations were applied in the 1994–95 CPS files, but the U.S. Census Bureau did not provide allocation flags that could have identified earnings nonresponse and subsequent imputations. Given that earnings nonrespondents in the CPS cannot be identified for 1994 and 1995, our CPS sample sizes for those 2 years are substantially larger than those for other years.

Table 7 provides information on the mean characteristics of workers in the nonprofit, for-profit, and public sectors in the early (1994–98) and late (2018–22) periods of our sample. In both periods, the hourly wages of workers in the nonprofit sector were higher than those of workers in the for-profit sector but lower than those of workers in the public sector. These wage differentials may be due to different types of workers being employed in the three sectors. Relative to wages in the for-profit sector, wages in the nonprofit sector could be higher because nonprofit workers tend to be older and more educated. For the 2018–22 period, the average age of nonprofit workers was 43.7 years, compared with 40.7 years for for-profit workers and 44.7 years for public sector workers. For the same period, the mean years of completed schooling for nonprofit, for-profit, and public sector workers were 15.5, 13.8, and 15.3, respectively. The fact that part-time work is more common in the nonprofit sector than in the for-profit or public sector could result in lower wages in the nonprofit sector.

Given the age, education, and occupation differences across workers in the three sectors, it is reasonable to expect differences in their mean hourly earnings. For the 2018–22 period, the mean hourly wages for the three groups (reported in 2022 dollars) were $35.10 for nonprofit workers, $34.12 for for-profit workers, and $35.37 for public sector workers.²⁸ The fact that nonprofit workers tend to be more educated and more experienced than for-profit workers may help explain their higher wages.

Wage differentials between nonprofit workers and for-profit and public sector workers: evidence

We address two main questions in this section. First, are nonprofit workers paid differently (on average) than for-profit or public sector workers who provide similar levels or types of work? Second, has the wage premium or penalty for being a nonprofit worker changed over time?

Table 8 (second column) and chart 1 (unadjusted series) show the unadjusted wage gaps between nonprofit and for-profit workers (i.e., log-wage differentials obtained without controlling for worker characteristics). Consistent with the evidence presented in table 7, the wages of nonprofit workers exceeded those of workers in the for-profit sector for every year since 1994. Relative to the for-profit sector, the nonprofit unadjusted wage premium fell from 0.08 in 1994 to 0.05 in 2001, subsequently rose to a peak of 0.13 in 2014, and then fell to 0.07 in 2022. Relative to wages in the public sector, nonprofit wages were lower for every year since 1994, although the shortfall diminished over time. (See table 8 and chart 2.) The nonprofit unadjusted wage penalty was the largest (0.16) in 1996, but then fell to 0.03 in 2022.

Although the unadjusted wage differentials are of interest, they do not shed light on whether workers with given skills and attributes receive a pay penalty or premium for working in the nonprofit sector. To investigate this question, we run wage regressions that control for worker characteristics, estimating wage differentials across sectors. Specifically, we estimate conditioned wage differentials between the nonprofit and for-profit sectors and between the nonprofit and public sectors, using the log of hourly earnings (in 2022 dollars) as the dependent variable. The independent variables used in the regressions include the following detailed worker attributes (number of categories shown in parentheses): education (6), female, union membership, potential experience and its square, female interacted with potential experience and its square, race or ethnic status (5), region (8), metropolitan population size (6), hours-worked category (2), and detailed occupation (451) and broad industry (21) controls that capture aspects of skills and work conditions. Table 8 (sixth column) and chart 1 (adjusted series) show estimates for the adjusted wage gap between the nonprofit and for-profit sectors. These estimates suggest nonprofit wage penalties ranging from 0.04 to 0.05 in the early (1994–98) period of our sample and from 0.06 to 0.07 in the late (2018–22) period. Over the entire study period, the nonprofit wage penalty (relative to the for-profit sector) has been remarkably stable after controlling for worker characteristics. The same is true for the nonprofit wage penalty relative to the public sector, with that penalty ranging between 0.03 and 0.06 over the period. (See table 8 and chart 2.) Overall, the analysis controlling for worker characteristics suggests that nonprofit workers incur a wage penalty of 0.03 to 0.06 relative to both for-profit and public sector workers.

Given the evidence summarized above, we conclude that the wage differentials between the nonprofit and for-profit sectors are rather modest once we account for worker attributes and skills. However, the standard cross-sectional wage differentials between the two sectors are not zero. Instead, using standard wage-level analysis, we find modest nonprofit penalties over the entire study period. One plausible explanation for this finding is that, over time, nonprofit workers have accumulated fewer skills than for-profit workers of the same age and with the same levels of schooling. This relative decline in accumulated skills for nonprofit workers is likely due to those workers having lower average hours worked per week (about 2 hours less) relative to their for-profit counterparts. We find no evidence that nonprofit workers systematically receive higher wages than equivalent for-profit workers. A reasonable conclusion is that the average wage differentials between equivalently skilled workers in the two sectors are negative but modest and not universal.

The focus so far has been on comparing nonprofit and for-profit wages or earnings. It should be noted, however, that workers’ total compensation includes a substantial share of nonwage benefits, including employer contributions to retirement benefits, health insurance, vacation, and leave time (e.g., sick days and paid family leave). A study using establishment data finds that nonprofits and for-profits have nearly identical shares of nonwage benefits with respect to total compensation, 31 and 30 percent, respectively.²⁹

To shed light on some of these nonwage benefits, table 9 shows rates of employment-based health insurance and pension coverage by sector, using data from the 2019–21 CPS Annual Social and Economic Supplement. As seen in the table, 60.0 percent of nonprofit workers held employment-based health insurance, compared with 51.5 percent of for-profit workers. As might be expected, public sector workers had higher health insurance coverage (69.3 percent). A similar pattern is observed for pension plans. Among nonprofit workers, 50.6 percent were offered pension plans and 41.9 percent were covered by those plans; these percentages compare with 41.0 and 32.5 percent, respectively, for for-profit workers. Public sector workers had substantially higher pension plan offer and coverage rates (72.7 and 65.4 percent, respectively) than did nonprofit and for-profit workers.

Conclusion

Considering the results of our analysis, we conclude that economy-wide differences in earnings for similar nonprofit and for-profit workers are rather modest, on average. Given evidence of similar levels of nonwage benefits across workers in the two sectors, we can expand that conclusion to total compensation (i.e., wages plus benefits). Economy-wide average compensation is highly similar for U.S. nonprofit and for-profit workers engaged in equivalent levels of work. Our data also suggest that, since 1994, nonprofit employment shares have increased, whereas public shares have declined and for-profit shares have changed little. In addition, we find that nonprofit workers tend to be more educated, older, and concentrated in service industries.