Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The Dark Money Subsidy? Tax Policy and Donations to Section 501(c)(4) Organizations

The Dark Money Subsidy? Tax Policy and Donations to Section 501(c)(4) Organizations Abstract This article presents the first empirical examination of giving to §501(c)(4) organizations, which have recently become important players in U.S. politics. Unlike gifts to charity, donations to a 501(c)(4) are not legally deductible. Yet, gifts to c(4) organizations are highly elastic to the after-tax price of charitable giving. At the lower end of the observed tax price range, c(4) giving falls with tax price, consistent with the hypothesis that giving to c(3) and c(4) organizations are substitutes. Over the top quarter of the distribution of tax price, however, gifts to c(4) organizations are negatively correlated with the after-tax price of giving to charity. That is, donors appear to respond as though the deduction subsidized their gift to a c(4). Donor responses to benefits for which they are not eligible may reflect the low salience of legal limitations or deliberate overclaiming. These results imply subsidies for charity can crowd out or in donations to c(4) organizations, with potential implications for U.S. politics. I cannot observe whether donors claim tax deductions for ineligible gifts, so the net results for the Treasury are unclear. 1. Introduction Although they have grown markedly in social importance since 2010, so-called “social welfare organizations” are seldom studied, with little known about their activities and finances. They are often better known as “c(4)’s,” as §501(c)(4) of the Tax Code grants these firms exemption from the U.S. corporate income tax. In contrast to more traditional charities exempted under §501(c)(3), donations to a 501(c)(4) are not deductible by donors. Many c(4) entities are community organizations, such as youth recreational associations or volunteer fire departments, that have failed to achieve charitable status because of tax-law technicalities. Others, however, are advocacy organizations, and it is these that have lent c(4) its newfound importance. In this article, I suggest that tax policy may have played a role in building the c(4) sector—not only by offering attractive regulatory features unavailable to charities but also by delivering accidental subsidies. Because they have become important political spending vehicles (Maguire, 2014), study of how these entities get their money should be of increasing scholarly interest. Yet although there are hundreds of papers examining the determinants of contributions to charitable organizations (see Bakija, 2013 for a review), none consider what factors encourage support for their c(4) cousins. Using several alternative measures of the after-tax price of donating to charity, I find that gifts to c(4) organizations in fact respond to changes in that price, even though donations to c(4) entities are not deductible. By including both linear and quadratic terms of the after-tax price of charitable giving (herein “tax price”) in the regression analysis, I show that the tax price is positively correlated with donations at the lower end of the tax price range, uncorrelated at the middle of the distribution, and negatively correlated in roughly the top quarter of the distribution of state-years. That is, for a substantial portion of the firm-years I observe, the charitable contribution deduction on net correlates with increased donations to ineligible c(4) recipients, to a statistically significant and economically large degree, with elasticities greater than |$-1$| in absolute magnitude. Since I rely on organizational data, not individual tax returns, I cannot tell whether this increased giving also results in a tax benefit—that is, whether donors are incorrectly claiming charitable contributions for gifts to a c(4). The c(4) organization has become a popular spending choice because of its unique regulatory status. While the c(4) firm must be “devoted exclusively to charitable, educational, or recreational purposes,” IRS interpretations of that limit allow essentially unlimited lobbying in pursuit of charitable, educational, or recreational ends (Halperin, 2014). Social welfare firms can also spend some limited fraction of their resources on campaigns for elective office, with the exact limits a matter of ongoing dispute (Aprill, 2012; Dougherty, 2013). In contrast, U.S. charities— §501(c)(3) organizations—cannot engage in “substantial” lobbying efforts, and may not electioneer at all. Until 2010, spending by social welfare organizations was somewhat constrained by federal election-law limits on firms other than candidates’ committees, but the Supreme Court’s decision in Citizens United, together with several key lower court decisions around the same time, released those shackles (Aprill, 2011; Galston, 2011). Since then, spending by “dark money” organizations, so-called because their donors are secret by law (IRC §§6103, 6104), and many with tax exemption under §501(c)(4), has skyrocketed (Maguire, 2014). Despite the increasing social importance of social welfare organizations, little is known about them. Few scholars have ever written systematically about c(4) firms. The rationales for the existence of c(4) as a category, and whether qualifying organizations should enjoy the subsidies it is known to provide, are few. Only Dougherty (2013), Halperin (2018), and Hackney (2020) explore these topics in any depth, with recent white papers by Aprill (2018) and Mayer (2018) also making important advances. Empirical work on social welfare organizations is equally scant, and largely focused on political activity (Horton Smith (1997) remarks that noncharitable exempt entities are part of the “dark matter” of unstudied voluntary associations). Chand (2017) looks at the lobbying activities of the subset of related c(3) and c(4) organizations that issue legislative scorecards. Child and Gronbjerg (2007) report survey evidence that noncharitable exempt organizations are more likely to engage in policy advocacy, a result Dimmery and Peterson (2016, 61) confirm using massive scraping of firms’ web sites. Nicholson-Crotty (2007,2009) employs collaboration with a c(4) as a measure of the political engagement of women’s health centers, while Kerlin and Reid (2010) drill deep using case studies of five environmental organizations that make use of multiple tax-exempt entities to conduct their advocacy work. As Koulish (2016) and Dimmery and Peterson (2016) show, however, most social welfare organizations are not political. Koulish (2016) provides the only available overview of the resources and activities of the whole subsector. In a more focused effort, Hofmann (2007) finds accounting evidence that noncharitable exempt organizations, including some social welfare firms, shift expenses to minimize the unrelated business income tax. At first glance it is unclear why the charitable contribution deduction would affect 501(c)(4) firms. Close observers have recognized that federal rules exempting 501(c)(4) income from the corporate income tax, and permitting donors of appreciated property to escape tax on built-in gains existing at the time of contribution, may encourage gifts, especially gifts in kind (Halperin, 2018). But investment earnings spent on political activity are subject to tax under §527(f) of the Tax Code, reducing this benefit. No one has suggested that the charitable contribution deduction might affect donations to a c(4), which again are not legally deductible. If anything, one might expect that the deduction would diminish giving to 501(c)(4) organizations. If charity and c(4) firms are substitutes, a decline in the price of giving to charity should (holding household income roughly level) shift a portion of the household budget towards charities. Charitable giving is a tiny fraction of average annual household spending (Giving USA 2018), so modest changes in the price of giving should not have meaningful income effects. Yet behavioral economics and past findings in tax compliance suggest a different potential story. As I will detail more below, c(3) and c(4) firms carry out similar missions under similar names, and tax rules for distinguishing them are likely highly opaque to most taxpayers. Feldman et al. (2016), Goldin and Listokin (2013), and Gallagher and Muehlegger (2011) have found evidence of taxpayers who overvalue various forms of tax subsidy. Perhaps some donors give to c(4) organizations believing they are making a charitable contribution. Alternately, some donors may understand the difference but conclude their erroneous claims are unlikely to be detected. Slemrod (1989) finds evidence of substantial overclaiming by charitable contributors. IRS procedures are poorly suited to audit whether deductions that are claimed in fact flow to eligible entities, especially for small cash gifts. I argue that my results can be reconciled with all three of these theoretical predictions about the effect of the charitable contribution deduction. When the tax benefits of supporting charity are relatively large, we observe a classic substitution effect towards charity and away from c(4) organizations. When tax benefits are smaller, however, subsidies for charity are positively correlated with giving to c(4)’s. This is consistent with, for example, the determination in Slemrod (1989) that misreporting is relatively inelastic to tax rates, as well as his finding (1989: Table 2) that the rates of misreported charitable contributions are hump-shaped in income.1 I emphasize, however, that I cannot verify whether taxpayers in fact claim donations to a c(4) as deductible on their returns. My findings would also fit a model in which donors are more apt to be confused when the stakes of confusion are smaller. It could be argued that the observed donation patterns might result from strategic firm behavior, especially fundraising. When I control for fundraising in regressions in which donations are the outcome variable, the estimated coefficients across the distribution of tax price are smaller, but for most of the distribution still statistically significant and larger than one in absolute magnitude.2 Further, when I repeat my analysis in a larger but less precisely measured set of data, I find that controlling for fundraising makes no difference in the price-elasticity of “public support,” a category that includes donations. In short, I find possible evidence of an unintended and previously unknown subsidy for social welfare organizations. As Faulhaber (2012) and Galle (2013a) observe, donors who over-estimate the value of the charitable contribution deduction may improve social welfare by providing public goods without the need for public support. Here, however, it is much less clear that any subsidy effect is salutary. For one, these donations may actually reduce taxes paid, as donors may well be erroneously deducting their gifts to c(4) organizations. In addition, Congress opted not to underwrite gifts to firms that can lobby and electioneer extensively with an unlimited matching grant, a subsidy structure that would tend to heavily favor wealthy interests (Galle, 2013b; Hasen, 1996; Tobin, 2007). Even if donations are not actually claimed, to the extent that misunderstandings encourage gifts to lobbying organizations, they may run contrary to good policy. While the implications are not as dramatic if fundraising is the main driver of my results, I still provide an initial window into the fiscal behavior of social welfare firms. The donative environment has a major impact on a c(4) organization. Changes to rules applicable to charities also impact c(4)’s, and the impact on net social welfare can potentially be opposite in sign from the charity-focused policy. 2. Background 2.1. What is a social welfare organization? Section 501(c)(4) serves mainly as a fallback option for entities that fail one of the qualifying tests for §501(c)(3) (Halperin, 2018). The statutory definition of “social welfare” has no clear meaning, and IRS guidance seems to define it as roughly parallel to the charitable purposes common to c(3) organizations (Dougherty, 2013). As Aprill (2013, 2018) notes, many social welfare organizations in fact resemble charities, but are disqualified from obtaining charitable status by their failure to serve a wide enough “charitable class”—either an indefinite group of beneficiaries or a definite but very large group. The organization cannot, however, serve purely private interests, suggesting that it must be organized on behalf of some small or mid-sized “community” (Dougherty, 2013). Arguably, government supports for organizations of this kind can serve as a mechanism for private allocation of public goods (Ayres, 2017). Empirically, social welfare firms are often community organizations, such as an ambulance corps or youth recreational association (Koulish, 2016). Many of the largest c(4)s in terms of assets held are health cooperatives, such as state HMO Blue and Delta Dental organizations, which were denied c(3) status but left eligible to be c(4)s by a series of IRS and judicial rulings (Mancino, 2005). To give a bit more sense of the makeup of the population, I generated a frequency plot of the words most commonly appearing in the names of c(4) organizations in the sample. The top words were, in order: fire, association, volunteer, housing, development, club, health, school, county, and community. While social welfare organizations cannot provide any substantial benefits to private individuals or firms, they can engage in extensive political activity. A c(4) can lobby without limit in pursuit of its social welfare mission. Commentators disagree about the extent of a c(4)’s permissible spending for political campaigns, with the most aggressive practitioners asserting that spending of up to 49% of firm income is permissible (for an overview of both sides, see Dougherty, 2013). Because donations to social welfare organizations are protected by tax secrecy laws, the c(4) is an attractive spending vehicle for actors who prefer to make anonymous political expenditures. Political organizations probably make up less than one-quarter of c(4) firms, however (Dimmery and Peterson, 2016, p. 60). Commonly, sophisticated charities that believe political engagement is part of their mission have a related c(4), often with partly or fully overlapping boards, staff, and assets (Kerlin and Reid, 2010). In some firms, the division is entirely on paper. Staff time and other expenditures can be booked to the c(4) when political in nature and to the c(3) otherwise (Leff, 2009). Similarly, some advocacy organizations maintain a related c(3) to serve as a recipient of deductible contributions and grants from private funders. A charity can transfer funds to a c(4) without limit, but cannot earmark transferred funds for activities it could not conduct itself. Most commentators believe that Supreme Court rulings from the early 1980s require the IRS to accept this