Are the Nonreligious Really Less Generous?
Perhaps the most important shift in religiosity in the US over the last thirty years has been the sustained increase in the percentage of the population that identifies as having no religious affiliation – the “nonreligious.”1 In 1990, the nonreligious in the US made up just 7% of the population (Kosmin et al., 2009). Surveys from the Pew Research Center suggest the percentage of unaffiliated adult Americans is close to 30% as of 2025 (Pew Research Center, 2025; G. A. Smith, 2021). This dramatic shift may have important ramifications for charitable giving if some scholars are correct in their assertion that religious individuals are more charitable than are nonreligious individuals (Li, 2017; Putnam & Campbell, 2012; Roberts & David, 2019). In their 2012 book, Putnam and Campbell asserted, “religious Americans give more generously than secular Americans, both to religious and to secular causes” (p. 448). Other scholars have found a similar gap in charitable giving between the religious and the nonreligious (Bekkers & Wiepking, 2011; A. C. Brooks, 2003; Helms & Thornton, 2012; Mathur, 2012) though other studies do not find such differences (A. C. Brooks, 2005; Eckel & Grossman, 2004; Gruber, 2004; Wang & Graddy, 2008; Wilhelm et al., 2008).
Putnam and Campbell’s work was published more than ten years ago and, while some scholars have investigated this question since then (Bekkers & Wiepking, 2011; Eagle et al., 2018), there are several reasons why we believe this should be re-examined. The first reason is because of the continued growth of the nonreligious, who have increased as a percentage of the US population by another 50% (from roughly 20% in 2010 to roughly 30% in 2021). Another factor that suggested this question warranted further investigation was Giving USA’s 2025 report (Giving USA, 2025), which showed that, during the time that the nonreligious have been on the rise (1990–2025), so, too, have charitable donations. While it is possible that the religious are giving more to make up for declining charitable giving among the nonreligious, the more likely interpretation is that the nonreligious are not less charitable than the religious.2 To test this, we use a dataset well-suited for addressing this question, the Panel Study of Income Dynamics (“PSID”; Survey Research Center 2025), which includes data on both religious and secular giving. We examine whether there is a meaningful difference in charitable giving between religious and nonreligious people.
Literature Review
Growth of the Nonreligious
If to be religious is to adhere to a constellation of beliefs, moral teachings, and/or other ideas centered on some conception of the otherworldly or supernatural, and to instantiate these things through particular behaviors and identities (e.g. religious rituals and organizational affiliations), then to be nonreligious is, simply put, not to do this (Cragun and Smith, 2024). The nonreligious have beliefs, behaviors, and identities; they are just less likely to be centered on supernaturalism or otherworldly interpretations of happenings in the universe.3 As previously mentioned, there has been a sharp rise in the numbers of Americans who claim no religion. Yet, there are many studies that support the claim that adult Americans, despite recent declines in religiosity, remain more religious than are citizens of many other industrialized nations (Berger et al., 2008; Norris & Inglehart, 2004; Pew Research Center, 2015). Even though Americans remain somewhat more religious than citizens of other developed countries, for the last thirty years there has been a consistent decline in religiosity (Pew Research Center, 2025; Voas & Chaves, 2016, 2018). Adult Americans are less likely to report having a religious affiliation, are less confident in their belief in a god or higher power (Voas, 2025) and have reduced their frequency of religious attendance since the 1980s (Chaves, 2011). These changes align with secularization theory (Stolz et al., 2016; Stolz, 2020), which asserts that modernization causes problems for religions, leading to religious decline over time (Bruce, 2013). The US is not alone in witnessing declining religiosity; similar trends are occurring in many countries (Kasselstrand, 2019; Kasselstrand et al., 2023; Stolz et al. 2021; Voas, 2007, 2009).
Differences in Giving
There is extensive literature on charitable giving generally (see Bekkers and Wiepking 2011), including psychological studies examining different mechanisms and relationships involved in generosity including individual cost-benefit analysis, the nature of altruism, and the psychological benefits of giving (Batson, 2011; Fehr and Fischbacher, 2003). However, far fewer studies undertake close sociological examination of giving specifically among the nonreligious. We argue this last part is important, given that, even while religiosity is declining in the US, charitable giving has actually been on the rise. Adjusted for inflation, total giving from 1979 through 2024 has consistently increased (Giving USA, 2025). Charitable giving totals more than $592 billion dollars in the U.S. (Giving USA, 2025). One notable shift in giving, however, has been a transition to giving to secular charities rather than religions (National Philanthropic Trust, 2021). The greater increase in giving to secular charitable organizations may reflect shifting religiosity in the US, with nonreligious individuals increasingly involved in secular activism and philanthropic organizations (Blankholm, 2014; Kettell, 2014) as nonreligiosity can also drive civic engagement (Frost & Edgell, 2018; Speed & Edgell, 2023).
Reflecting the secularization that has taken place in the US, there has been an increase in the number of secular community and philanthropic groups (Cimino & Smith, 2014; García & Blankholm, 2016). Logically, individuals who are not affiliated with religions should be less likely to donate to religions and instead give to secular organizations, whether they are affirmatively secular (Cragun et al., 2017) or secular in the sense that they are not involved in any religion but focus on other social issues, like the environment, education, or health. This is not to say that nonreligious Americans are more generous in their giving overall, only that they are more likely to target their donations away from religions and toward secular charities.
It is also important to note that the increase in giving to secular organizations likely involves factors that go beyond the rising numbers of nonreligious individuals. For example, the explosive growth of online social networks in recent decades may provide greater reach, access, awareness, and/or more opportunities to influence people to give to various public causes. There is evidence that social media not only allows users to discover charitable causes that they may be more personally tied to or moved by but also enhances donation behavior and fundraising success (Di Lauro et al. 2019). Nonprofits are able to leverage social media (i.e. charity badges) to encourage peer-to-peer giving (Saxton & Wang 2014) and sharing or reposting donation appeals promotes prosocial action (Boulianne 2022). The increase in philanthropic opportunities and organization combined with heightened awareness due to social media may be a factor in the rise of charitable giving.