structure, even if it permits c(3) firms to pass through considerable value to related c(4)s (for a dissenting view on what the Court’s holdings require, see Galle, 2013b). In many advocacy organizations, a c(4) further shares resources and staff with a political action committee (Kerlin and Reid, 2010). These arrangements can be a source of “dark money,” as the c(4) shields donors’ funds from disclosure that would otherwise apply to PAC contributions (Dougherty, 2013). The other core difference between §501(c)(3) and §501(c)(4) is the ability to receive deductible charitable contributions. Both sets of organizations are exempt from the U.S. corporate income tax, thereby subsidizing contributions of investment property, except to the extent investment income is spent on political activity (T.C. §527(f); Halperin, 2011,2018). Further, contributions to either type of firm are not treated as a taxable event for the donor, so that transfers of assets with untaxed appreciation effectively eliminate any tax on the asset’s built-in gains.3 In essence, the donor enjoys a zero tax rate on any property used to fund gifts to charity or social welfare groups.4 Most donated property is reported to have very low basis (Ackerman and Auten, 2011, p. 660), so the value of this exclusion may approximate the tax due on the full value of the property. Whether the average donor appreciates the distinction between these two entity forms is uncertain. Often, similarly named firms can appear in either category, such as the ACLU (a c(4)) and the ACLU Foundation (a charity). A 1987 federal law requires that mass tv, print, phone, or radio solicitations by noncharitable firms must include a “prominent” statement that gifts are not tax deductible (T.C. §6113; see IRS Notice 88-120 for additional detail). Application of these rules to the internet remains unclear. An investigation of a random sample of 20 c(4) organizations determined that all disclosed the nondeductibility of gifts somewhere on their website, but often this information was located at the bottom of a page reached only after multiple clicks. The IRS offers little guidance in its instructions for the individual tax return. The instructions to schedule A of the Form 1040 do not mention §501(c)(4) or even social welfare organizations (IRS, 2006a).5 Instead, the instructions offer examples of organizations that would be eligible for donations: “Boy Scouts, Boys and Girls Club of America, CARE, Girl Scouts, Goodwill Industries, Red Cross, Salvation Army, United Way, etc.” The closest the instructions come to identifying the existence of ineligible donee firms is when they note that “[g]ifts to civic leagues, social and sports clubs, labor unions, and chambers of commerce” are not deductible. These appear to be references to firms eligible for tax exemption, but not deductible contributions, under sections 501(c)(5), (6), (7), and (8). It is also unclear whether the IRS can reliably identify donations to social welfare organizations that are erroneously (or deliberately) deducted. Donors must keep records evidencing their donations in the event of audit, but generally need not disclose the identity of donee firms on their tax return (IRS, 2006a). In-kind gifts of more than |${\$}$|500 must be separately reported on the Form 8283, where the taxpayer must provide the name of the recipient organization (IRS, 2006b). Donee organizations during the period of my sample reported all contributors of |${\$}$|5,000 or more on Schedule B of the Form 990 organizational tax return, but it is unknown whether IRS had the technical capability to match the Schedule B to individual donor returns. Even in the event an individual were audited and required to show their donation records, the auditor would need to take the additional step of individually verifying the charitable status of each donee firm; since firm names do not clearly reveal an organization’s status, it is not obvious why auditors would do so (unless they had read this article).6 In July of 2018, the IRS announced a new policy in which it would not require 501(c)(4) organizations to report their donors to the government, making verification even more challenging (Rev. Proc. 2018-38). Notably, the accompanying IRS press release states that there is no administrative need for donor information (U.S. Department of the Treasury, 2018), a premise my results suggest may be mistaken. 2.2. Hypotheses: Charitable Contribution Deductions and c(4) Firms These facts suggest a potentially complex relationship between government subsidies for charitable contributions and donations to c(4) organizations. Most straightforwardly, if gifts to c(3) and c(4) organizations are (uncompensated) substitutes, donations to c(4) firms should fall when giving to charity is relatively less expensive. There is no theoretical reason to expect that donations to the two forms would be instead be complements, except in the case where one organization is conducting business through linked c(3) and c(4) firms.7 Therefore, as the after-tax price of charitable giving falls, donors who understand and abide by deductibility rules should more strongly prefer to make a deductible contribution to a c(3) entity rather than a nondeductible contribution to a c(4).8 Deliberate misclaiming could instead produce a negative correlation between the tax price of charitable giving and donations to a c(4). In the model described informally in Slemrod (1989), taxpayers’ propensity to overclaim charitable contribution deductions is a function of tax rates, detection probability, and magnitude of sanction (see also Feldstein, 1999, which offers a more general model in which taxpayers adjust their reporting to claim tax benefits until the marginal cost of doing so exceeds marginal returns). It appears that at least for cash contributions of under |${\$}$|5,000, and perhaps more, the ability of the IRS to detect c(4) donations reported as charitable contributions is small, and may be near zero. This should be evident to taxpayers who can comprehend the instructions on IRS forms. Slemrod’s model therefore predicts a positive association between tax rates and misreported c(4) contributions, though empirically he finds that the misreporting elasticity is rather smaller in magnitude than the price-elasticity of actual charitable giving. Behavioral models in which taxpayers understand the tax system imperfectly could also produce a negative correlation between c(4) contributions and the tax price of charitable giving. Again, several prior studies find that taxpayers can be overly responsive to complex tax incentives because they fail to understand legal limits on those incentives (Gallagher and Muehlegger, 2011; Goldin and Listokin, 2013; Katuscak and Kawano, 2016). Eckel and Grossman (2017) also report a field experiment in which they find donors have varying awareness of government subsidies. As in those settings, charitable giving is a complex regime in which the average donor is presented with very limited information about whether her contribution is deductible. For misinformed donors, we might expect that contributions to c(4)’s will mirror their giving to charities and have similar price-elasticities. The prediction that price-elasticities of giving will be similar for both organizational forms assumes, though, that donor confusion is exogenous to tax price. Chetty et al. (2009) suggest that high stakes may motivate greater investment in learning to understand the tax system. That is, suppose that there are two groups of givers in the population, those who understand tax rules and those who do not. Following Chetty et al. (2009), we might call these naïves and sophisticates. Observable population-level elasticities for giving to c(4)’s will reflect a weighted mix of highly negative-elasticity naïves and zero- or positive-elasticity sophisticates. The “high stakes” hypothesis is that the share of sophisticates in the population is larger when the marginal tax savings of a charitable contribution deduction is greater. Accordingly, the high-stakes prediction would be that observed elasticities would be closer to zero when the tax price of giving is closer to unity (i.e. if |$\beta _s $| is the charitable-price-elasticity of giving to a c(4) organization, and is typically negative, then |$\frac{\partial \beta _s} {\partial p}$| is positive). Empirical evidence for the high-stakes theory is mixed, however, and largely derives from experimental settings. Taubinsky and Rees-Jones (2017) find that very large increases in tax rates increase consumer attentiveness to taxes. Feldman et al. (2018) find to the contrary, with higher rates actually lowering attentiveness, albeit over less-dramatic variation in simulated tax rates. They suggest that this effect may be produced by confirmation bias: consumers selectively ignore tax when it is inconsistent with their pre-tax preferences (Galle, 2013a describes this theory in more detail). In sum, we should not expect the price-elasticity of donations to c(4) organizations to be constant across the distribution of tax prices, but instead that donations will have a nonlinear relationship to tax price. Although the rival behavioral theories make contending predictions about how elasticity will vary with tax price, my setting likely does not allow for clean identification of one or the other. Contending factors within a purely rational framework—substitution effects and misreporting—already may tend to produce elasticities that are positive over some range of the distribution and negative over others, confounding any effort to test the behavioral hypotheses. In any event, these likely nonlinearities motivate my use of higher-order terms of tax price in the regression analysis, as detailed more below. 3. Data Data on the fiscal behavior of social welfare organizations are drawn from individual Form 990 tax returns filed annually by exempt organizations and compiled by the National Center on Charitable Statistics in their SOI-Other files. I collect data for tax years ranging from 1991 to 2007.9 My main outcome variable of interest is “direct public support,” or donations received from the public and not via charities or other donation aggregators. Each annual SOI-Other file is a stratified sample of the population of tax returns filed by noncharitable exempt organizations, with overweighting of firms with greater assets; firms with more than |${\$}$|10 million in assets are always included. Stratification naturally raises concerns about whether the criteria for weighting may be correlated with outcomes of interest. My results are robust to sample weighting. The alternative source of information is the Core-Other files, also available via NCCS, which provide a more limited set of variables for the full population of firms. Unfortunately, the Core files fail to disaggregate individual contributions from government grants and “indirect” support, or contributions from other firms, but I employ them for robustness testing. I clean the data following protocols described in Galle (2016). These steps result in the omission of 608 firm-year observations, leaving a total of 21,040 firm-years for c(4) organizations. I also restate negative expense items at their absolute value, resulting in 148 changes. Most of the regression results limit the sample to firms that ever report receiving donations, yielding 9,186 firm-years. For some regression analysis, I attempt to match social welfare organizations with related c(3) firms. Guided by a fuzzy matching algorithm based on firm names and zip codes, I hand match c(4) entities in the SOI-Other database to related charitable organizations in the IRS Business Master File listing of all approved charities. I code a social welfare firm as paired with a matching charity if either entity reports the other as “related” on Schedule R of its tax return. I also code firms as related if they share a common address and identify one or more common individuals in their listings of highly compensated employees or board members. Finally, I code two firms as related if either makes reference to the other on its official web page.10 Ultimately, I have 314 firm-year c(4) observations in the SOI-Other file with matched c(3) data. For analysis using the Core-Other files, I repeated this process using the much larger set of Core firms, yielding 1,558 matched firm-years. I measure the after-tax cost of donating to charity using three alternative but similar measures. In the reported regressions, I employ the state-year price facing taxpayers in each state’s top tax bracket, as computed by Bakija and Heim (2011) using the calculator described in Bakija (2009). This variable, which I call the “after-tax price” or (following the standard term in the literature) “tax price of giving,” represents the net cost of an additional dollar of charitable contributions in a given state and year, taking account of the combined impact of state and federal taxes. A larger charitable contribution deduction results in a lower after-tax price. As the Supplementary Appendix Section A.1 describes in more detail, most of the within-state tax price variation in the sample is caused by changing federal law and its interactions with cross-sectional variations in state rules. For instance, increases in federal marginal rates reduce the value of a state charitable contribution deduction, but for a set of “control” states with no income tax or no deduction, those alterations have no impact. Figure 1 plots the log after-tax price of giving in these two groups of states over time. The vertical line at 2002 represents the January 1, 2003 effective date of “JGTRRA,” the Jobs Growth and Tax Relief Reconciliation Act of 2003. Figure 1. Open in new tabDownload slide Log After-Tax Price of Giving in States With and Without Income Tax. Notes: Prices shown are mean cost of donating |${\$}$|1 to charity, net of state and federal tax, for simulated top-bracket taxpayers in each state. Prices derived from Bakija and Heim (2011). Figure 1. Open in new tabDownload slide Log After-Tax Price of Giving in States With and Without Income Tax. Notes: Prices shown are mean cost of donating |${\$}$|1 to charity, net of state and federal tax, for simulated top-bracket taxpayers in each state. Prices derived from Bakija and Heim (2011). For robustness, I also use the dollar-weighted mean tax price paid by all taxpayers in each state-year, as computed by Duquette (2016). A third method is described in Supplementary Appendix Section A.1. I additionally control for capital gains tax rates. State and federal top effective capital gains rates are taken from NBER taxsim calculations, as summarized in Feenberg and Coutts (1993). In essence, all of these sets of tax data draw a random sample of exogenously itemizing 1984 taxpayers, deflate their data to a subsequent tax year, and compute the marginal effect on tax liability of an additional dollar of contribution or capital gains, respectively, under combined state and federal tax rules applicable in that year.11 I compute average yearly S&P 500 value using monthly S&P means from Robert Schiller’s web site, with each firm-year observation representing the mean of the 12 monthly observations falling within the firm’s fiscal year. All dollar values are deflated to 2007 dollars using the chained-CPI index for the year and month ending the firm’s fiscal year. I obtain state demographic and fiscal variables from the U.S. Census. For variables reported on a calendar-year basis, including year fixed effects, I match firm fiscal years to calendar year based on the calendar year in which a majority of the fiscal-year months fall. Fiscal years ending in June are assigned to the prior calendar year. Table 1 provides simple summary data of the variables I employ in the regression analysis, with separate means for firms that do or do not report receiving donations during the sample period. Table 1. Summary Statistics . All Firms . Firms with Donations . Variables . Mean . SD . Mean . SD . Firm variables Direct public support 628.5 13.2|$^{*}$| 1.44|$^{*}$| 19.9|$^{*}$| Indirect support 165.5 4.3|$^{*}$| 171.5 4.37|$^{*}$| Grants 984.8 22.0|$^{*}$| 1.25|$^{*}$| 26.0|$^{*}$| Program Svc. revenues 29.8|$^{*}$| 203|$^{*}$| 8.47|$^{*}$| 61.3|$^{*}$| Interest revenue 309.1 1.95|$^{*}$| 106.9 984.3 Dividend revenue 485.2 8.3|$^{*}$| 296.1 2.78|$^{*}$| Total revenues 34.9|$^{*}$| 215|$^{*}$| 13.3|$^{*}$| 74.8|$^{*}$| Officer compensation 153.4 1.2|$^{*}$| 89.8 524.6 Program Svc. expenditures 30.9|$^{*}$| 194|$^{*}$| 11.0|$^{*}$| 66.5|$^{*}$| Management expenses 2.6|$^{*}$| 20.7|$^{*}$| 1.19|$^{*}$| 7.82|$^{*}$| Fundraising expenses 102.6 1.4|$^{*}$| 232.8 2.07|$^{*}$| Outside fundraising exp. 8.2 143.5 18.6 216.7 Gross assets 48.0|$^{*}$| 547|$^{*}$| 15.2|$^{*}$| 69.3|$^{*}$| Liabilities 33.8|$^{*}$| 520|$^{*}$| 6.05|$^{*}$| 35.7|$^{*}$| Fundraising share .016 .080 .031 10.6 S&P 12-month avg. 929.11 377.0 961.6 365.3 State variables State tax price |$-$|0.086 0.042 |$-$|0.086 0.040 Population 11.2|$^{*}$| 9.8|$^{*}$| 10.4|$^{*}$| 8.7|$^{*}$| Share under 26 0.36 0.02 0.35 0.02 Share over 64 0.13 0.019 0.13 0.020 Net capital gains rate 26.5 5.40 26.0 5.26 . All Firms . Firms with Donations . Variables . Mean . SD . Mean . SD . Firm variables Direct public support 628.5 13.2|$^{*}$| 1.44|$^{*}$| 19.9|$^{*}$| Indirect support 165.5 4.3|$^{*}$| 171.5 4.37|$^{*}$| Grants 984.8 22.0|$^{*}$| 1.25|$^{*}$| 26.0|$^{*}$| Program Svc. revenues 29.8|$^{*}$| 203|$^{*}$| 8.47|$^{*}$| 61.3|$^{*}$| Interest revenue 309.1 1.95|$^{*}$| 106.9 984.3 Dividend revenue 485.2 8.3|$^{*}$| 296.1 2.78|$^{*}$| Total revenues 34.9|$^{*}$| 215|$^{*}$| 13.3|$^{*}$| 74.8|$^{*}$| Officer compensation 153.4 1.2|$^{*}$| 89.8 524.6 Program Svc. expenditures 30.9|$^{*}$| 194|$^{*}$| 11.0|$^{*}$| 66.5|$^{*}$| Management expenses 2.6|$^{*}$| 20.7|$^{*}$| 1.19|$^{*}$| 7.82|$^{*}$| Fundraising expenses 102.6 1.4|$^{*}$| 232.8 2.07|$^{*}$| Outside fundraising exp. 8.2 143.5 18.6 216.7 Gross assets 48.0|$^{*}$| 547|$^{*}$| 15.2|$^{*}$| 69.3|$^{*}$| Liabilities 33.8|$^{*}$| 520|$^{*}$| 6.05|$^{*}$| 35.7|$^{*}$| Fundraising share .016 .080 .031 10.6 S&P 12-month avg. 929.11 377.0 961.6 365.3 State variables State tax price |$-$|0.086 0.042 |$-$|0.086 0.040 Population 11.2|$^{*}$| 9.8|$^{*}$| 10.4|$^{*}$| 8.7|$^{*}$| Share under 26 0.36 0.02 0.35 0.02 Share over 64 0.13 0.019 0.13 0.020 Net capital gains rate 26.5 5.40 26.0 5.26 Notes: Dollar values in thousands of 2007 dollars. *millions Open in new tab Table 1. Summary Statistics . All Firms . Firms with Donations . Variables . Mean . SD . Mean . SD . Firm variables Direct public support 628.5 13.2|$^{*}$| 1.44|$^{*}$| 19.9|$^{*}$| Indirect support 165.5 4.3|$^{*}$| 171.5 4.37|$^{*}$| Grants 984.8 22.0|$^{*}$| 1.25|$^{*}$| 26.0|$^{*}$| Program Svc. revenues 29.8|$^{*}$| 203|$^{*}$| 8.47|$^{*}$| 61.3|$^{*}$| Interest revenue 309.1 1.95|$^{*}$| 106.9 984.3 Dividend revenue 485.2 8.3|$^{*}$| 296.1 2.78|$^{*}$| Total revenues 34.9|$^{*}$| 215|$^{*}$| 13.3|$^{*}$| 74.8|$^{*}$| Officer compensation 153.4 1.2|$^{*}$| 89.8 524.6 Program Svc. expenditures 30.9|$^{*}$| 194|$^{*}$| 11.0|$^{*}$| 66.5|$^{*}$| Management expenses 2.6|$^{*}$| 20.7|$^{*}$| 1.19|$^{*}$| 7.82|$^{*}$| Fundraising expenses 102.6 1.4|$^{*}$| 232.8 2.07|$^{*}$| Outside fundraising exp. 8.2 143.5 18.6 216.7 Gross assets 48.0|$^{*}$| 547|$^{*}$| 15.2|$^{*}$| 69.3|$^{*}$| Liabilities 33.8|$^{*}$| 520|$^{*}$| 6.05|$^{*}$| 35.7|$^{*}$| Fundraising share .016 .080 .031 10.6 S&P 12-month avg. 929.11 377.0 961.6 365.3 State variables State tax price |$-$|0.086 0.042 |$-$|0.086 0.040 Population 11.2|$^{*}$| 9.8|$^{*}$| 10.4|$^{*}$| 8.7|$^{*}$| Share under 26 0.36 0.02 0.35 0.02 Share over 64 0.13 0.019 0.13 0.020 Net capital gains rate 26.5 5.40 26.0 5.26 . All Firms . Firms with Donations . Variables . Mean . SD . Mean . SD . Firm variables Direct public support 628.5 13.2|$^{*}$| 1.44|$^{*}$| 19.9|$^{*}$| Indirect support 165.5 4.3|$^{*}$| 171.5 4.37|$^{*}$| Grants 984.8 22.0|$^{*}$| 1.25|$^{*}$| 26.0|$^{*}$| Program Svc. revenues 29.8|$^{*}$| 203|$^{*}$| 8.47|$^{*}$| 61.3|$^{*}$| Interest revenue 309.1 1.95|$^{*}$| 106.9 984.3 Dividend revenue 485.2 8.3|$^{*}$| 296.1 2.78|$^{*}$| Total revenues 34.9|$^{*}$| 215|$^{*}$| 13.3|$^{*}$| 74.8|$^{*}$| Officer compensation 153.4 1.2|$^{*}$| 89.8 524.6 Program Svc. expenditures 30.9|$^{*}$| 194|$^{*}$| 11.0|$^{*}$| 66.5|$^{*}$| Management expenses 2.6|$^{*}$| 20.7|$^{*}$| 1.19|$^{*}$| 7.82|$^{*}$| Fundraising expenses 102.6 1.4|$^{*}$| 232.8 2.07|$^{*}$| Outside fundraising exp. 8.2 143.5 18.6 216.7 Gross assets 48.0|$^{*}$| 547|$^{*}$| 15.2|$^{*}$| 69.3|$^{*}$| Liabilities 33.8|$^{*}$| 520|$^{*}$| 6.05|$^{*}$| 35.7|$^{*}$| Fundraising share .016 .080 .031 10.6 S&P 12-month avg. 929.11 377.0 961.6 365.3 State variables State tax price |$-$|0.086 0.042 |$-$|0.086 0.040 Population 11.2|$^{*}$| 9.8|$^{*}$| 10.4|$^{*}$| 8.7|$^{*}$| Share under 26 0.36 0.02 0.35 0.02 Share over 64 0.13 0.019 0.13 0.020 Net capital gains rate 26.5 5.40 26.0 5.26 Notes: Dollar values in thousands of 2007 dollars. *millions Open in new tab 4. Identification Donation regressions in the main text are estimated using an equation of the form: $$\begin{align}\label{eq1} D_{it}&= \alpha _{i}+\beta _1 \textit{TaxPrice}_{st} +\beta _2 TaxPrice_{st} \ \ast\ \textit{TaxPrice}_{st} +\beta _3 CG_{st}\nonumber\\ &\quad +\beta _4 S\& P_{it} +\beta _5 S\& P_{it - 1} +\beta _6 X_{it} +\beta _7 W_{st} +\lambda _{t}+\phi _s \lambda _{t}+\varepsilon_{it}, \end{align}$$(1) where TaxPrice is the mean after-tax cost of donating to charity for top-bracket taxpayers in a given state-year, where |$s$| indexes states and |$t$| time, CG is the proxy for the net capital gain savings from excluding built-in gains, and S&P and S&P|$_{t-1}$| are the same-year and lagged values of the S&P variable, with |$i$| indexing firms.12|$X$| is a vector of firm controls, while |$W$| is a vector of state-level controls. |$\varphi _s \lambda _{t}$| is a set of state by year trends. Tax price and its square are my main variables of interest. In essence, I am looking at the correlation between within-firm changes in tax prices confronting in-state donors and the reported donations received by the firm. The results reported rely on the top-bracket tax price computed in Bakija and Heim (2011). That is, the tax price variable represents the mean cost of donating one dollar to charity, net of both federal and state charitable contribution deductions, for donors who face the highest tax rate in a given state-year.13 Alternatively, I also estimate results (not tabulated) using the dollar-weighted mean for all state-year taxpayers, as computed in Duquette (2016). Results are qualitatively similar using this estimate. As Yetman and Yetman (2012) do, I prefer the top-bracket tax price. Because there is little year-over-year variation in the tax price for most middle-bracket donors (Bakija and Heim, 2011), any effects are more difficult to identify using averages that include a large number of these donors. As Supplementary Appendix Section A.1 details, I also extend the Galle (2016) methodology for identifying the price-elasticity of giving to charities to donations to c(4) organizations. Galle (2016) uses a form of difference-in-differences approach to control nonparametrically for unobservables. Results using the tax price measure just described are almost identical to those obtained using the DD method. For simplicity, therefore, the main text omits the DD results. In a sense, however, even the vanilla tax price represents a kind of DD methodology. With year effects in the regression, states without an income tax should see little year-over-year variation in tax price, as the year effects will absorb the impact of federal tax changes. Firms in these states can thus be thought of as the control group.14 Identification accordingly depends on the assumption that there are no unobserved shocks correlated with both being located in a state with an income tax and with the timing of JGTRRA or the Pease phaseout, and which would not be accounted for with state by year trends. Donations to both c(3) and c(4) organizations allow the donor to exclude from tax any built-in gains from in-kind property. Bakija and Heim (2011, pp. 621–22) include in their price variable an estimate of the value of making gifts with unrealized built-in gain. I would like to identify separately the effects of permissible exclusion of built-in gain from impermissible “charitable” contribution deductions to 501(c)(4) organizations. I therefore include a separate control for these excluded capital gains, in order to isolate in the Bakija and Heim price the impact of the charitable contribution deduction.15 To measure capital gains rates, I attempt to replicate as closely as possible the calculation Bakija and Heim (2011) employ, but my data are organized at the firm level and not by individual donor. As they do, I use the combined state-federal marginal burden on capital gains for state taxpayers, as computed by NBER taxsim (described more in 3.0 above). I then modify this figure, multiplying it by the product, |$n_{bt}$| * |$s_{bt}$| * |$a$| * |$d$| * mcg|$_{bt + 1}$|⁠. As Bakija and Heim (2011) explain, their calculation of the value of untaxed appreciation derives from using the lead of the capital gains rate, mcg|$_{t+1}$|⁠, on the assumption that donors’ alternative to donation now is often a gift in a later year. They then multiply this value by |$n_{it}$|⁠, the share of gifts made in kind by like-bracket donors, then by |$s_{it}$|⁠, the portion of in-kind gifts made in stock for individuals in the same bracket, and finally by |$a$|⁠, the average historic gain-to-value ratio of donated stock. This figure is then discounted to present-year value at the rate |$d$|⁠. To replicate this method as closely as possible, I repeat these calculations, using top-bracket (⁠|$b)$| means for each year |$t$|⁠. I call the resulting variable “net capital gain rate.”16 The use of state-level tax variables may lead to some attenuation. Tax benefits are generally determined by the state of the donor, not the recipient organization. Therefore, a key assumption of my model is that most giving is within state, and that cross-state donations will tend to average out over time. A number of studies find that more than 75% of individual and foundation gifts are to within-state or otherwise spatially proximate organizations (Center on Philanthropy at Indiana University, 2008a, b; Glückler and Ries, 2012). Given that many c(4) organizations are community groups, that figure may overstate the geographic dispersion of many c(4) supporters.17 To test the extent to which attenuation contributes to the results, I also hand-code each donation-receiving c(4) firm for indicators of whether the firm provides regional or nation-wide services, and therefore is more likely to receive cross-state donations. Firms are coded based on geographic indicators in their titles, mission descriptions on their tax returns, or current web site. I code organizations as regional or national when these descriptors include terms such as national, nationwide, “across the United States,” global, and so on. Firms are also coded as national when it is clear from context that they serve members or beneficiaries across the country even if none of these exact geographic descriptors are used. Stock values and stock momentum can impact gifts to charities List and Peysakhovich 2010. To allow for the possibility that wealth effects and recent directional changes in stock values may have independent effects on giving, I control for lagged and same-year average value of the S&P 500. I further control for basic firm-level variables, such as investment revenue, officer compensation, liabilities, and lagged assets (since donations would affect same-year assets). In theory, if changes in the tax environment for charities affect giving to social welfare firms, those changes may also affect other firm outcomes, such as revenues from alternative sources, and especially fundraising expenditures (Supplementary Appendix Section A.2 provides evidence that tax price affects fundraising by c(4) firms). To avoid “bad controls” selection effects, therefore, I omit these outcome variables from reported regressions with donations on the left-hand side. It is possible that some secular trend might affect charitable giving among “treated” states—those with an income tax—but not untreated states. Such a trend could result in a spurious correlation between tax price and donations. I therefore control for state by year trends. Results are similar when excluding trends or including both linear and quadratic trends, although the specification in which I include controls for fundraising loses statistical significance with quadratic trends included. Additionally, I include calendar-year fixed effects and a set of state demographic variables, comprising GDP, unemployment rate, median income, population, and share of the population over 64 and under 26. Variables are estimated in logs, with the exception of the outcome variable and of course indicators. Many firms in the sample report zero donations or zero fundraising for some or all of their firm-years. When either of these variables appears on the left-hand side I do not log them. Instead, I estimate marginal effects using a fixed-effects panel negative binomial regression. Negative binomial estimation is often used for count variables, but since it is an exponential function, when the left-hand variable is estimated in levels and the right-hand variables are in logs, it can be interpreted similarly to, and substituted for, traditional log-log estimates (Woolridge, 2010; see Almunia et al., 2020 for application to charitable giving). It therefore avoids the need to add some arbitrary, and potential biasing, amount to zero outcomes before logging. The implementation of the fixed-effects panel negative binomial estimator in Stata omits firms for which the outcome variable is always zero. That approach makes theoretical sense in my sample. Many social welfare organizations, such as health insurance cooperatives, do not have a business model that involves donations. Including these firms would presumably bias any result towards zero. For similar reasons, Andreoni and Payne (2013) omit firms with zero reported gifts, grants, or fundraising from their sample (Duquette, 2016 similarly omits firms without donations). In regressions with fundraising on the left-hand side (Supplementary Appendix Section A.2), I omit firms that never report gifts as well as firms that never report fundraising. I cluster standard errors at the state level because that is the level of treatment. Due to technical limitations in Stata, I am obliged to implement clustering using wild bootstrap standard errors (Cameron et al., 2008). 