The Claim about Differences
One widely-cited source for the claim that religious Americans are more generous than secular Americans is Putnam and Campbell (2012), who argued that religious people are more generous even when donating to secular organizations. This claim is intermingled with other claims by Putnam and Campbell about the pro-sociality of religious individuals over less religious individuals, including the assertion that religious people are nicer, more likely to volunteer, and are more generous in their giving toward both secular and religious causes (see Galen 2012 for an alternative perspective). Additionally, Putnam and Campbell assert that philanthropic organizations dedicated to helping the needy benefit more from donations by the religious than donations from less religious Americans.
That Putnam and Campbell found such differences in the early 2000s when they gathered their data led us to wonder if, with the continued secularization of the US, those findings were still accurate. While we return to this explanation in our discussion below, it may be the case that Putnam and Campbell, along with other scholars (Bekkers & Schuyt, 2008; Bekkers & Wiepking, 2011; Brady & Hapenny, 2010; A. C. Brooks, 2003; Carabain & Bekkers, 2012; Helms & Thornton, 2012; Li, 2017; Mathur, 2012; Roberts & David, 2019; Wilson & Son, 2018), found these differences during a time when the nonreligious were still a distinct minority in the US population (Strawn, 2019). As more recent data indicate, with continued secularization, the nonreligious now make up more than one-fourth of the US population and are increasingly similar to the rest of the US population (Strawn, 2019). Thus, there may have been differences in giving to charitable organizations between the religious and nonreligious a decade ago, but the continued growth of the nonreligious may have changed that dynamic, justifying further examination of this claimed difference in giving.
Additionally, recent research focused on the nonreligious by Frost and Edgell (2018) and others (Baker et al., 2018; LeDrew, 2015) has found that there is substantial variation within the broad category of the nonreligious, building on the work of Hwang, Hammer, and Cragun (2011). These scholars show that affirmatively secular individuals, those who identify as Atheists, Humanists, freethinkers, and so forth, are substantively different from individuals who identify as “nothing in particular” (Pew Forum on Religion, 2012; we refer to them as “nones”). Frost and Edgell (2018) found that Atheists, Agnostics, and the spiritual but not religious were much more likely to be engaged in civic life than those who identified as “nothing in particular” (see also Speed and Edgell 2023). These more recent findings again raise questions about previous studies that suggested highly religious individuals were more charitable than less religious individuals. Our wording here is intentional; many prior studies have failed to separate affirmatively secular individuals (e.g., Atheists, Humanists, etc.) from individuals who report a religious affiliation but who infrequently attend (i.e., the less religious). These groups are also substantively different from individuals who do not report a religious affiliation but either do not consider religion relevant to them (Quack and Schuh 2017) or retain some elements of religiosity outside of organized religion (Beider 2022). This is the case in Putnam and Campbell (2012), who grouped marginally religious individuals (i.e., those who reported a religious affiliation but did not regularly attend services) together with nones and affirmatively secular individuals for their analyses. Given the latest research on the nonreligious as well as the growth of the nonreligious, such an approach is problematic as it hides the variation that exists among the nonreligious.
Secular vs Non-Secular Giving
At the core of any such comparisons lies a distinction between religious and secular charities. As noted above, it would, indeed, be surprising to find that nonreligious individuals are more likely to give or give more (or even the same amount) to religions than do religious individuals. People are more likely to participate in and give to organizations they care about or value (Bekkers & Schuyt, 2008; Bekkers & Wiepking, 2011; Breeze, 2013; Yeung, 2018). It is not surprising that those who consider religion important donate more to religious causes (Li, 2017; Wang & Graddy, 2008). Religious individuals are the “members” of their religious organizations and receive the primary benefits (McBride, 2007), though the charitable works of religions may spread beyond the core beneficiaries as well. Some religions “require” membership dues (e.g., some Jewish synagogues) while others encourage or even expect members to donate a percentage of their yearly income (i.e., tithes) to be considered in “good standing” in the religion (e.g., the LDS Church; Hammarberg, 2013). While specific amounts or percentages are not strictly enforced in most religions, many expect members to give as part of their regular participation in the religion. Importantly, both likelihood of giving and how much people give warrant analysis, as less affluent individuals may be just as likely to make charitable donations, but will not give as much. Secular charitable organizations do not have the same expectations of giving as religious organizations do; they do not typically have the same structures in place that allow them to exert similar kinds of pressure on individuals to donate to them. Choi and DiNitto (2012) found that religious identification was the strongest predictor of religious giving while Bekkers and Wiepking (2011) argued that church members were more likely to report charitable giving and report higher donation amounts. As a result, donations to religions tend to be higher, particularly among some religious groups, than are donations to secular organizations (A. C. Brooks, 2003, 2005; Indiana University Lilly Family School of Philanthropy at IUPUI, 2019).
Although our study is not entirely novel, as other scholars have examined differences between religious and nonreligious giving (including some that use the PSID), it remains the case that many prior studies have failed to distinguish between donating to religions and donating to secular charities (Eckel & Grossman, 2004; Frost & Edgell, 2018; Li, 2017; Wang & Graddy, 2008). Our study is unique in that we not only make this distinction, but we also parse the generic term “nonreligious” into more meaningful and instructive subcategories. It is true that religiously affiliated individuals are more likely to give to religious organizations (J. E. Brooks, 2006; Lincoln et al., 2008; C. Smith et al., 2008), but what about secular individuals? With improvements in study design, research examining differences in giving between the religious and nonreligious must necessarily take into consideration that nonreligious individuals will be less likely to donate to religions than religious individuals, net of controls (Bekkers & Wiepking, 2011; A. C. Brooks, 2005; Eagle et al., 2018; Eckel & Grossman, 2004). This leads to our first two hypotheses:
Hypothesis 1a: Net of controls, nonreligious individuals – both atheists and nones – will have a lower likelihood of making donations to religions than will religious individuals.
Hypothesis 1b: Net of controls, nonreligious individuals – both atheists and nones – will donate less money to religions than will religious individuals.