4.1. Linear or Quadratic Estimates? Once more, theory suggests that the relationship between tax price and donations to 501(c)(4) organizations may not be linear. Graphical analysis confirms that quadratic estimates may better capture the relationship between tax price and donations.18Figure 2 plots 40 equal-width bins of log donations by their mean tax price, limiting the sample to c(4) firms that ever report receiving donations. A quadratic best-fit line through the bin means forms a parabola. Figure 2. Open in new tabDownload slide Scatterplot of Forty Equal-Width Bins of Residualized Log Donations vs. Log Tax Price Notes: Solid line = quadratic fit line. Bins defined by log donations and log tax price, residualized by controlling for firm and state fixed effects. Sample limited to 501(c)(4) organizations ever reporting nonzero donations. Figure 2. Open in new tabDownload slide Scatterplot of Forty Equal-Width Bins of Residualized Log Donations vs. Log Tax Price Notes: Solid line = quadratic fit line. Bins defined by log donations and log tax price, residualized by controlling for firm and state fixed effects. Sample limited to 501(c)(4) organizations ever reporting nonzero donations. As a result, in regressions for which donations are the outcome variable, I include both tax price and tax price squared as predictors. 5. Results I investigate the determinants of two sets of social welfare firm outcomes. Section 5.1 examines the predictors of donations, and Section 5.2 reports robustness testing using the larger Core-Other files. 5.1. Determinants of Direct Public Support I first consider the determinants of giving to social welfare organizations among firms that ever report receiving donations from the public. For my baseline estimates, I omit other “outcome” variables that might also be influenced by the tax price of giving, such as fundraising. These estimates thus correspond to what Galle (2016) dubs the “reduced-form” elasticity, |$E_{pD}^R $|⁠. It is also of interest, however, to separately identify the extent to which any effects may be caused by fundraising. The reduced-form elasticity can be decomposed into its structural components, as in equation 2: $$\begin{align}\label{eq2} E_{pD}^R = E_{pD}^S + E_{pF} \cdot E_{FD} \end{align}$$(2) where the right-hand side values are the “structural” components of donations. That is, |$E_{pD}^S $| is the price-elasticity of giving with fundraising held constant, or |$\frac{\partial D(P,F)}{\partial P}\cdot \frac{P}{D(P,F)}$| The value |$E_{pF} $| is the price-elasticity of fundraising, |$\frac{dF(P)}{dP}\cdot \frac{P}{F}$|⁠. And the last value, |$E_{FD} $|⁠, is the fundraising elasticity of donations, |$\frac{\partial D(P,F)}{\partial F} \cdot \frac{F}{D(P,F)}$|⁠.19 Columns 1 and 2 of Table 2 report specifications in which I directly estimate |$E_{pD}^R $|⁠, and in which I separately estimate |$E_{pD}^S$| and |$E_{FD} $|⁠, respectively. Again, in both cases I include both linear and quadratic terms for the tax price variables.20 Table 2. Effects of Tax Price on Donations to Social Welfare Organizations Variables . (1) . (2) . (3) . (4) . State tax price -10.84*** -6.513** -8.492** -10.50*** (3.368) (2.865) (3.525) (2.973) State tax price square -12.82*** -8.455** -10.04** -12.07*** (3.787) (3.302) (4.044) (3.424) Log net capital gains -0.459 -0.0162 -1.024* -1.216*** (0.887) (0.402) (0.523) (0.423) Log S&P 500 -0.0326 -0.231 -0.175 -0.0196 (0.365) (0.332) (0.458) (0.361) Lag of Log S&P 500 0.236 0.0327 -0.484 0.0506 (0.356) (0.318) (0.437) (0.340) Log fundraising 0.353*** (0.0170) Log fundraising share -0.177*** (0.0224) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of Tax Price 3.34*** 2.84*** 2.61** 2.84*** (1.16) (0.99) (1.20) (1.03) |$\quad$||$-1$|SD of Tax Price 2.31** 2.16** 1.81* 1.88** (0.95) (0.78) (0.95) (0.81) |$\quad$| Median Tax Price 0.47 0.94 0.365 0.144 (0.73) (0.54) (0.67) (0.56) |$\quad$||$+$|1 SD of Tax Price -1.66* -0.46 -1.30 -1.86*** (0.93) (0.70) (0.87) (0.73) |$\quad$||$+$|2 SD of Tax Price -3.17*** -1.46 -2.49** -3.28*** (1.25) (1.00) (1.24) (1.03) Observations 8,723 8,676 6,539 8,448 Number of Firms 1139 1139 933 1100 Variables . (1) . (2) . (3) . (4) . State tax price -10.84*** -6.513** -8.492** -10.50*** (3.368) (2.865) (3.525) (2.973) State tax price square -12.82*** -8.455** -10.04** -12.07*** (3.787) (3.302) (4.044) (3.424) Log net capital gains -0.459 -0.0162 -1.024* -1.216*** (0.887) (0.402) (0.523) (0.423) Log S&P 500 -0.0326 -0.231 -0.175 -0.0196 (0.365) (0.332) (0.458) (0.361) Lag of Log S&P 500 0.236 0.0327 -0.484 0.0506 (0.356) (0.318) (0.437) (0.340) Log fundraising 0.353*** (0.0170) Log fundraising share -0.177*** (0.0224) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of Tax Price 3.34*** 2.84*** 2.61** 2.84*** (1.16) (0.99) (1.20) (1.03) |$\quad$||$-1$|SD of Tax Price 2.31** 2.16** 1.81* 1.88** (0.95) (0.78) (0.95) (0.81) |$\quad$| Median Tax Price 0.47 0.94 0.365 0.144 (0.73) (0.54) (0.67) (0.56) |$\quad$||$+$|1 SD of Tax Price -1.66* -0.46 -1.30 -1.86*** (0.93) (0.70) (0.87) (0.73) |$\quad$||$+$|2 SD of Tax Price -3.17*** -1.46 -2.49** -3.28*** (1.25) (1.00) (1.24) (1.03) Observations 8,723 8,676 6,539 8,448 Number of Firms 1139 1139 933 1100 Notes: Fixed-effects panel regressions using negative binomial estimation. Wild bootstrap standard errors clustered by state in (parenthesis). Nonindicator independent variables logged. Column three omits firms with zero fundraising. Column four omits firms with a matched c(3) sister organization. All columns include controls for calendar year and state by year trends, firm assets, liabilities, program expenses, investment gain/loss, officer compensation, dividend income, and interest income, as well as state median income, gdp, population, population share under 26, and population share over 64. Predicted Marginal Effects rows report estimated combined impact of tax price and price squared at the identified points in the distribution of tax price; SD = standard deviation. **Statistically significant at 5% level. ***Statistically significant at 1% level. Open in new tab Table 2. Effects of Tax Price on Donations to Social Welfare Organizations Variables . (1) . (2) . (3) . (4) . State tax price -10.84*** -6.513** -8.492** -10.50*** (3.368) (2.865) (3.525) (2.973) State tax price square -12.82*** -8.455** -10.04** -12.07*** (3.787) (3.302) (4.044) (3.424) Log net capital gains -0.459 -0.0162 -1.024* -1.216*** (0.887) (0.402) (0.523) (0.423) Log S&P 500 -0.0326 -0.231 -0.175 -0.0196 (0.365) (0.332) (0.458) (0.361) Lag of Log S&P 500 0.236 0.0327 -0.484 0.0506 (0.356) (0.318) (0.437) (0.340) Log fundraising 0.353*** (0.0170) Log fundraising share -0.177*** (0.0224) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of Tax Price 3.34*** 2.84*** 2.61** 2.84*** (1.16) (0.99) (1.20) (1.03) |$\quad$||$-1$|SD of Tax Price 2.31** 2.16** 1.81* 1.88** (0.95) (0.78) (0.95) (0.81) |$\quad$| Median Tax Price 0.47 0.94 0.365 0.144 (0.73) (0.54) (0.67) (0.56) |$\quad$||$+$|1 SD of Tax Price -1.66* -0.46 -1.30 -1.86*** (0.93) (0.70) (0.87) (0.73) |$\quad$||$+$|2 SD of Tax Price -3.17*** -1.46 -2.49** -3.28*** (1.25) (1.00) (1.24) (1.03) Observations 8,723 8,676 6,539 8,448 Number of Firms 1139 1139 933 1100 Variables . (1) . (2) . (3) . (4) . State tax price -10.84*** -6.513** -8.492** -10.50*** (3.368) (2.865) (3.525) (2.973) State tax price square -12.82*** -8.455** -10.04** -12.07*** (3.787) (3.302) (4.044) (3.424) Log net capital gains -0.459 -0.0162 -1.024* -1.216*** (0.887) (0.402) (0.523) (0.423) Log S&P 500 -0.0326 -0.231 -0.175 -0.0196 (0.365) (0.332) (0.458) (0.361) Lag of Log S&P 500 0.236 0.0327 -0.484 0.0506 (0.356) (0.318) (0.437) (0.340) Log fundraising 0.353*** (0.0170) Log fundraising share -0.177*** (0.0224) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of Tax Price 3.34*** 2.84*** 2.61** 2.84*** (1.16) (0.99) (1.20) (1.03) |$\quad$||$-1$|SD of Tax Price 2.31** 2.16** 1.81* 1.88** (0.95) (0.78) (0.95) (0.81) |$\quad$| Median Tax Price 0.47 0.94 0.365 0.144 (0.73) (0.54) (0.67) (0.56) |$\quad$||$+$|1 SD of Tax Price -1.66* -0.46 -1.30 -1.86*** (0.93) (0.70) (0.87) (0.73) |$\quad$||$+$|2 SD of Tax Price -3.17*** -1.46 -2.49** -3.28*** (1.25) (1.00) (1.24) (1.03) Observations 8,723 8,676 6,539 8,448 Number of Firms 1139 1139 933 1100 Notes: Fixed-effects panel regressions using negative binomial estimation. Wild bootstrap standard errors clustered by state in (parenthesis). Nonindicator independent variables logged. Column three omits firms with zero fundraising. Column four omits firms with a matched c(3) sister organization. All columns include controls for calendar year and state by year trends, firm assets, liabilities, program expenses, investment gain/loss, officer compensation, dividend income, and interest income, as well as state median income, gdp, population, population share under 26, and population share over 64. Predicted Marginal Effects rows report estimated combined impact of tax price and price squared at the identified points in the distribution of tax price; SD = standard deviation. **Statistically significant at 5% level. ***Statistically significant at 1% level. Open in new tab In Table 2, the bottom panel labeled “Predicted Net Marginal Effect of Tax Price” reports post-estimation results for the combined impact of tax price, accounting for both its linear and quadratic terms, as calculated using the margins command in Stata 15. To give a sense of the change in coefficient across the distribution, for each column I report the combined effect at the median tax price, as well as at one and two standard deviations above and below the median. These span the range of an after-tax cost of 58 cents for a dollar of giving at the low end to 74 cents at the top end. In line with my predictions, the impact of tax price on giving varies as the tax price changes. For the baseline case, reported in Column 1 of Table 2, the net correlation of tax price with donations is negative and highly significant for prices above the median, where the government’s subsidy per dollar of giving is smallest. At the median, the impact is negative but significant only at the 5% level. Two deviations below median, where government per-dollar subsidies are the largest, tax price is positively and highly significantly correlated with giving—that is, in this range, donations to c(4)’s are declining as the subsidy for giving to c(3)’s increases. Figure 3 offers a more detailed summary of this variation. In this figure, I plot predicted net coefficients by tax price decile. It is evident that the marginal effects of tax price are negative and significant for observations on the right-hand side of the figure, where the after-tax cost of giving is highest. Figure 3. Open in new tabDownload slide Net Marginal Effect of Tax Price on Donations to 501(c)(4) Firms, by Decile of Tax Price Notes: Summarizes results of regression in which the outcome variable is log donations. Predicted marginal effect coefficient is combined linear and quadratic effect of log tax price of charitable contributions. Fixed-effects panel regression using negative binomial estimation. Standard errors clustered by state. Includes controls for capital gains, same-year and lagged S&P 500, firm assets, liabilities, expenses, officer compensation, dividend, and other investment income, as well as calendar year and state median income, state by year trends, gdp, population, population share under 26, and population share over 64. Figure 3. Open in new tabDownload slide Net Marginal Effect of Tax Price on Donations to 501(c)(4) Firms, by Decile of Tax Price Notes: Summarizes results of regression in which the outcome variable is log donations. Predicted marginal effect coefficient is combined linear and quadratic effect of log tax price of charitable contributions. Fixed-effects panel regression using negative binomial estimation. Standard errors clustered by state. Includes controls for capital gains, same-year and lagged S&P 500, firm assets, liabilities, expenses, officer compensation, dividend, and other investment income, as well as calendar year and state median income, state by year trends, gdp, population, population share under 26, and population share over 64. In short, I find evidence that for roughly one-quarter to one-third of all 501(c)(4) firm-year observations, tax subsidies for contributions to charity are correlated with increased contributions to c(4)s, even though these donations are not legally eligible for the subsidy. These effects are relatively large in magnitude. For example, in the deciles where tax price begins to show statistical significance on the negative side, the elasticity of donations received to changes in tax price is about -1, comparable to the price-elasticity of giving to charity reported in other firm-side studies (Yetman and Yetman, 2012; Galle, 2016). Among firms in jurisdictions where the relative subsidy is largest, however, tax price correlates with smaller donations, consistent with a prediction that gifts to charities and social welfare organizations are substitutes.21 Figure 4 further explores this result graphically. To isolate tax price from capital gains and the stock market, I first regress log donations on log capital gains and the S&P 500, and capture the residuals. I then plot these “predicted” gifts against tax price in treatment and control states (i.e. states with no income tax). Although the results are not easy to interpret with the naked eye, it does seem that the relationship between price and giving is negative for treatment states (where the after-tax price of a donation is lower) and positive in control states (where the price is higher). Figure 4. Open in new tabDownload slide Plot of Predicted Gifts Against After-Tax Price of Giving Notes: Predicted gifts are residuals obtained after regressing actual log gifts against log capital gains and log S&P 500. Treated states are states with an income tax, control states without. Figure 4. Open in new tabDownload slide Plot of Predicted Gifts Against After-Tax Price of Giving Notes: Predicted gifts are residuals obtained after regressing actual log gifts against log capital gains and log S&P 500. Treated states are states with an income tax, control states without. Columns 2 through 4 of Table 2 shed some additional light on these surprising results. When I control for fundraising as well as the share of firm expenditures devoted to fundraising, coefficients are still positive and significant for the lower half of the tax price distribution. Coefficients for the upper half of the distribution are still negative but no longer significant. As expected, fundraising increases donations, while the share of firm expenses devoted to fundraising depresses them.22 One potential interpretation of these results is that fundraising by c(4)’s does not affect crowding out but does in part explain the positive relationship between high after-tax costs of charitable giving and donations to c(4)’s. However, in several alternative specifications, such as with instruments for fundraising or regressions omitting state-by-year trends, coefficients on tax price remain negative and significant to the right of median. Another angle on the relative importance of fundraising is to examine the effects of tax price only among firm-years that report zero fundraising expenditures.23 These results are given in Column 3. Omitting firm-years with nonzero fundraising has only modest impact on the estimated coefficients, suggesting that fundraising is not the main driver of my results. Yet another possible explanation for the influence of tax price on social welfare giving is that the results are driven by charities with c(4) affiliates. Changes in tax price would naturally impact gifts to the c(3) entity, but its managers might steer some donations to the c(4).24 I have anecdotal evidence of this phenomenon from anonymous sources who have worked closely with c(3)–c(4) affiliated groups. Alternately, since related firms often have very similar names, some donors might simply confuse the two. Lastly, reputational spillovers from one firm might influence donations to the other (Minkoff, 2002). I find no direct support for these possibilities. As described in Section 3, I hand match c(4) firms in my sample with c(3) entities in the IRS Business Master File. In Column 4, I report results only for firms with no matched c(3) related firm. If anything, the coefficients for the combined tax price variables are a bit larger in absolute magnitude than those for Column 1 at both ends of the distribution. To be sure, it is likely that my matching technique has some number of false negatives, so that the Column 4 results still include some members of affiliated groups. But the fact that the effect of tax price increases when I omit 314 firm-year observations from related firms tends to suggest that these related groups are not driving the result.25 Overall, the data lend support to my hypothesis of a complex relationship between subsidies for charity and giving to noncharitable firms. As I described earlier, however, these results do not allow us to cleanly identify the exact drivers of that complexity. The shift from positive to negative coefficients as the tax price of charitable giving rises has several plausible explanations. If the results were driven entirely by salience and taxpayer confusion, they would align with the experimental findings by Taubinsky and Rees-Jones (2017). As in their study, I find evidence consistent with the theory that naïve donors are a larger share of the donor population when per-dollar subsidies are relatively smaller. When donors have less at stake, they may be more apt to make mistakes. It might be argued that this theory is more appealing in a cross-sectional setting than in one, such as mine, where identification relies on variation within states. Sophisticated donors arguably should not become naïve when subsidies shrink. As Chetty et al. (2009) find, however, individuals may be aware of tax rules without necessarily being willing to invest the cognitive effort of accounting for them. But I cannot clearly disentangle mistaken responses from rational behavior. My results are also potentially explainable by the contending influences of substitution effects and intentional misclaiming. It could be that substitution away from c(4) giving prevails when the tax price of charitable giving is very low, while misclaiming is predominant when tax price is high. In addition, there is some limited evidence to suggest that fundraising might be relevant over some portions of the distribution of tax price, although there is also evidence to the contrary. I investigate the relationship between tax price and fundraising more completely in Supplementary AppendixSection A.2. 5.2. Robustness Analysis 5.2.1. Alternative specifications. Donations to certain veterans organizations may be deductible even if the organization is not recognized as a 501(c)(3) charity (Aprill, 2018). To rule out the possibility that these organizations drive the results, I repeat the regressions (not tabulated) omitting any firm self-reporting itself as a veterans’ organization, or one in which the words “veteran” or “veterans” appear in the entity name. The impact of state tax price and capital gains are slightly larger and more precisely estimated when these 221 firm-years are omitted. As described earlier, attenuation may also be a concern given that I cannot observe the tax rules governing donors. Results (untabulated) are essentially identical when I omit the 674 firm-years coded as regional or national in focus.26 Although in my view estimation using negative binomial regression is the most appropriate response to the presence of firms with zero donations,27 some readers may also be interested in OLS results. I therefore repeat the analysis employing OLS, but using a two-stage Heckman selection model. The first stage is a probit regression where the dependent variable is whether an organization ever reports receiving donations, and the second stage is an OLS estimate, conditioning on the inverse Mills ratio obtained in the first stage. This results in coefficients (untabulated) that are similar to those reported, but with rather larger confidence intervals. 5.2.2. Replication using the core-other files. In order to obtain greater statistical power, I attempt to repeat my regression analysis using a larger set of tax returns. This alternative database, the Core-Other files, contains a much more limited set of variables from each tax return than the SOI files, but includes returns from every filing firm. Most problematically, the Core file does not separately report individual contributions, but instead has a single field, “total support,” that aggregates individual donations, government grants, and contributions from other exempt organizations. Another difficulty for the Core files is that they have been given less editorial attention by IRS staff. A large fraction of firm-year observations have missing or obviously incorrect fiscal years, and many lack address information. I am able to recover or correct some fiscal years by linking reported beginning-of-year and end-of-year asset and liability balances. I also impute missing middle-value zip codes. However, approximately 70,000 observations are omitted because one or both of these errors were uncorrectable. Of the remaining 426,000 or so firm-year observations, only 275,675 ever report nonzero values for total support. In general, results for this subset of 275,675 are broadly similar to the results reported in Sections 5.1 and the Supplementary Appendices. As might be expected given that total support includes not only donations but also two other sources of revenue that might potentially move in the opposite direction from donations, the coefficients in donation regressions are smaller. Table 2 Columns 1 and 2 report full-sample regressions in which “total support” is the outcome variable. The tax price of giving to charity still has a negative and significant sign for a substantial portion of the distribution, with elasticities near or exceeding |$-$|1.0. The effect when controlling for fundraising (Column 2) is essentially the same. I interpret this as additional evidence that fundraising does not explain all of donor responses to tax price. Likewise, coefficients are basically unchanged in the sub-samples with zero fundraising (Table 3 Column 3) or with no linked c(3) organization (Table 3 Column 4). Table 3. Effects of Tax Price on Total Support Among Core-Other 501(c)(4) Organizations Variables . (1) . (2) . (3) . (4) . Log tax price -6.104*** -5.931*** -5.072*** -6.333*** (0.468) (0.465) (0.515) (0.471) Log tax price square -7.066*** -6.880*** -5.914*** -7.365*** (0.538) (0.534) (0.591) (0.541) Log net capital gains 0.325*** 0.342*** 0.217*** 0.303*** (0.0650) (0.0646) (0.0719) (0.0653) Log fundraising exp 0.126*** (0.00148) Log S&P 500 -0.256*** -0.235*** -0.147** -0.258*** (0.0582) (0.0578) (0.0635) (0.0585) Lag of log S&P 0.0315 0.0278 -0.0454 0.0310 (0.0564) (0.0557) (0.0623) (0.0567) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of tax price 1.711*** 1.678*** 1.469*** 1.812*** 0.161 0.160 0.177 0.162 |$\quad$||$-1$|SD of tax price 1.146*** 1.1279*** 0.996*** 1.223*** 0.128 0.127 0.140 0.1283 |$\quad$| Median tax price 0.128 0.137 0.144 0.162* 0.0897 0.0894 0.0984 0.090 |$\quad$||$+$|1 SD of tax price -1.0448*** -1.005*** -0.838*** -1.060*** 0.116 0.115 0.128 0.116 |$\quad$||$+$|2 SD of tax price -1.879*** -1.817*** -1.535*** -1.929*** 0.164 0.163 0.181 0.165 Observations 275,675 275,675 241,965 273,413 Number of firms 30,519 30,519 28,357 30,289 Variables . (1) . (2) . (3) . (4) . Log tax price -6.104*** -5.931*** -5.072*** -6.333*** (0.468) (0.465) (0.515) (0.471) Log tax price square -7.066*** -6.880*** -5.914*** -7.365*** (0.538) (0.534) (0.591) (0.541) Log net capital gains 0.325*** 0.342*** 0.217*** 0.303*** (0.0650) (0.0646) (0.0719) (0.0653) Log fundraising exp 0.126*** (0.00148) Log S&P 500 -0.256*** -0.235*** -0.147** -0.258*** (0.0582) (0.0578) (0.0635) (0.0585) Lag of log S&P 0.0315 0.0278 -0.0454 0.0310 (0.0564) (0.0557) (0.0623) (0.0567) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of tax price 1.711*** 1.678*** 1.469*** 1.812*** 0.161 0.160 0.177 0.162 |$\quad$||$-1$|SD of tax price 1.146*** 1.1279*** 0.996*** 1.223*** 0.128 0.127 0.140 0.1283 |$\quad$| Median tax price 0.128 0.137 0.144 0.162* 0.0897 0.0894 0.0984 0.090 |$\quad$||$+$|1 SD of tax price -1.0448*** -1.005*** -0.838*** -1.060*** 0.116 0.115 0.128 0.116 |$\quad$||$+$|2 SD of tax price -1.879*** -1.817*** -1.535*** -1.929*** 0.164 0.163 0.181 0.165 Observations 275,675 275,675 241,965 273,413 Number of firms 30,519 30,519 28,357 30,289 Notes: Fixed-effects panel regressions using negative binomial estimation. Wild bootstrap standard errors clustered by state in (parenthesis). Nonindicator independent variables logged. Column three omits firms with zero fundraising. Column four omits firms with a matched c(3) sister organization. All columns include controls for calendar year and state-by-year trends, as well as gross assets, liabilities, program service revenue, investment income, investment losses, officer compensation, state median income, gdp, population, population share under 26, and population share over 64. Predicted Marginal Effects rows report estimated combined impact of tax price and price squared at the identified points in the distribution of tax price; SD= standard deviation. *: statistically significant at 10% level. **: statistically significant at 5% level. ***: statistically significant at 1% level. Open in new tab Table 3. Effects of Tax Price on Total Support Among Core-Other 501(c)(4) Organizations Variables . (1) . (2) . (3) . (4) . Log tax price -6.104*** -5.931*** -5.072*** -6.333*** (0.468) (0.465) (0.515) (0.471) Log tax price square -7.066*** -6.880*** -5.914*** -7.365*** (0.538) (0.534) (0.591) (0.541) Log net capital gains 0.325*** 0.342*** 0.217*** 0.303*** (0.0650) (0.0646) (0.0719) (0.0653) Log fundraising exp 0.126*** (0.00148) Log S&P 500 -0.256*** -0.235*** -0.147** -0.258*** (0.0582) (0.0578) (0.0635) (0.0585) Lag of log S&P 0.0315 0.0278 -0.0454 0.0310 (0.0564) (0.0557) (0.0623) (0.0567) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of tax price 1.711*** 1.678*** 1.469*** 1.812*** 0.161 0.160 0.177 0.162 |$\quad$||$-1$|SD of tax price 1.146*** 1.1279*** 0.996*** 1.223*** 0.128 0.127 0.140 0.1283 |$\quad$| Median tax price 0.128 0.137 0.144 0.162* 0.0897 0.0894 0.0984 0.090 |$\quad$||$+$|1 SD of tax price -1.0448*** -1.005*** -0.838*** -1.060*** 0.116 0.115 0.128 0.116 |$\quad$||$+$|2 SD of tax price -1.879*** -1.817*** -1.535*** -1.929*** 0.164 0.163 0.181 0.165 Observations 275,675 275,675 241,965 273,413 Number of firms 30,519 30,519 28,357 30,289 Variables . (1) . (2) . (3) . (4) . Log tax price -6.104*** -5.931*** -5.072*** -6.333*** (0.468) (0.465) (0.515) (0.471) Log tax price square -7.066*** -6.880*** -5.914*** -7.365*** (0.538) (0.534) (0.591) (0.541) Log net capital gains 0.325*** 0.342*** 0.217*** 0.303*** (0.0650) (0.0646) (0.0719) (0.0653) Log fundraising exp 0.126*** (0.00148) Log S&P 500 -0.256*** -0.235*** -0.147** -0.258*** (0.0582) (0.0578) (0.0635) (0.0585) Lag of log S&P 0.0315 0.0278 -0.0454 0.0310 (0.0564) (0.0557) (0.0623) (0.0567) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of tax price 1.711*** 1.678*** 1.469*** 1.812*** 0.161 0.160 0.177 0.162 |$\quad$||$-1$|SD of tax price 1.146*** 1.1279*** 0.996*** 1.223*** 0.128 0.127 0.140 0.1283 |$\quad$| Median tax price 0.128 0.137 0.144 0.162* 0.0897 0.0894 0.0984 0.090 |$\quad$||$+$|1 SD of tax price -1.0448*** -1.005*** -0.838*** -1.060*** 0.116 0.115 0.128 0.116 |$\quad$||$+$|2 SD of tax price -1.879*** -1.817*** -1.535*** -1.929*** 0.164 0.163 0.181 0.165 Observations 275,675 275,675 241,965 273,413 Number of firms 30,519 30,519 28,357 30,289 Notes: Fixed-effects panel regressions using negative binomial estimation. Wild bootstrap standard errors clustered by state in (parenthesis). Nonindicator independent variables logged. Column three omits firms with zero fundraising. Column four omits firms with a matched c(3) sister organization. All columns include controls for calendar year and state-by-year trends, as well as gross assets, liabilities, program service revenue, investment income, investment losses, officer compensation, state median income, gdp, population, population share under 26, and population share over 64. Predicted Marginal Effects rows report estimated combined impact of tax price and price squared at the identified points in the distribution of tax price; SD= standard deviation. *: statistically significant at 10% level. **: statistically significant at 5% level. ***: statistically significant at 1% level. Open in new tab 6. Discussion Over some portion of the range of tax prices of giving, donors to c(4) organizations appear to respond to the charitable contribution deduction as though it applied to gifts to c(4)’s, at least with respect to their giving behavior, if not on their tax returns. The theoretical and policy implications of that fact depend in some measure on what underlies it. Most dramatically, if donors in fact are responding to the charitable