Since much of the prior research on this question also finds that nonreligious individuals are less generous in donating to secular charities, we are inclined to hypothesize the same. However, it is worth noting that nonreligious individuals tend to be younger than religious individuals and tend to be disproportionately male (Baker & Smith, 2015; though more recent data suggest the gender ratio may be shifting; see PRRI 2025). Age and being male are both associated with lower levels of charitable giving (Eagle et al., 2018; Wang & Graddy, 2008). Some nonreligious individuals – particularly atheists and agnostics – tend to have higher levels of educational attainment and higher incomes, but nones do not have higher levels of educational attainment or higher incomes than the average religious person and may fall on the lower end of the education and income spectrum (Cragun & Smith, 2024; Strawn, 2019). Many studies have found education to have a positive relationship with time spent volunteering and secular and non-secular charitable giving (Bekkers & Wiepking, 2011; Choi & DiNitto, 2012; Wang & Graddy, 2008). Given the demographic differences that exist between the religious and the nonreligious and even within the broad nonreligious category (Baker & Smith, 2015; Strawn, 2019; Thiessen & Wilkins-Laflamme, 2020), it is important to control for core demographic variables like age, education, and income when examining differences in giving to secular charities. Recognizing the differences in characteristics between the religious and nonreligious, we propose two additional hypotheses:
Hypothesis 2a: Net of controls, there will not be differences in the likelihood of making donations to secular charities between religious and nonreligious individuals.
Hypothesis 2b: Net of controls, there will not be differences in how much money religious and nonreligious individuals give to secular charities.
In our analyses that follow, we test these two hypotheses utilizing the 2023 wave of the Panel Study of Income Dynamics.
Methods
Data for the study come from the Panel Study of Income Dynamics (PSID) 2023 wave (Survey Research Center, 2025).4 In the 2023 wave, there were 9,152 participants, but religious affiliation was not reported or missing for 445, reducing our sample to 8,707 individuals. The PSID is a highly detailed survey focused on household income and spending. Relevant to our purposes, the PSID includes questions about participants’ charitable giving that separate out giving to religions from giving to secular charities.
Dependent Variables
We have sixteen dependent variables, two variables for each type of organization. Participants were asked whether they had made a donation to each of the following types of organizations: religious organizations, organizations for the needy, health organizations, international peace organizations, educational organizations, youth organizations, cultural organizations, environmental organizations, and other organizations. (See PSID codebook for specific wording for the questions.) These variables were all recoded into a dummy code with: 0 = did not donate; 1 = did donate. Our first set of analyses focus on the probability of donating using logistic regression.
PSID participants were also asked how much they donated to each type of organization. Our second set of analyses examine how much individuals donated using Tobit regression (Henningsen, 2021), which is specifically designed for use with censored and skewed distributions as the majority of PSID participants did not donate, resulting in a very large number of “0” responses, and a much smaller number of individuals who made very large donations. Readers should be aware at the outset that donation distributions are heavily skewed and leptokurtic. This raises some challenges when analyzing how much people give, as we detail in the results section.
Independent Variable
Our independent variable is a measure of participants’ religious affiliation. Participants were asked their religious preference. The PSID included close to 30 options. We recoded these into 10 religious affiliations, each of which had a minimum of 30 participants in the PSID:5 None (n = 1,667), Atheist/Agnostic (n = 404), Catholic (n = 1,240), Baptist (n = 1,678), other Christian (n = 3,385), Jewish (n = 116), Muslim (n = 52), Buddhist (n = 30), Latter-day Saint (a.k.a., Mormon; n = 60), and other religion (n = 75). Recognizing that 30 participants is quite small, we include dispersion statistics in all of our analyses to reflect the lack of precision that results from analyzing smaller groups. Preliminary analyses suggested that Catholics were often near the middle in terms of their donations to each of the different types of organizations. We use Catholics as the reference group in our regression analyses. Where there are significant differences for religious groups in our regression tables (see S1 and S2), they indicate differences relative to Catholics. In our marginal effects plots, Catholics are included in the figures. Also, readers should note that we use casewise deletion instead of listwise deletion given how many dependent variables we have. The number of cases included in an analysis is specified in the corresponding results table.
Control Variables
As noted in the literature review, how much individuals give to charitable organizations is highly dependent upon their annual income and their wealth. We control for both of these. The PSID has very robust measures of household income and wealth. However, both of these variables were also highly skewed and leptokurtic. The median family or household income was $64,917, the mean was $93,472.18, the minimum was –$170,200 (due to losses in either a business or the stock market), and the maximum was $5,750,500, with a standard deviation of $132,532.3 (skew = 15.34; kurtosis = 499.33). Likewise, wealth was significantly skewed. Median wealth was $67,000, the mean was $391,512.8, the minimum was –$3,249,000 and the maximum was $48,337,000 (skew = 15.6; kurtosis = 386.21). To address the skew and kurtosis, we used a two-step approach. First, we set all negative values and zeros to “1”. Because we were going to transform the variables, setting those values to “1” meant they would all calculate to “0” when transformed. Second, we used Box-Cox transformations based on the suggested lambda and gamma from the “powerTransform” function from the “car” package in R (Hawkins and Weisberg 2017). For wealth, this resulted in: lambda = 0.1283197; gamma = 0.1. For income: lambda = .320902; gamma = 0.1. After applying these power transformations, skewness and kurtosis fell within acceptable parameters. For wealth: skew = –0.29; kurtosis = –0.92. For income: skew = 0.08; kurtosis = 2.51.
In addition to the two variables described above, we included respondent’s age, respondent’s completed education (in years), the total number of individuals in the family unit or household, the respondent’s sex (male = 1), whether the respondent lived in a metropolitan area or not (metro = 1), and three dummy codes for race: White (n = 5,166), Black (n = 3,713), and Asian (n = 141); American Indian or Alaska Native (n = 62), native Hawaiian or Pacific Islander (n = 13), and other (n = 391) are reflected in the control group. An initial check for collinearity concerns between income and wealth resulted in a correlation below .40. We also ran a separate ordinary least squares regression model testing for VIF. The highest VIF was for one of the religion dummy codes (other Christian) at 2.334. We did not see any indicators that our models suffered from multicollinearity problems.
To calculate probabilities of donating, we use binary logistic regression. To estimate amount donated, because of the left censoring of our dependent variables due to large numbers of people giving no money to the various organizations, we use Tobit regression. All analyses were conducted using R version 4.5.1 (R Core Team 2025). Tobit regression analyses used the “AER” package (Kleiber and Zeileis, 2008). Marginal effects used the “marginaleffects” package (Arel-Bundock et al., 2024).