contribution deduction by giving to c(4) organizations, that would upend many important assumptions about our current political system. Charitable organizations are prohibited from electioneering, and are limited to “insubstantial” lobbying, because political activities are supposed to “be conducted without public subvention” (Slee v. Commissioner, 42 F.2d 184, 185 (2d Cir. 1930)). The charitable contribution deduction provides an almost unlimited matching grant from taxpayers to donors. Uncapped subsidies are not usually optimal, even for public goods, when those goods are congestible. Lobbying expenditures cause congestion externalities as interest groups fight for legislative attention (see Galle, 2013b for a review of the supporting empirical literature). Political subsidies structured like the charitable contribution deduction would also have dramatic implications for the role of wealth in our political system (Hasen, 2008; Galle, 2013b). Deductions would tend to systematically favor wealthier interests in legislative and electoral contests, and to do so through a mechanism that leaves the source of expenditures completely opaque to the voting public (for evidence that funding sources matter to voters, see Grose and Wood, 2020). My results would not signal a complete abolition of the barrier between politics and public support but would represent incremental steps in that direction. It is unlikely that large gifts made with advice of counsel would erroneously be made to firms ineligible for deductible contributions, and similarly unlikely that single multi-million dollar gifts could be wrongfully deducted beneath the IRS’s notice. In the aggregate, though, the charitable contribution deduction might well be encouraging meaningful amounts of untraceable donations to c(4) political organizations. At the same time, crowdfunding has become a major feature of the modern political-contribution landscape (Karpf, 2012; Johnson, 2016). It is possible that on net subsidized c(4) contributions tend to aid grassroots or lower-income donors over those with greater resources. To the extent that is a desirable outcome, however, it likely would be preferable to do so through a mechanism explicitly designed to that end, such as through small-dollar matching grants (Hasen, 1996). Politics aside, it is possible that the social value of contributions to social welfare firms might be lower than for gifts to charity. Local public goods and lobbying activity may produce smaller positive externalities per average dollar than, say, international relief organizations or research institutions (e.g. because free riding incentives are smaller, see Leshem and Tabbach, 2017). If so, tax subsidies intended for charity may be less cost-effective to the extent they are claimed by c(4) organizations. Of course if this is true it may well also argue against the existing subsidy for contributions of built-in gain property to c(4) organizations. That provision appears not to have received serious scrutiny from policy makers and does not even appear in the Joint Committee on Taxation’s list of tax expenditures (for more extended criticism of the built-in gain provision, see Halperin, 2002; Colinvaux, 2013). What if donors are mistakenly giving to c(4) organizations on the belief that they will receive a subsidy, but then in fact not claiming the contributions on their tax returns? Faulhaber (2012) and Galle (2013a) argue that taxpayers who mistakenly overestimate the value of charitable contribution deductions in essence provide free public goods to the community at large. On the other side of the balance, these donors are likely not maximizing their own preferred consumption choices, and the net results for consumer welfare likely depend on how donors adjust to new information about their past mistakes (Chetty et al., 2009). The implications are more modest if most of the observed effects are caused by fundraising. I cannot clearly rule out a role for fundraising. In the smaller SOI-Other sample, controlling for fundraising leaves coefficients statistically insignificant in some portions of the tax price distribution, while in the larger Core-Other sample it does not. In Supplementary Appendix Section A.2, I report that national firms, many of them advocacy organizations, are the most apt to adjust their fundraising in response to tax policy. With donations on the left-hand side, SOI-Other specifications without these firms retain significant coefficients even when controlling for fundraising. Thus, it seems possible that fundraising explains some of the donation response to taxes at these national firms but not among the community-oriented organizations that comprise the bulk of the sample. 7. Conclusion I have presented evidence suggesting new and previously unknown linkages between the U.S. system of supports for charitable organizations and the behavior of social welfare firms. It appears that subsidies for charity simultaneously pressure and lift up c(4) organizations via a variety of channels. Policy makers may wish to consider these results in a number of contexts. Donor confusion raises significant issues. Donors who mistakenly contribute to social welfare organizations, without claiming the deduction, are providing the public with free benefits, albeit perhaps in a way that might raise consumer-protection concerns. On the other hand, subsidies of the level I find are potentially troubling for organizations that engage in partisan politics to the degree that social welfare firms now do. Regulators may also wish to consider reforms if donors are claiming the deduction, and if social welfare firm outputs are on average of lower net social benefit than charity. Possibilities include better informing donors or more closely scrutinizing their tax returns. Alternately, Congress could consider paring back the official subsidies c(4) firms now claim, such as the exclusion for donations of built-in gain property. Lastly, I find evidence that the net payoff of dollars diverted away from charity may be higher than prior literature suggests. Slemrod (1989) models the treasury efficiency of the charitable contribution deduction under the assumption that misclaimed donations are not socially productive. Chetty (2009) notes that avoidance behaviors can still have social value. Here, I find potential evidence of one such value-creating avoidance activity: while contributions to c(4) organizations may not be what Congress intended for the charitable contribution deduction, many likely do create positive externalities of some magnitude. The net treasury efficiency of tax subsidies for charity thus depends in some part on their incidental effects on social welfare firms. Supplementary material Supplementary material is available at American Law and Economics Review Journal online. Footnotes 1. One possible psychological explanation for this phenomenon is suggested by recent experimental results in which subjects cheat less when the stakes from cheating are larger (Rahwan et al., 2018). 2. Supplementary Appendix Section A.2 additionally reports regressions examining the extent to which the charitable contribution deduction affects fundraising and returns to fundraising in c(4) firms. 3. On the other hand, transfers also eliminate built-in losses. Well-advised donors do not contribute depreciated property, at least if that property would have generated a deduction when sold by the donor. 4. Gifts to charity of appreciated stock, and in some cases of other assets, can be deducted at their fair market value (see Ackerman and Auten, 2011 for more detailed discussion of the applicable rules). The FMV deduction in combination with the exclusion of tax on built-in gain can produce a negative effective tax rate. 5. The Schedule A instructions do not change in relevant respects between 1991 and 2006. 6. Slemrod (1989) reports that IRS audits in the early 1980’s detected mis-claimed charitable contributions of between 5% and 10% of total claimed contributions, with smaller fractions mostly at the top of the income distribution. This result could still be consistent with large shares of donations to c(4) organizations being mis-claimed. In the Core file that covers all c(4) returns, total annual reported “support” from all c(4) filers ranges from |${\$}$|2.9 to |${\$}$|5.1 billion (in 2007 dollars) during my sample period, 1991 to 2007. U.S. charitable contribution deductions in 2007 totaled about |${\$}$|193 billion (IRS, 2017). If Slemrod’s sample was representative, his figures would imply between |${\$}$|9.5 and |${\$}$|19 billion in inaccurately claimed donations. 7. That is, for any single donor the marginal effect on the output of a charitable organization will not be affected by the extent of their contribution to an unrelated social welfare firm. 8. In theory, a decline in the after-tax price of charitable contributions could also increase donations to competing c(4) entities through an income effect. That is, the subsidy may expand the household’s budget enough that it consumes more of all goods, including c(4) donations. But this is very implausible in our setting, where household giving averages less than 2% of household income (Giving USA 2018) and the effective subsidies to the marginal price of charity are only a fraction of that. 9. Major revisions to the ways in which donations are reported on IRS Form 990, effective in 2008, make comparisons of donation levels between the two editions somewhat problematic. 10. I did not encounter any instances in which a web site referred to another organization in order to explain that the two were unrelated and should not be confused. 11. An “exogenous itemizer” is a taxpayer who would have itemized their deductions under the standard income tax (not necessarily the AMT) even with zero charitable contribution deductions. Using this population avoids the endogeneity problem that arises if increased donations change the tax price, such as would result if additional donations would lead the donor to itemize. See Bakija & Heim (2011) for more discussion. 12. Since I control for calendar year, identification for the S&P variables depends on their impact in firms that follow a non-calendar fiscal year. 13. The tax price computation also includes adjustments for the effects of claiming a charitable contribution deduction on other determinants of taxable income. For example, because state and local taxes are federally deductible, a state contribution deduction increases federal taxes. See Bakija and Heim (2011) and Galle (2016) for more detail on these interactions. 14. Most of the tax-price variation in the sample occurs as a result of federal-law changes in 2003 and 2006. The change in 2006 was caused by the phaseout of the Pease limitation, a provision that tended to sharply reduce the value of charitable giving for some high-bracket taxpayers in states without an income tax. The 2003 changes were largely a product of complex interactions of state taxes with the federal rate structure, and in particular the AMT. See Supplementary Appendix Section A.1 for more discussion. 15. Although some state capital gains rates match the state rate on other sources of income, federal rules often interact with ordinary rates to a differing extent than capital gains rates. The coefficient of correlation between the state tax price and net capital gains rate variables is 0.80. 16. Results are robust to simply using the unadjusted NBER capital gains rate. 17. To offer a sense of the possible degree of attenuation, I implement a Monte Carlo exercise in which I use actual observed tax prices but adjust donations so that the price-elasticity of giving is |$-$|1. I then randomly replace the after-tax price for 25% of the observations with the same-year price of another state. After replacement, the estimated coefficient falls from |$-$|1 to |$-$|0.787. If I repeat the exercise replacing only 10% of the price observations, the estimated coefficient is |$-$|0.911. 18. RESET analysis also suggested that quadratic forms better fit the data. 19. |$E_{FD}$| can also be thought of as appearing in reduced form. Galle (2016) does not consider the potential separate impact of the share of firm expenditures devoted to fundraising. Donors may dislike firms with high fundraising shares. If so, |$E_{FD} $| measures the net effect of an additional dollar of fundraising on donations through two channels, one directly, and the other through fundraising share. 20. Results are similar when omitting any trends and when including both linear and quadratic state-by-year trends. In the latter case, some coefficients of tax price at below-median price lose significance. 21. An alternative explanation might be that tax price is correlated with the share of itemizers in a jurisdiction, and that itemizer share might in some way drive the outcome. This is implausible, as the coefficient of correlation between the two is only -0.147. Nonetheless, I run additional regressions in which I control for log share of itemizers. For the years (1997–2007) for which I have these data available, coefficients for the two sets of regressions are essentially identical. 22. Fundraising is endogenous to donations. Unfortunately, there are few convincing instruments for fundraising that would satisfy the exclusion restriction. To nonetheless give a sense of what IV results might look like, I instrument for fundraising using the control function approach of Woolridge (2013), and employ occupancy (as in Andreoni and Payne, 2011; Heutel, 2014), a measure of expected fundraising (as in Galle, 2016), and printing expenditures as instruments. In these regressions, the coefficient on fundraising falls in about the same range as that reported in Table 2, around 0.29, or 0.47 if I omit expected fundraising as an instrument. The coefficient on the state tax price in these specifications remains significant for most portions of the distribution of tax price, as in regressions without fundraising controls, but is larger in magnitude. 23. Since firms are likely to under-report fundraising effort or classify it in another category, firms with zero reported fundraising might still be expending resources on solicitations. Some firms might also rely on volunteer fundraising campaigns. 24. Kerlin and Reid (2010) report that in some advocacy organizations, the c(3) arm transfers large portions of its revenues to the c(4). Why, then, would firms also steer donations to the c(4), if gifts to the c(3) receive a much larger tax benefit? Recall that a c(3) cannot make grants earmarked for funding political activities if it would itself be prohibited from engaging in those same political activities. Of course, money is fungible. Still, donors who want to fund electioneering or large lobbying campaigns may prefer to give directly to the c(4) in order to ensure that the funds are spent on preferred projects. 25. I may also obtain more precise results when omitting linked firms because these political organizations are more apt to receive cross-state donations. Dropping them therefore reduces attenuation. 26. When I control for fundraising in sub-samples omitting either c(3)-linked or national firms, I obtain statistically significant and negative coefficients at the top of the distribution of tax price, unlike in regressions using the full sample. This could be due to the possibility that strategic fundraising responses to the charitable contribution deduction are mostly a feature of national advocacy organizations, a possibility I explore more fully in Supplementary Appendix Section A.2. 27. Unsurprisingly, results are very similar when alternately using cluster-robust poisson estimates. References Ackerman, Deena , and Auten Gerald. 2011 . “ Tax Expenditures for Noncash Charitable Contributions ,” 64 National Tax Journal 651 – 88 . Google Scholar Crossref Search ADS WorldCat Almunia, Miguel , Guceri Irem, Lockwood Benjamin, and Scharf Kimberley. 2020 . “ More Giving or More Givers? The Effects of Tax Incentives on Charitable Donations in the UK ,” 183 Journal of Public Economics 104 – 114 . Google Scholar Crossref Search ADS WorldCat Andreoni, James , and Payne A. Abigail. 2013 . “ Is Crowding Out Due Entirely to Fundraising? Evidence from a Panel of Charities ,” 95 Journal of Public Economics 334 – 43 . Google Scholar Crossref Search ADS WorldCat Andreoni, James , and Payne A. Abigail. 2003 . “ Do Government Grants to Private Charities Crowd Out Giving or Fundraising? ,” 93 American Economic Review 792 – 812 . Google Scholar Crossref Search ADS WorldCat Aprill, Ellen P . 2012 . “ Once and Future Gift Taxation of Transfers to Section 501(c)(4) Organizations: Current Law, Constitutional Issues, and Policy Considerations ,” 15 Journal of Legislation and Public Policy 289 – 327 . Google Scholar OpenURL Placeholder Text WorldCat Aprill, Ellen P . 2013 . “ Regulating the Political Speech of Noncharitable Exempt Organizations After Citizens United ,” 10 Election Law Journal 363 . Google Scholar Crossref Search ADS WorldCat Aprill, Ellen P . 2018 . “ Examining the Landscape of §501(c)(4) Social Welfare Organizations ,” 21 N.Y.U. Journal of Legislation and Public Policy 345 – 411 . Google Scholar OpenURL Placeholder Text WorldCat Ayres, Ian . 2017 . “ Voluntary Taxation and Beyond: The Promise of Social-Contract Voting Mechanisms ,” 19 American Law & Economics Review 1 – 48 . Google Scholar Crossref Search ADS WorldCat Cameron, A Colin. , Gelbach Jonah B., and Miller, Douglas L. 2008 . “ Bootstrap-Based Improvements for Inference with Clustered Errors ,” 90 Review of Economics and Statistics 414 – 427 . Google Scholar Crossref Search ADS WorldCat Chand, Daniel E . 2017 . “ Lobbying and Nonprofits: Money and Membership Matter—But Not for All ,” 98 Social Science Quarterly , 1297 – 1312 . Google Scholar Crossref Search ADS WorldCat Chetty, Rah . 2009 . “ Is the taxable income elasticity sufficient to calculate deadweight loss? The implications of evasion and avoidance ,” 1 American Economic Journal: Economic Policy 31 – 52 . Google Scholar Crossref Search ADS WorldCat Chetty, Raj , Looney Adam, and Kroft Korey. 2009 . “ Salience and Taxation: Theory and Evidence ,” 99 American Economic Review 1145 – 77 . Google Scholar Crossref Search ADS WorldCat Child, Curtis D. , and Gronbjerg Kirsten A.. 2007 . “ Nonprofit Advocacy Organizations: Their Characteristics and Activities ,” 88 Social Science Quarterly 259 – 81 . Google Scholar Crossref Search ADS WorldCat Colinvaux, Roger . 2013 . “ Charitable Contributions of Property: A Broken System Reimagined ,” 50 Harvard Journal on Legislation 263 – 329 . Google Scholar OpenURL Placeholder Text WorldCat Dimmery, Drew , and Peterson Andrew. 2016 . “ Shining the Light on Dark Money: Political Spending by Nonprofits ,” 2 Russell Sage Foundation Journal of the Social Sciences 51 – 68 . Google Scholar Crossref Search ADS WorldCat Dougherty, Terence . 2013 . “ Section 501(c)(4) Advocacy Organizations: Political Candidate-Related and Other Partisan Activities in Furtherance of Social Welfare ,” 36 Seattle Law Review 1377 – 1412 . Google Scholar OpenURL Placeholder Text WorldCat Duquette, Nicolas J . 2016 . “ Do Tax Incentives Affect Charitable Contributions? Evidence from Public Charities’ Reported Revenues ,” 137 Journal of Public Economics 51 – 69 . Google Scholar Crossref Search ADS WorldCat Eckel, Catherine , and Grossman Philip J.. 2017 . “ Comparing Rebate and Matching Subsidies Controlling for Donors’ Awareness: Evidence from the Field ,” 66 Journal of Behavioral and Experimental Economics 88 – 95 . Google Scholar Crossref Search ADS WorldCat Faulhaber, Lilian V . 2012 . “ The Hidden Limits of the Charitable Deduction: An Introduction to Hypersalience ,” 92 Boston University Law Review 1307 – 1348 . Google Scholar OpenURL Placeholder Text WorldCat Feenberg, Daniel , and Coutts Elisabeth. 1993 . “ An Introduction to the TAXSIM Model ,” 12 Journal of Policy Analysis and Management 189 – 94 . Google Scholar Crossref Search ADS WorldCat Feldman, Naomi E. , Goldin Jacob, and Homonoff Tatiana. 2018 . “ Raising the Stakes: Experimental Evidence on the Endogeneity of Taxpayer Mistakes .” 71 National Tax Journal 201 – 230 . Google Scholar Crossref Search ADS WorldCat Feldman, Naomi E. , Katuscak Peter, and Kawano Laura. 2016 . “ Taxpayer Confusion: Evidence from the Child Tax Credit ,” 106 American Economic Review 807 – 35 . Google Scholar Crossref Search ADS WorldCat Feldstein, Martin . 1999 . “ Tax Avoidance and the Deadweight Loss of the Income Tax ,” 81 Review of Economics and Statistics 674 – 80 . Google Scholar Crossref Search ADS WorldCat Gallagher, Kelly , and Muehlegger Eric. 2011 . “ Giving Green to Get Green: Incentives and Consumer Adoption of Hybrid Vehicle Technology ,” 61 Journal of Environmental Economics and Management 1 – 15 . Google Scholar Crossref Search ADS WorldCat Galle, Brian . 2013a . “ Carrots, Sticks, and Salience ,” 67 Tax Law Review 53 – 109 . Google Scholar OpenURL Placeholder Text WorldCat Galle, Brian . 2013b . “ Charities in Politics: A Reappraisal ,” 54 William & Mary Law Review 1561 – 1632 . Google Scholar OpenURL Placeholder Text WorldCat Galle, Brian . 2016 . “ How Do Nonprofit Firms Respond to Tax Policy? ,” 45 Public Finance Review 364 – 94 . Google Scholar Crossref Search ADS WorldCat Galston, Miriam . 2011 . “ When Statutory Regimes Collide: Will Citizens United and Wisconsin Right to Life Make Federal Tax Regulation of Campaign Activity Unconstitutional? ,” 13 University of Pennsylvania Journal of Constitutional Law 867 . Google Scholar OpenURL Placeholder Text WorldCat Goldin, Jacob , and Listokin Yair. 2013 . “ Tax Expenditure Salience ,” 16 American Law & Economics Review 144 – 76 . Google Scholar Crossref Search ADS WorldCat Grose, Christian , and Wood Abby K.. 2020 . “ Campaign Finance Transparency Affects Legislative Candidate Performance at the Polls ,” American Journal of Political Science . Google Scholar OpenURL Placeholder Text WorldCat Hackney, Philip . 2020 . “ Political Justice and Tax Policy: The Social Welfare Organization Case ,” 8 Texas A&M Law Review . Google Scholar OpenURL Placeholder Text WorldCat Halperin, Daniel . 2002 . “ A Charitable Contribution of Appreciated Property and the Realization of Built-In Gains ,” 56 Tax Law Review 1 – 38 . Google Scholar OpenURL Placeholder Text WorldCat Halperin, Daniel . 2011 . “ Is Income Tax Exemption for Charities a Subsidy? ,” 64 Tax Law Review 283 – 312 . Google Scholar OpenURL Placeholder Text WorldCat Halperin, Daniel . 2018 . “ The Tax Exemption Under Section 501(c)(4) ,” 21 N.Y.U. Journal of Legislation and Public Policy 519 – 35 . Google Scholar OpenURL Placeholder Text WorldCat Hasen, Richard L . 1996 . “ Clipping Coupons for Democracy: An Egalitarian/Public Choice Defense of Campaign Finance Vouchers ,” 84 California Law Review 1 – 58 . Google Scholar Crossref Search ADS WorldCat Hofmann, Mary Ann . 2007 . “ Tax-Motivated Expense Shifting by Tax-Exempt Associations ,” 29 Journal of the American Taxation Association 43 – 60 . Google Scholar Crossref Search ADS WorldCat Horton Smith, David . 1997 . “ The Rest of the Nonprofit Sector: Grassroots Associations as the Dark Matter Ignored in Prevailing ‘Flat Earth’ Maps of the Sector ,” 26 Nonprofit & Voluntary Sector Quarterly 114 – 31 . Google Scholar Crossref Search ADS WorldCat Hungerman, Daniel M. , and Rinz, Kevin “ Where Does Voucher Funding Go? How Large-Scale Subsidy Programs Affect Private-School Revenue, Enrollment, and Prices ,” 136 Journal of Public Economics 62 – 85 . Crossref Search ADS WorldCat Internal Revenue Service. 2006a . “ Instructions for Schedule A, Itemized Deductions .”, Available at: https://www.irs.gov/pub/irs-prior/i1040sa–2006.pdf (accessed September 17, 2020 ). Internal Revenue Service. 2006b . “ Instructions for Form 8283: Noncash Contributions ,” Available at: https://foundation.mtech.edu/give/i8283.pdf (accessed September 17, 2020 ). Internal Revenue Service. 2017 . “ Individual Income Tax Returns: Selected Income and Tax Items for Tax Years 1999 – 2016 ,” Available at: https://www.irs.gov/statistics/soi-tax-stats-historical-table-1 (accessed September 17, 2020 ). Johnson, Dennis . 2016 . Campaigning in the Twenty-First Century: Activism, Big Data, and Dark Money Second Edition . New York, NY : Routledge . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Karpf, David . 2012 . The MoveOn Effect: The Unexpected Transformation of American Political Advocacy . New York, NY : Oxford University Press . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Kerlin, Janelle A. , and Reid Elizabeth J.. 2010 . “ The Financing and Programming of Advocacy in Complex Nonprofit Structures ,” 39 Nonprofit and Voluntary Sector Quarterly 802 – 24 . Google Scholar OpenURL Placeholder Text WorldCat Kimberlin, Sara E . 2010 . “ Advocacy by Nonprofits: Roles and Practices of Core Advocacy Organizations and Direct Service Agencies ,” 9 Journal of Policy Practice 164 – 82 . Google Scholar Crossref Search ADS WorldCat Koulish, Jeremy . 2016 . “ From Camps to Campaign Funds: The History, Anatomy, and Activities of 501(c)(4) Organizations ,” Urban Institute Research Report , Available at: https://www.urban.org/research/publication/camps-campaign-funds-history-anatomy-and-activities-501c4-organizations. (accessed September 17, 2020 ). Google Scholar Leff, Benjamin Moses . 2009 . “ Sit Down and Count the Cost: A Framework for Constitutionally Enforcing the 501(c)(3) Campaign Intervention Ban ,” 28 Virginia Tax Review 673 – 731 . Google Scholar OpenURL Placeholder Text WorldCat Leshem, Shmuel , and Tabbach Avraham. 2017 . “ Solving the Volunteer’s Dilemma: The Efficiency of Rewards Versus Punishments ,” 18 American Law & Economics Review 1 – 32 . Google Scholar Crossref Search ADS WorldCat List, John , and Peysakhovich Yana. 2010 . “ Charitable Donations Are More Responsive to Stock Market Booms than Busts ,” 110 Economic Letters 166 – 69 . Google Scholar Crossref Search ADS WorldCat Maguire, Robert . 2014 . “ Dark Money Hits |${\$}$|50 Million, Most Still to Come ,” Center for Responsive Politics, August 28 . Available at: https://www .opensecrets.org/news/2014/08/dark-money-hits-50-million-most-still-to-come. Mancino, Douglas . 2005 . “ The Impact of Federal Tax Exemption Standards on Health Care Policy and Delivery ,” 15 Health Matrix 5 . Google Scholar PubMed OpenURL Placeholder Text WorldCat Mayer, Lloyd Hitoshi . 2018 . “ A (Partial) Defense of Section 501(c)(4)’s Catchall Nature ,” 21 N.Y.U. Journal of Legislation and Public Policy 439 – 80 . Google Scholar OpenURL Placeholder Text WorldCat Minkoff, Debra C . 2002 . “ The Emergence of Hybrid Organizational Forms: Combining Identity-Based Service Provision and Political Action ,” 31 Nonprofit and Voluntary Sector Quarterly 377 – 401 . Google Scholar Crossref Search ADS WorldCat Nicholson-Crotty, Jill . 2007 . “ Politics, Policy, and the Motivations for Advocacy in Nonprofit Reproductive Health and Family Planning Providers ,” 36 Nonprofit and Voluntary Sector Quarterly 5 – 21 . Google Scholar Crossref Search ADS WorldCat Nicholson-Crotty, Jill . 2009 . “ The Stages and Strategies of Advocacy Among Nonprofit Reproductive Health Providers ,” 38 Nonprofit and Voluntary Sector Quarterly 1044 – 53 . Google Scholar Crossref Search ADS WorldCat O’Neil, Cherie , Steinberg Richard S., and Thompson, G. Rodney “ Reassessing the Tax-Favored Status of the Charitable Deduction for Gifts of Appreciated Assets ,” 49 National Tax Journal 215 – 233 . OpenURL Placeholder Text WorldCat Rahwan, Zoe , Hauser Oliver P., Kochanowska Ewa, and Fasolo Barbara, “ High Stakes: A Little More Cheating, A Lot Less Charity ,” 152 Journal of Economic Behavior & Organization 276 – 95 . Crossref Search ADS WorldCat Roberts, Russell D . 1987 . “ Financing Public Goods ,” 95 Journal of Political Economy 420 – 37 . Google Scholar Crossref Search ADS WorldCat Slemrod, Joel . 1989 . “ Are estimated tax elasticities really just tax evasion elasticities? The case of charitable contributions ,” 71 Revue of Economics and Statistics 517 – 22 . Google Scholar Crossref Search ADS WorldCat Slemrod, Joel , and Yitzhaki Shlomo. 2002 . “ Tax Avoidance, Evasion, and Administration ,” in Alan J. Auerbach, and Martin Feldstein, eds. Handbook of Public Economics , Vol. 3. Google Scholar OpenURL Placeholder Text WorldCat Steinberg, Richard . 1991 . “ Does Government Spending Crowd Out Donations? Interpreting the Evidence ,” 62 Annals of Public and Cooperative Economics 591 - 617 . Google Scholar Crossref Search ADS WorldCat Steinberg, Richard . 1997 . “On the Regulation of Fund Raising,” in Burlingame D. ed., Critical Issues in Fund Raising . San Francisco, CA : Jossey-Bass , pp. 234 - 44 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Taubinsky, Dmitry , and Rees-Jones Alex. 2017 . “ Attention Variation and Welfare: Theory and Evidence from a Tax Salience Experiment ,” 85 Review of Economics & Statistics 2462 – 2496 . Google Scholar Crossref Search ADS WorldCat Tobin, Donald B . 2007 . “ Political Campaigning by Churches and Charities: Hazardous for 501(c)(3)s, Dangerous for Democracy ,” 95 Georgetown Law Journal 1313 . Google Scholar OpenURL Placeholder Text WorldCat U.S. Department of the Treasury. 2018 . “ Treasury Department and IRS Announce Significant Reform to Protect Personal Donor Information to Certain Tax-Exempt Organizations .” Available at: https://home.treasury.gov/news/press-releases/sm426 (accessed July 18, 2018 ). Yetman, Michelle H. , and Yetman, Robert 2013. “ How Does the Incentive Effect of the Charitable Deduction Vary Across Charities? ,” 88 The Accounting Review 1069 . Crossref Search ADS WorldCat © The Author 2020. Published by Oxford University Press on behalf of the American Law and Economics Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png American Law and Economics Review Oxford University Press

The Dark Money Subsidy? Tax Policy and Donations to Section 501(c)(4) Organizations

American Law and Economics Review , Volume 22 (2) – Dec 1, 2020

Loading next page...