Results
Table 1 presents descriptive statistics for the independent and control variables for each (non)religious group as well as for the entire sample. There are notable differences in income and wealth by (non)religious affiliation. Baptists have the lowest mean income ($71,216.9) and wealth ($225,471.2) while Jews have the highest mean income ($193,321.1) and wealth ($1,366,952.0). We included standard deviations in Table 1 to illustrate how skewed wealth and income are in the sample before transforming them. We also included the transformed means and standard deviations as those are the values that are in the models. There are also differences in age and race across groups. Atheists/agnostics are the youngest on average (38.34); LDS participants were the oldest (54.52). Baptists are the most likely to identify as Black (.75); Jews were the least likely (.03). All of these variables are controlled in the regression models.
Table 1
Descriptive Statistics by Religious Affiliation.
| NONE (N = 1667) | ATHEIST/AGNOSTIC (N = 404) | CATHOLIC (N = 1240) | BAPTIST (N = 1678) | OTHER CHRISTIAN (N = 3385) | JEWISH (N = 116) | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MEAN | SD | MEAN | SD | MEAN | SD | MEAN | SD | MEAN | SD | MEAN | SD | |
| Total Family Income | 93,574.3 | 172,713.8 | 121,794.6 | 132,917.4 | 112,448.0 | 170,145.4 | 71,216.9 | 101,193.4 | 90,027.6 | 99,589.4 | 193,321.1 | 168,979.3 |
| Total Family Income (transformed) | 104.50 | 41.54 | 118.53 | 38.65 | 113.22 | 40.21 | 96.01 | 37.59 | 105.38 | 38.67 | 141.11 | 38.99 |
| Wealth | 318,176.4 | 1,083,152.0 | 469,219.6 | 1,933,506.0 | 549,187.8 | 1,830,686.0 | 225,471.2 | 914,678.7 | 428,274.8 | 1,420,237.0 | 1,366,952.0 | 1,774,995.0 |
| Wealth (transformed) | 19.27 | 14.22 | 23.27 | 13.38 | 25.18 | 13.23 | 18.82 | 13.29 | 21.91 | 14.11 | 34.68 | 10.55 |
| Age – respondent | 41.64 | 14.07 | 38.34 | 12.78 | 51.97 | 16.73 | 52.38 | 15.55 | 47.89 | 16.42 | 54.76 | 18.15 |
| Educational Attainment – respondent | 13.45 | 2.32 | 14.86 | 1.99 | 13.23 | 3.28 | 13.17 | 2.11 | 13.69 | 2.32 | 16.09 | 1.30 |
| Household Size | 2.57 | 1.51 | 2.20 | 1.26 | 2.79 | 1.58 | 2.38 | 1.46 | 2.54 | 1.47 | 2.29 | 1.39 |
| Male (proportion) | 0.70 | 0.46 | 0.81 | 0.39 | 0.74 | 0.44 | 0.55 | 0.50 | 0.65 | 0.48 | 0.76 | 0.43 |
| Metro Area (proportion) | 0.86 | 0.35 | 0.90 | 0.30 | 0.91 | 0.28 | 0.81 | 0.39 | 0.83 | 0.37 | 0.98 | 0.13 |
| White (proportion)* | 0.57 | 0.50 | 0.87 | 0.34 | 0.76 | 0.43 | 0.24 | 0.43 | 0.55 | 0.50 | 0.97 | 0.18 |
| Black (proportion) | 0.37 | 0.48 | 0.07 | 0.26 | 0.07 | 0.26 | 0.75 | 0.43 | 0.41 | 0.49 | 0.03 | 0.16 |
| Asian (proportion) | 0.02 | 0.13 | 0.04 | 0.19 | 0.02 | 0.14 | 0.00 | 0.04 | 0.01 | 0.09 | 0.00 | 0.00 |
| MUSLIM (N = 52) | BUDDHIST (N = 30) | LDS (N = 60) | OTHER (N = 75) | TOTAL (N = 8,707) | ||||||||
| MEAN | SD | MEAN | SD | MEAN | SD | MEAN | SD | MEAN | SD | |||
| Total Family Income | 77,848.8 | 112,774.3 | 90,149.5 | 86,702.0 | 132,646.6 | 98,140.2 | 107,724.2 | 155,432.9 | 93,498.4 | 132,500.3 | ||
| Total Family Income (transformed) | 92.30 | 46.93 | 107.67 | 34.87 | 125.71 | 33.52 | 106.66 | 47.85 | 105.69 | 40.02 | ||
| Wealth | 116,696.2 | 345,527.8 | 361,686.8 | 799,765.8 | 593,519.0 | 848,042.3 | 391,072.7 | 729,638.8 | 398,465.0 | 1,377,538.0 | ||
| Wealth (transformed) | 13.43 | 13.72 | 20.86 | 15.28 | 28.18 | 12.84 | 23.93 | 13.12 | 21.51 | 14.03 | ||
| Age – respondent | 46.33 | 15.55 | 49.37 | 15.28 | 53.52 | 18.62 | 44.64 | 16.06 | 47.79 | 16.33 | ||
| Educational Attainment – respondent | 13.76 | 2.45 | 14.47 | 2.00 | 15.33 | 2.26 | 14.28 | 2.82 | 13.59 | 2.47 | ||
| Household Size | 2.88 | 1.94 | 2.43 | 1.43 | 3.53 | 1.97 | 2.73 | 1.64 | 2.54 | 1.50 | ||
| Male (proportion) | 0.79 | 0.41 | 0.77 | 0.43 | 0.78 | 0.42 | 0.73 | 0.45 | 0.70 | 0.47 | ||
| Metro Area (proportion) | 0.90 | 0.30 | 0.97 | 0.18 | 0.93 | 0.25 | 0.85 | 0.36 | 0.85 | 0.35 | ||
| White (proportion) | 0.23 | 0.43 | 0.37 | 0.49 | 0.90 | 0.30 | 0.39 | 0.49 | 0.54 | 0.50 | ||
| Black (proportion) | 0.63 | 0.49 | 0.13 | 0.35 | 0.05 | 0.22 | 0.12 | 0.33 | 0.39 | 0.49 | ||
| Asian (proportion) | 0.08 | 0.27 | 0.37 | 0.49 | 0.00 | 0.00 | 0.32 | 0.47 | 0.02 | 0.12 | ||
[i] * “Other” race is reflected in the constant in the regression models. White, Black, and Asian will likely not sum to 1.0; the balance for any given religious group is individuals in the “Other” racial category.
Table 2 shows the probability of donating to the various target organizations by (non)religious group. Table 3 shows the mean donation in dollars to the various charitable organizations by (non)religious group in dollars. Table 3 includes standard deviations to illustrate the necessity of using Tobit regression since it is robust to censoring and skewed distributions.
Table 2
Probability of Donating to Various Organizations by Religious Affiliation Without Controls.
| NONE | ATHEIST/AGNOSTIC | CATHOLIC | BAPTIST | OTHER CHRISTIAN | JEWISH | MUSLIM | BUDDHIST | LDS | OTHER | TOTAL | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Religion | 0.07 | 0.05 | 0.27 | 0.28 | 0.33 | 0.33 | 0.13 | 0.10 | 0.69 | 0.26 | 0.25 |
| Organization for the Needy | 0.13 | 0.23 | 0.19 | 0.15 | 0.18 | 0.43 | 0.25 | 0.28 | 0.26 | 0.19 | 0.17 |
| Health Organization | 0.07 | 0.08 | 0.12 | 0.06 | 0.08 | 0.36 | 0.09 | 0.10 | 0.14 | 0.07 | 0.08 |
| International Peace Organization | 0.03 | 0.09 | 0.06 | 0.03 | 0.04 | 0.20 | 0.06 | 0.17 | 0.05 | 0.07 | 0.04 |
| Educational Organization | 0.06 | 0.09 | 0.11 | 0.07 | 0.09 | 0.22 | 0.10 | 0.07 | 0.23 | 0.13 | 0.08 |
| Youth Organization | 0.04 | 0.05 | 0.06 | 0.05 | 0.05 | 0.09 | 0.08 | 0.10 | 0.14 | 0.04 | 0.05 |
| Cultural Organization | 0.06 | 0.12 | 0.05 | 0.02 | 0.04 | 0.20 | 0.12 | 0.03 | 0.14 | 0.06 | 0.05 |
| Environmental Organization | 0.06 | 0.14 | 0.07 | 0.02 | 0.05 | 0.19 | 0.04 | 0.13 | 0.05 | 0.09 | 0.06 |
| Other Organization | 0.03 | 0.07 | 0.06 | 0.01 | 0.03 | 0.13 | 0.04 | 0.10 | 0.05 | 0.04 | 0.04 |
| Any Organization | 0.30 | 0.44 | 0.49 | 0.40 | 0.49 | 0.75 | 0.48 | 0.57 | 0.80 | 0.49 | 0.44 |
| Secular Organizations Only | 0.28 | 0.44 | 0.38 | 0.27 | 0.34 | 0.71 | 0.46 | 0.50 | 0.48 | 0.39 | 0.33 |
Table 3
Mean Donations in Dollars (not transformed) to Charitable Organizations by Religious Affiliation.
| NONE | ATHEIST/AGNOSTIC | CATHOLIC | BAPTIST | OTHER CHRISTIAN | JEWISH | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MEAN | SD | MEAN | SD | MEAN | SD | MEAN | SD | MEAN | SD | MEAN | SD | |
| Religion | 49.98 | 443.90 | 49.12 | 563.25 | 328.14 | 1041.61 | 847.21 | 3151.40 | 1268.97 | 4806.75 | 595.95 | 2199.48 |
| Organization for the Needy | 43.32 | 239.16 | 130.64 | 646.43 | 93.98 | 567.63 | 47.88 | 237.18 | 127.74 | 1582.62 | 290.04 | 743.47 |
| Health Organization | 21.49 | 230.80 | 37.70 | 326.14 | 49.69 | 351.68 | 18.94 | 230.73 | 24.34 | 248.56 | 204.97 | 845.73 |
| International Peace Organization | 12.14 | 159.86 | 19.33 | 125.67 | 26.49 | 356.79 | 4.94 | 56.74 | 15.91 | 165.48 | 74.96 | 348.07 |
| Educational Organization | 41.43 | 475.96 | 295.32 | 4985.14 | 60.93 | 634.86 | 14.16 | 126.32 | 99.80 | 2103.46 | 82.27 | 259.28 |
| Youth Organization | 9.22 | 88.13 | 32.41 | 291.18 | 21.54 | 197.42 | 10.22 | 132.65 | 24.74 | 410.46 | 17.35 | 76.03 |
| Cultural Organization | 18.07 | 144.95 | 45.61 | 300.62 | 14.88 | 164.96 | 3.56 | 57.70 | 18.73 | 573.32 | 90.95 | 369.88 |
| Environmental Organization | 23.61 | 426.27 | 22.31 | 93.63 | 30.47 | 345.60 | 2.03 | 20.46 | 18.06 | 450.53 | 89.78 | 492.03 |
| Other Organization | 17.07 | 223.58 | 31.55 | 267.09 | 80.79 | 995.24 | 14.04 | 300.32 | 31.05 | 502.14 | 104.74 | 620.45 |
| Any Organization | 25.54 | 153.89 | 73.36 | 580.87 | 78.82 | 262.42 | 108.95 | 370.89 | 182.50 | 787.62 | 173.58 | 381.92 |
| Secular Organizations Only | 22.90 | 149.08 | 77.39 | 644.14 | 47.46 | 230.93 | 14.48 | 78.29 | 45.51 | 474.30 | 120.78 | 262.21 |
| MUSLIM | BUDDHIST | LDS | OTHER | TOTAL | ||||||||
| MEAN | SD | MEAN | SD | MEAN | SD | MEAN | SD | MEAN | SD | |||
| Religion | 173.08 | 814.12 | 25.00 | 85.85 | 7340.25 | 8903.92 | 224.23 | 695.18 | 779.75 | 3496.72 | ||
| Organization for the Needy | 85.00 | 358.75 | 84.33 | 212.01 | 182.92 | 609.71 | 59.05 | 186.89 | 92.99 | 1034.61 | ||
| Health Organization | 7.21 | 30.65 | 3.17 | 10.54 | 44.43 | 169.98 | 12.33 | 63.32 | 29.24 | 278.14 | ||
| International Peace Organization | 3.92 | 22.01 | 20.67 | 83.54 | 2.67 | 14.36 | 15.20 | 67.71 | 15.38 | 191.83 | ||
| Educational Organization | 19.62 | 81.09 | 22.00 | 84.09 | 51.00 | 164.47 | 27.13 | 129.40 | 73.77 | 1727.11 | ||
| Youth Organization | 19.83 | 97.06 | 20.33 | 76.04 | 124.17 | 472.69 | 14.93 | 115.88 | 19.33 | 285.85 | ||
| Cultural Organization | 9.52 | 29.44 | 16.00 | 87.64 | 50.25 | 310.59 | 8.67 | 58.49 | 17.41 | 378.34 | ||
| Environmental Organization | 0.98 | 7.00 | 8.50 | 29.04 | 5.83 | 29.24 | 24.13 | 134.00 | 18.78 | 366.96 | ||
| Other Organization | 2.88 | 15.38 | 30.00 | 105.54 | 411.67 | 2456.82 | 15.33 | 116.11 | 35.52 | 563.23 | ||
| Any Organization | 35.42 | 111.37 | 25.56 | 39.18 | 914.22 | 1073.50 | 45.07 | 95.52 | 120.68 | 561.50 | ||
| Secular Organizations Only | 17.35 | 58.20 | 25.62 | 40.34 | 109.12 | 382.25 | 22.30 | 61.80 | 37.98 | 349.31 | ||
We ran two regression models for each target charitable organization in the interest of model testing and improvement for each question (Rodgers 2010). The first model includes just the (non)religious groups to determine how much variation they explain in either the probability of giving or how much is given. Variation explained in giving probability or amount given is calculated using a variety of pseudo-R2 measures. We interpret the R2 values in line with standard cutoffs for how meaningful the relationships are: an R2 below 0.02 is considered negligible; from 0.02 to 0.12 is considered small; from 0.13 to 0.25 is medium; 0.26 and up is considered large (Cohen 1988). The second model introduces the control variables to compare whether the model with demographics is a meaningful improvement over a model with just religious affiliation. We examined the residuals from the full models and did not see any indications of nonlinearity or heteroskedasticity. In the models, a dummy code for Catholic is not included as a predictor variable, meaning Catholics are reflected in the constant and all groups are compared to Catholics.
This approach to model testing resulted in 36 regression models that generated 18 tables. We have included all of the regression models in two online supplemental files (S1 includes models estimating the probability of giving and S2 includes models estimating how much groups gave). Here we focus on religious affiliation and effect sizes. Readers interested in the contributions from the other variables should examine the supplemental files. Additionally, we created figures using average marginal effects to illustrate the results. There is a figure for each regression model, resulting in 18 figures as well. These are included with the supplemental files. The average marginal effects plots have the added benefit that Catholics are shown in those models and we can more easily compare them to the other (non)religious groups, something we cannot show in the regression tables. The average marginal effects models generally work well for illustrating the results, but the models derived from the Tobit regressions are counter-intuitive (see Supplemental File 2) because of the skewed nature of charitable contributions. The models almost exclusively predict negative donation amounts as they are based on the average person but the underlying regression model is heavily influenced by the very large contributions of a very small number of people.6 As a result, the models nearly uniformly predict negative giving. It may help readers to think of the models in Supplemental File 2 less as illustrating what the average person gives and more of a heuristic to help them see relative differences in how much members of the various (non)religious groups give on average.
As shown in Figure S1A and Table S1A, atheists/agnostics (p = 0.049)7 and nones (p = 0.064) are the least likely to donate to religions, though the margins of error overlap with Muslims and Buddhists, providing partial support for Hypothesis 1a. With the exception of Mormons (p = 0.68), the probability of donating to a religion for the other religious groups is below p = 0.33, meaning less than 1 in 3 religiously affiliated individuals donate to a religion in the US, with some exceptions. Nones and atheists are significantly less likely (S1A) to donate to religions than are Catholics (the comparison group in the model). This is also the only model where the amount of variation explained by religious affiliation is clearly in the small range, with pseudo-R2 measures varying between 0.071 and 0.114. Religious affiliation, including those who are not religiously affiliated, accounts for somewhere between 7.0% and 11.4% of the variation in the probability of giving to religions.
Aside from giving to religions, the average marginal effects shown in Supplemental File 1 illustrate that Atheists generally fall in the upper half of (non)religious groups included in the models when it comes to probability of donating, along with Jews and Mormons. The figures in Supplemental File 1 suggest that nones are generally in the lower half when it comes to their probability of donating. However, the error bars in those figures are important, as are the tables in S1. When looking at the probability of donating to all of the other types of organizations – those that help the needy, health, international peace, educational, youth, cultural, environmental, and other – the amount of variation explained by religious affiliation is never estimated above 4.3% (Table S1H – probability of donating to environmental organization) and is generally below the cutoff for small, hovering in the 1% range. When just looking at the models that only include religious affiliation (S1 tables), nones are often significantly less likely to donate than are Catholics; but when the control variables are included, the difference is typically no longer statistically significant. In all of the models (S1), the control variables explain a greater percentage of the likelihood to donate than does religious affiliation, as the LogLikelihood statistics in the tables in S1 illustrate. These results support Hypothesis 2a.
The same pattern holds with the amount donated, as shown in the tables and figures in Supplemental File 2 (S2). Atheists and nones donate significantly less to religions than do Catholics (Table S2A and Figure S2A), but the estimated R2 for that model is just 0.015, suggesting religious affiliation explains a negligible 1.5% of the variation in amount donated to religions. That is the largest pseudo-R2 for all of the models in S2 and the models in S2 that only include religious affiliation. This suggests weak support for Hypothesis 1b.
Income, wealth, education, and age do explain some of the variation in how much money people donate to various types of charitable organizations (see the full models in S2), but religious affiliation contributes a negligible amount to explaining how much people donate to charitable organizations, including religions. The models and average marginal effects plots in S2 support Hypothesis 2b that there are not meaningful differences in how much religious and nonreligious people give to secular charities. The full models in S2 are better models than those that include just religious affiliation, per the Log Likelihood statistics, but they are still not particularly good models at predicting how much money people give.
Discussion
In this paper, we addressed two questions. First, is there a difference in probability of giving to religions and secular charities by religious (non)affiliation? Second, are there meaningful differences in how much religious and nonreligious individuals give to religions and secular charities by religious (non)affiliation?
Nones and atheists are significantly less likely to make donations to religions than are religiously affiliated individuals. Hypothesis 1a was supported. This aligns with prior research showing that people tend to donate to causes that share their values and interests (Bekkers & Schuyt, 2008; Breeze, 2013; Yeung, 2018). Likewise, nones and atheists give significantly less to religions than do religious individuals, but the amount of variation explained in the amount given by religious affiliation was in the negligible range, suggesting that religious affiliation is not a meaningful predictor of how much money people give to religions. Hypothesis 1b was marginally supported.
When it comes to donations to secular organizations, the story is very different. Controlling for demographics that are related to giving (e.g., wealth, income, age, and education) results in generally negligible differences between the religious and nonreligious in both the probability of giving (Hypothesis 2a) and how much people give (Hypothesis 2b) to organizations that serve the needy, health organizations, international peace organizations, educational organizations, youth organizations, cultural organizations, environmental organizations, and other organizations. Nones and atheists are generally just as charitable as are the religious groups examined in this study. If any group stands out, it may be Jews, who are particularly generous, even after controlling for their much higher incomes and wealth.
Our results differ from some prior work on charitable giving and religious affiliation, particularly that of Putnam and Campbell (2012), though other scholars have also found that there are either no differences or negligible differences in giving to charities by religious affiliation (Eckel and Grossman, 2004; Vaidyanathan et al., 2011; Wang & Graddy, 2008). Why might our results differ from those of Putnam and Campbell? Analyzing charitable giving is complicated for many reasons, not the least of which are the problems of left censoring, obtaining sufficiently detailed data from large samples, and skewed and leptokurtic distributions. Another factor that complicates this is how religious and nonreligious individuals are grouped. As we noted in the literature review, in prior decades it was not uncommon to group the affirmatively secular (Frost & Edgell, 2018) with the less religious (i.e., individuals who report a religious affiliation but who are less engaged), as Putnam and Campbell (2012) did. When such individuals are grouped with affirmatively secular individuals who do donate to secular charities, the consequence is that the charitable giving behavior of the affirmatively secular may be canceled out by the lack of giving of the less religious. As Putnam (2001) himself notes, civic engagement is predictive of other forms of civic engagement. Wang and Graddy (2008) affirm that social networks, social trust, and civic engagement increase the amount given to both secular and non-secular causes. The same authors found that volunteering, along with social and financial capital, positively affect giving towards both causes. As a result, prior research may have inadvertently masked the civic engagement of the affirmatively secular. Scholars should consider re-examining previously analyzed data with this possibility in mind.
Kelly et al. (2024), in a meta-analysis of religion and prosociality, argued that there is a general albeit weak relationship between religiosity and prosociality. Our findings run counter to those conclusions, though Kelly et al. provide some reasons why our findings are likely to differ from their general conclusions. We used a behavioral measure of prosociality – charitable giving – rather than a self-report measure. Kelly et al. (2024) found that behavioral measures of prosociality have an even weaker correlation with religiosity. We used religious affiliation as our primary measure of religiosity, which Kelly et al. find is the measure most weakly correlated with prosocial behaviors. We have a large sample and are using secondary (or “archival”) data, both of which are related to smaller effects in their meta-analysis. Kelly et al. also, for unclear reasons, excluded studies that separated charitable giving by religious and secular targets, which is exactly what this paper does. As we noted above, there is a large body of research showing that people are more likely to give to causes they value (Bekkers & Schuyt, 2008; Breeze, 2013; Yeung, 2018). Our primary goal with the study was to investigate whether giving differs by the target, which our data support: religiously affiliated individuals are more likely to give to religions, but no more or less likely to give to other charities than are nonreligious individuals. By excluding such studies, Kelly et al. (2024) may have inadvertently strengthened the correlation between religiosity and charitable giving. In short, our study has all of the characteristics of the studies in Kelly et al. (2024) that show the weakest relationship between religiosity and prosociality: behavioral measure, large sample size, secondary data, and religious affiliation as the religiosity measure. Intriguingly, the first two of those four criteria are also considered to be indicators of better study quality, as they are associated with greater generalizability. A future study may consider examining whether alternative measures of religiosity like frequency of religious attendance generates the same results as what we found.
Another reason our findings may differ from prior research is due to a problem that plagues a lot of the research in the social science of religion: a misguided focus on p-values without taking into consideration effect sizes (Ellis, 2010). It is possible other scholars could do the same analyses we did and focus on the statistically significant differences in probability of giving and how much different groups give (see the tables in S1 and S2), ignoring the far more important effect size and proportionate reduction of error indicators. We can envision a scenario in which researchers look only at religious affiliation, disregard effect sizes, focus only on p-values, and conclude that nones are less charitable than Catholics. Control variables matter. But so, too, does how much of the variation religious affiliation explains. When it comes to the probability of giving to secular charities and how much one gives, religious affiliation is not a meaningful predictor, generally explaining less than 2% of the variation, which, per Cohen (1988), falls in the negligible range.
There are two important implications from our study. First, secularization does not appear to equate with people being less charitable toward their communities. Second, secular individuals are likely motivated by the same forces that religious people are to give to charities. Secular individuals donate to charitable organizations that share their values. Accompanying the growth of the nonreligious has been the founding and spread of political and community groups that share the values of the nonreligious (Cimino & Smith, 2014; Frost & Edgell, 2018; García & Blankholm, 2016). Prior research suggests that individuals give to charities for various reasons, including altruism (Harbaugh et al., 2007; Ottoni-Wilhelm et al., 2017) and because giving makes people happy (J. L. Aaker & Akutsu, 2009; J. Aaker & Liu, 2008; Dunn et al., 2008; Mogilner et al., 2012; Reed et al., 2007). Giving also improves subjective well-being (Harbaugh et al., 2007; Lyubomirsky et al., 2005). Moreover, research indicates that individuals with higher subjective well-being invest more hours in volunteering and donate more money (Wang & Graddy, 2008), leading to a self-reinforcing cycle. None of these studies exclude nonreligious individuals from their analyses, suggesting nonreligious individuals likely have the same motivations to give and they receive the same benefits as do religious individuals. It therefore makes sense why secular individuals would be as likely to give to secular charities as are religious individuals – they are motivated by the same underlying factors to give. Additionally, individuals who have a greater connection to specific causes are more likely to donate to those causes, which helps to explain why secular individuals are more likely to donate to secular causes than religious causes, as our models show (Cryder et al., 2013).
Our study is of particular importance because the number of nonreligious people in the United States continues to increase. Given this trend, it is possible that donations to religions will eventually plateau and perhaps even begin to decline as the religious continue to shrink as a percentage of the population. The outlook for secular charities, based on our findings, is brighter. Secular individuals are as likely to give to secular organizations as are religious individuals. Assuming the US population continues to secularize (Stinespring & Cragun, 2015), secular nonprofits and charities will likely continue to see increased donations.
Limitations
While the PSID is ideal for this type of analysis, it is possible that some of the organizations that we are classifying as secular were not secular in the minds of PSID participants. For instance, a Mormon could donate to Brigham Young University or a Catholic could donate to a Catholic hospital. When asked if they made a charitable donation to an educational organization, the Mormon could clearly say that they had. The Catholic could say that they had donated to a health-related organization. Both individuals could also consider those donations to their religion. It is not entirely clear how to disentangle donations to charitable organizations that have a religious affiliation (e.g., Notre Dame University, Adventist hospitals, or Church World Service). It may be impossible to disentangle such donations.
Conclusion
This study examined whether nonreligious individuals are less likely to engage in charitable giving than their religious counterparts, based on data from the 2023 wave of the PSID. Overall, our results suggest that the nonreligious are not inherently less likely to give than the religious, and that secularization in the United States has not led to a decline in charitable giving. This study contributes to the literature by carefully analyzing the giving patterns of nonreligious Americans in comparison to religious Americans. We found that religious Americans give significantly more money to religions, but there are largely negligible differences in giving to secular charities and organizations. Our findings run counter to some prior studies that have found differences in giving between the religious and nonreligious, most notably the assertion made by Putnam and Campbell (2012) in their study that the religious are simply more generous than the nonreligious.
Our findings are important for two reasons. Assuming the US population continues to secularize, it may be the case that donations to religions will eventually slow or even decline. However, donations to secular charities and nonprofits will likely continue to increase as secular individuals give to organizations that share their values.
Future studies could add to our findings by exploring further variations within the nonreligious population. There are other constituent groups beyond atheists and the nones, such as agnostics, the spiritual but not religious, and those occupying ambiguous states between the religious and secular (Smith and Dougherty 2025). Although such studies would likely require more complicated analyses, they could yield interesting insights regarding (non)religious generosity as well as shed more light on what researchers might expect in the future given America’s increasingly secular landscape. More concretely, studies that seek to disentangle donations to religiously affiliated institutions from donations to explicitly secular ones would likewise be useful. Finally, longitudinal and qualitative analyses would help us better understand how ongoing secularization in the United States is likely to shape charitable giving patterns over time.
Data Accessibility Statement
The data used for this study, the Panel Study of Income Dynamics, is available here: https://psidonline.isr.umich.edu/. Our analyses of the data constitute secondary data analysis. R code for the analyses in this paper is available here: https://osf.io/c835p/overview?view_only=d3cccf53be1948cda9f50b1c5b78110e.
Additional Files
The additional files for this article can be found as follows:
Supplementary File 1
Figures S1A to S1I and Tables S1A to S1I. DOI: https://doi.org/10.5334/snr.164.s1
Supplementary File 2
Figures S2A to S2I and Tables S2A to S2I. DOI: https://doi.org/10.5334/snr.164.s2
Notes
[2] Throughout this paper, we use “nonreligious” to refer to all individuals with no religious affiliation – atheists, agnostics, and those who don’t report a religious affiliation but don’t provide additional information. We use “nones” to refer to the “nothing in particular” group identified by Pew Research Center and distinguish them from those who self-identify as “atheists” or “agnostics,” who we often refer to as just “atheists” for parsimony. As such, our definitions of “religious” and “nonreligious” rely on identification rather than behaviors or personal beliefs. We acknowledge it’s more complicated than this, and that some people we classify as nonreligious will hold some religious beliefs or engage in some behaviors that could be construed as religious. Our focus on identification is a function of what the data set used can tell us. Despite this limitation, (non)religious identification is a useful proxy for making claims about religious and nonreligious groups in society.
[3] We estimated that, for religious individuals to compensate for lower charitable giving by the nonreligious, they would have had to increase their giving by roughly 30% on average in order for charitable giving to all organizations to remain at 1990 levels.
[4] This is not the only definition or set of criteria for what constitutes “religious” and “nonreligious,” but this is what we mean when we invoke the terms throughout this paper.
[5] The original version of this paper used the 2019 wave of the PSID and found very similar results. Delays in revising the manuscript resulted in a shift from the 2019 to the 2023 wave. We note that the results were similar when analyzing two waves of the PSID to suggest that our findings may not reflect a new trend but rather that more sophisticated analyses are better able to illustrate the relationship between (non)religiosity and charitable giving.
[6] Note that Atheist/Agnostic is not a religious affiliation but rather a belief (or lack of belief) toward the existence of a god or higher power. We recognize this but retain it as a separate category because it distinguishes between those who are “nothing in particular” (a.k.a. “nones” in this paper) and those who are affirmatively secular.
[7] We attempted to create marginal effects models using just the highest values for income and wealth (99th percentile). Even when setting income and wealth that high, the predicted amounts in the figures were negative, illustrating just how influential the donations of one or two very wealthy individuals can be.
Ethics and Consent
We did not obtain IRB approval for this project because we were analyzing secondary data that were anonymized.
Acknowledgements
We thank the editor and anonymous reviewers for their helpful feedback on this paper.
Competing Interests
One of the authors of this article, David Speed, is an editor of the journal Secularism & Nonreligion. However, a different editor (Isabella Kasselstrand) managed this paper throughout the review process and David Speed was only involved as an author and not in the review process.
Author Contributions
Cragun and Rodriguez conceptualized the paper. Rodriguez wrote the original literature review, which was then edited and added to by all four authors. Cragun conducted the analyses, with helpful feedback and suggestions from the other three authors. Smith and Speed provided vital feedback and edited the manuscript multiple times. Revisions were primarily made by Cragun, though all revisions were edited and approved by all authors.