 
/lp/oxford-university-press/the-dark-money-subsidy-tax-policy-and-donations-to-section-501-c-4-47IUKjIYkk
Publisher
Oxford University Press
Copyright
© The Author 2020. Published by Oxford University Press on behalf of the American Law and Economics Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
1465-7252
eISSN
1465-7260
DOI
10.1093/aler/ahaa009
Publisher site
See Article on Publisher Site

Abstract

Abstract This article presents the first empirical examination of giving to §501(c)(4) organizations, which have recently become important players in U.S. politics. Unlike gifts to charity, donations to a 501(c)(4) are not legally deductible. Yet, gifts to c(4) organizations are highly elastic to the after-tax price of charitable giving. At the lower end of the observed tax price range, c(4) giving falls with tax price, consistent with the hypothesis that giving to c(3) and c(4) organizations are substitutes. Over the top quarter of the distribution of tax price, however, gifts to c(4) organizations are negatively correlated with the after-tax price of giving to charity. That is, donors appear to respond as though the deduction subsidized their gift to a c(4). Donor responses to benefits for which they are not eligible may reflect the low salience of legal limitations or deliberate overclaiming. These results imply subsidies for charity can crowd out or in donations to c(4) organizations, with potential implications for U.S. politics. I cannot observe whether donors claim tax deductions for ineligible gifts, so the net results for the Treasury are unclear. 1. Introduction Although they have grown markedly in social importance since 2010, so-called “social welfare organizations” are seldom studied, with little known about their activities and finances. They are often better known as “c(4)’s,” as §501(c)(4) of the Tax Code grants these firms exemption from the U.S. corporate income tax. In contrast to more traditional charities exempted under §501(c)(3), donations to a 501(c)(4) are not deductible by donors. Many c(4) entities are community organizations, such as youth recreational associations or volunteer fire departments, that have failed to achieve charitable status because of tax-law technicalities. Others, however, are advocacy organizations, and it is these that have lent c(4) its newfound importance. In this article, I suggest that tax policy may have played a role in building the c(4) sector—not only by offering attractive regulatory features unavailable to charities but also by delivering accidental subsidies. Because they have become important political spending vehicles (Maguire, 2014), study of how these entities get their money should be of increasing scholarly interest. Yet although there are hundreds of papers examining the determinants of contributions to charitable organizations (see Bakija, 2013 for a review), none consider what factors encourage support for their c(4) cousins. Using several alternative measures of the after-tax price of donating to charity, I find that gifts to c(4) organizations in fact respond to changes in that price, even though donations to c(4) entities are not deductible. By including both linear and quadratic terms of the after-tax price of charitable giving (herein “tax price”) in the regression analysis, I show that the tax price is positively correlated with donations at the lower end of the tax price range, uncorrelated at the middle of the distribution, and negatively correlated in roughly the top quarter of the distribution of state-years. That is, for a substantial portion of the firm-years I observe, the charitable contribution deduction on net correlates with increased donations to ineligible c(4) recipients, to a statistically significant and economically large degree, with elasticities greater than |$-1$| in absolute magnitude. Since I rely on organizational data, not individual tax returns, I cannot tell whether this increased giving also results in a tax benefit—that is, whether donors are incorrectly claiming charitable contributions for gifts to a c(4). The c(4) organization has become a popular spending choice because of its unique regulatory status. While the c(4) firm must be “devoted exclusively to charitable, educational, or recreational purposes,” IRS interpretations of that limit allow essentially unlimited lobbying in pursuit of charitable, educational, or recreational ends (Halperin, 2014). Social welfare firms can also spend some limited fraction of their resources on campaigns for elective office, with the exact limits a matter of ongoing dispute (Aprill, 2012; Dougherty, 2013). In contrast, U.S. charities— §501(c)(3) organizations—cannot engage in “substantial” lobbying efforts, and may not electioneer at all. Until 2010, spending by social welfare organizations was somewhat constrained by federal election-law limits on firms other than candidates’ committees, but the Supreme Court’s decision in Citizens United, together with several key lower court decisions around the same time, released those shackles (Aprill, 2011; Galston, 2011). Since then, spending by “dark money” organizations, so-called because their donors are secret by law (IRC §§6103, 6104), and many with tax exemption under §501(c)(4), has skyrocketed (Maguire, 2014). Despite the increasing social importance of social welfare organizations, little is known about them. Few scholars have ever written systematically about c(4) firms. The rationales for the existence of c(4) as a category, and whether qualifying organizations should enjoy the subsidies it is known to provide, are few. Only Dougherty (2013), Halperin (2018), and Hackney (2020) explore these topics in any depth, with recent white papers by Aprill (2018) and Mayer (2018) also making important advances. Empirical work on social welfare organizations is equally scant, and largely focused on political activity (Horton Smith (1997) remarks that noncharitable exempt entities are part of the “dark matter” of unstudied voluntary associations). Chand (2017) looks at the lobbying activities of the subset of related c(3) and c(4) organizations that issue legislative scorecards. Child and Gronbjerg (2007) report survey evidence that noncharitable exempt organizations are more likely to engage in policy advocacy, a result Dimmery and Peterson (2016, 61) confirm using massive scraping of firms’ web sites. Nicholson-Crotty (2007,2009) employs collaboration with a c(4) as a measure of the political engagement of women’s health centers, while Kerlin and Reid (2010) drill deep using case studies of five environmental organizations that make use of multiple tax-exempt entities to conduct their advocacy work. As Koulish (2016) and Dimmery and Peterson (2016) show, however, most social welfare organizations are not political. Koulish (2016) provides the only available overview of the resources and activities of the whole subsector. In a more focused effort, Hofmann (2007) finds accounting evidence that noncharitable exempt organizations, including some social welfare firms, shift expenses to minimize the unrelated business income tax. At first glance it is unclear why the charitable contribution deduction would affect 501(c)(4) firms. Close observers have recognized that federal rules exempting 501(c)(4) income from the corporate income tax, and permitting donors of appreciated property to escape tax on built-in gains existing at the time of contribution, may encourage gifts, especially gifts in kind (Halperin, 2018). But investment earnings spent on political activity are subject to tax under §527(f) of the Tax Code, reducing this benefit. No one has suggested that the charitable contribution deduction might affect donations to a c(4), which again are not legally deductible. If anything, one might expect that the deduction would diminish giving to 501(c)(4) organizations. If charity and c(4) firms are substitutes, a decline in the price of giving to charity should (holding household income roughly level) shift a portion of the household budget towards charities. Charitable giving is a tiny fraction of average annual household spending (Giving USA 2018), so modest changes in the price of giving should not have meaningful income effects. Yet behavioral economics and past findings in tax compliance suggest a different potential story. As I will detail more below, c(3) and c(4) firms carry out similar missions under similar names, and tax rules for distinguishing them are likely highly opaque to most taxpayers. Feldman et al. (2016), Goldin and Listokin (2013), and Gallagher and Muehlegger (2011) have found evidence of taxpayers who overvalue various forms of tax subsidy. Perhaps some donors give to c(4) organizations believing they are making a charitable contribution. Alternately, some donors may understand the difference but conclude their erroneous claims are unlikely to be detected. Slemrod (1989) finds evidence of substantial overclaiming by charitable contributors. IRS procedures are poorly suited to audit whether deductions that are claimed in fact flow to eligible entities, especially for small cash gifts. I argue that my results can be reconciled with all three of these theoretical predictions about the effect of the charitable contribution deduction. When the tax benefits of supporting charity are relatively large, we observe a classic substitution effect towards charity and away from c(4) organizations. When tax benefits are smaller, however, subsidies for charity are positively correlated with giving to c(4)’s. This is consistent with, for example, the determination in Slemrod (1989) that misreporting is relatively inelastic to tax rates, as well as his finding (1989: Table 2) that the rates of misreported charitable contributions are hump-shaped in income.1 I emphasize, however, that I cannot verify whether taxpayers in fact claim donations to a c(4) as deductible on their returns. My findings would also fit a model in which donors are more apt to be confused when the stakes of confusion are smaller. It could be argued that the observed donation patterns might result from strategic firm behavior, especially fundraising. When I control for fundraising in regressions in which donations are the outcome variable, the estimated coefficients across the distribution of tax price are smaller, but for most of the distribution still statistically significant and larger than one in absolute magnitude.2 Further, when I repeat my analysis in a larger but less precisely measured set of data, I find that controlling for fundraising makes no difference in the price-elasticity of “public support,” a category that includes donations. In short, I find possible evidence of an unintended and previously unknown subsidy for social welfare organizations. As Faulhaber (2012) and Galle (2013a) observe, donors who over-estimate the value of the charitable contribution deduction may improve social welfare by providing public goods without the need for public support. Here, however, it is much less clear that any subsidy effect is salutary. For one, these donations may actually reduce taxes paid, as donors may well be erroneously deducting their gifts to c(4) organizations. In addition, Congress opted not to underwrite gifts to firms that can lobby and electioneer extensively with an unlimited matching grant, a subsidy structure that would tend to heavily favor wealthy interests (Galle, 2013b; Hasen, 1996; Tobin, 2007). Even if donations are not actually claimed, to the extent that misunderstandings encourage gifts to lobbying organizations, they may run contrary to good policy. While the implications are not as dramatic if fundraising is the main driver of my results, I still provide an initial window into the fiscal behavior of social welfare firms. The donative environment has a major impact on a c(4) organization. Changes to rules applicable to charities also impact c(4)’s, and the impact on net social welfare can potentially be opposite in sign from the charity-focused policy. 2. Background 2.1. What is a social welfare organization? Section 501(c)(4) serves mainly as a fallback option for entities that fail one of the qualifying tests for §501(c)(3) (Halperin, 2018). The statutory definition of “social welfare” has no clear meaning, and IRS guidance seems to define it as roughly parallel to the charitable purposes common to c(3) organizations (Dougherty, 2013). As Aprill (2013, 2018) notes, many social welfare organizations in fact resemble charities, but are disqualified from obtaining charitable status by their failure to serve a wide enough “charitable class”—either an indefinite group of beneficiaries or a definite but very large group. The organization cannot, however, serve purely private interests, suggesting that it must be organized on behalf of some small or mid-sized “community” (Dougherty, 2013). Arguably, government supports for organizations of this kind can serve as a mechanism for private allocation of public goods (Ayres, 2017). Empirically, social welfare firms are often community organizations, such as an ambulance corps or youth recreational association (Koulish, 2016). Many of the largest c(4)s in terms of assets held are health cooperatives, such as state HMO Blue and Delta Dental organizations, which were denied c(3) status but left eligible to be c(4)s by a series of IRS and judicial rulings (Mancino, 2005). To give a bit more sense of the makeup of the population, I generated a frequency plot of the words most commonly appearing in the names of c(4) organizations in the sample. The top words were, in order: fire, association, volunteer, housing, development, club, health, school, county, and community. While social welfare organizations cannot provide any substantial benefits to private individuals or firms, they can engage in extensive political activity. A c(4) can lobby without limit in pursuit of its social welfare mission. Commentators disagree about the extent of a c(4)’s permissible spending for political campaigns, with the most aggressive practitioners asserting that spending of up to 49% of firm income is permissible (for an overview of both sides, see Dougherty, 2013). Because donations to social welfare organizations are protected by tax secrecy laws, the c(4) is an attractive spending vehicle for actors who prefer to make anonymous political expenditures. Political organizations probably make up less than one-quarter of c(4) firms, however (Dimmery and Peterson, 2016, p. 60). Commonly, sophisticated charities that believe political engagement is part of their mission have a related c(4), often with partly or fully overlapping boards, staff, and assets (Kerlin and Reid, 2010). In some firms, the division is entirely on paper. Staff time and other expenditures can be booked to the c(4) when political in nature and to the c(3) otherwise (Leff, 2009). Similarly, some advocacy organizations maintain a related c(3) to serve as a recipient of deductible contributions and grants from private funders. A charity can transfer funds to a c(4) without limit, but cannot earmark transferred funds for activities it could not conduct itself. Most commentators believe that Supreme Court rulings from the early 1980s require the IRS to accept this structure, even if it permits c(3) firms to pass through considerable value to related c(4)s (for a dissenting view on what the Court’s holdings require, see Galle, 2013b). In many advocacy organizations, a c(4) further shares resources and staff with a political action committee (Kerlin and Reid, 2010). These arrangements can be a source of “dark money,” as the c(4) shields donors’ funds from disclosure that would otherwise apply to PAC contributions (Dougherty, 2013). The other core difference between §501(c)(3) and §501(c)(4) is the ability to receive deductible charitable contributions. Both sets of organizations are exempt from the U.S. corporate income tax, thereby subsidizing contributions of investment property, except to the extent investment income is spent on political activity (T.C. §527(f); Halperin, 2011,2018). Further, contributions to either type of firm are not treated as a taxable event for the donor, so that transfers of assets with untaxed appreciation effectively eliminate any tax on the asset’s built-in gains.3 In essence, the donor enjoys a zero tax rate on any property used to fund gifts to charity or social welfare groups.4 Most donated property is reported to have very low basis (Ackerman and Auten, 2011, p. 660), so the value of this exclusion may approximate the tax due on the full value of the property. Whether the average donor appreciates the distinction between these two entity forms is uncertain. Often, similarly named firms can appear in either category, such as the ACLU (a c(4)) and the ACLU Foundation (a charity). A 1987 federal law requires that mass tv, print, phone, or radio solicitations by noncharitable firms must include a “prominent” statement that gifts are not tax deductible (T.C. §6113; see IRS Notice 88-120 for additional detail). Application of these rules to the internet remains unclear. An investigation of a random sample of 20 c(4) organizations determined that all disclosed the nondeductibility of gifts somewhere on their website, but often this information was located at the bottom of a page reached only after multiple clicks. The IRS offers little guidance in its instructions for the individual tax return. The instructions to schedule A of the Form 1040 do not mention §501(c)(4) or even social welfare organizations (IRS, 2006a).5 Instead, the instructions offer examples of organizations that would be eligible for donations: “Boy Scouts, Boys and Girls Club of America, CARE, Girl Scouts, Goodwill Industries, Red Cross, Salvation Army, United Way, etc.” The closest the instructions come to identifying the existence of ineligible donee firms is when they note that “[g]ifts to civic leagues, social and sports clubs, labor unions, and chambers of commerce” are not deductible. These appear to be references to firms eligible for tax exemption, but not deductible contributions, under sections 501(c)(5), (6), (7), and (8). It is also unclear whether the IRS can reliably identify donations to social welfare organizations that are erroneously (or deliberately) deducted. Donors must keep records evidencing their donations in the event of audit, but generally need not disclose the identity of donee firms on their tax return (IRS, 2006a). In-kind gifts of more than |${\$}$|500 must be separately reported on the Form 8283, where the taxpayer must provide the name of the recipient organization (IRS, 2006b). Donee organizations during the period of my sample reported all contributors of |${\$}$|5,000 or more on Schedule B of the Form 990 organizational tax return, but it is unknown whether IRS had the technical capability to match the Schedule B to individual donor returns. Even in the event an individual were audited and required to show their donation records, the auditor would need to take the additional step of individually verifying the charitable status of each donee firm; since firm names do not clearly reveal an organization’s status, it is not obvious why auditors would do so (unless they had read this article).6 In July of 2018, the IRS announced a new policy in which it would not require 501(c)(4) organizations to report their donors to the government, making verification even more challenging (Rev. Proc. 2018-38). Notably, the accompanying IRS press release states that there is no administrative need for donor information (U.S. Department of the Treasury, 2018), a premise my results suggest may be mistaken. 2.2. Hypotheses: Charitable Contribution Deductions and c(4) Firms These facts suggest a potentially complex relationship between government subsidies for charitable contributions and donations to c(4) organizations. Most straightforwardly, if gifts to c(3) and c(4) organizations are (uncompensated) substitutes, donations to c(4) firms should fall when giving to charity is relatively less expensive. There is no theoretical reason to expect that donations to the two forms would be instead be complements, except in the case where one organization is conducting business through linked c(3) and c(4) firms.7 Therefore, as the after-tax price of charitable giving falls, donors who understand and abide by deductibility rules should more strongly prefer to make a deductible contribution to a c(3) entity rather than a nondeductible contribution to a c(4).8 Deliberate misclaiming could instead produce a negative correlation between the tax price of charitable giving and donations to a c(4). In the model described informally in Slemrod (1989), taxpayers’ propensity to overclaim charitable contribution deductions is a function of tax rates, detection probability, and magnitude of sanction (see also Feldstein, 1999, which offers a more general model in which taxpayers adjust their reporting to claim tax benefits until the marginal cost of doing so exceeds marginal returns). It appears that at least for cash contributions of under |${\$}$|5,000, and perhaps more, the ability of the IRS to detect c(4) donations reported as charitable contributions is small, and may be near zero. This should be evident to taxpayers who can comprehend the instructions on IRS forms. Slemrod’s model therefore predicts a positive association between tax rates and misreported c(4) contributions, though empirically he finds that the misreporting elasticity is rather smaller in magnitude than the price-elasticity of actual charitable giving. Behavioral models in which taxpayers understand the tax system imperfectly could also produce a negative correlation between c(4) contributions and the tax price of charitable giving. Again, several prior studies find that taxpayers can be overly responsive to complex tax incentives because they fail to understand legal limits on those incentives (Gallagher and Muehlegger, 2011; Goldin and Listokin, 2013; Katuscak and Kawano, 2016). Eckel and Grossman (2017) also report a field experiment in which they find donors have varying awareness of government subsidies. As in those settings, charitable giving is a complex regime in which the average donor is presented with very limited information about whether her contribution is deductible. For misinformed donors, we might expect that contributions to c(4)’s will mirror their giving to charities and have similar price-elasticities. The prediction that price-elasticities of giving will be similar for both organizational forms assumes, though, that donor confusion is exogenous to tax price. Chetty et al. (2009) suggest that high stakes may motivate greater investment in learning to understand the tax system. That is, suppose that there are two groups of givers in the population, those who understand tax rules and those who do not. Following Chetty et al. (2009), we might call these naïves and sophisticates. Observable population-level elasticities for giving to c(4)’s will reflect a weighted mix of highly negative-elasticity naïves and zero- or positive-elasticity sophisticates. The “high stakes” hypothesis is that the share of sophisticates in the population is larger when the marginal tax savings of a charitable contribution deduction is greater. Accordingly, the high-stakes prediction would be that observed elasticities would be closer to zero when the tax price of giving is closer to unity (i.e. if |$\beta _s $| is the charitable-price-elasticity of giving to a c(4) organization, and is typically negative, then |$\frac{\partial \beta _s} {\partial p}$| is positive). Empirical evidence for the high-stakes theory is mixed, however, and largely derives from experimental settings. Taubinsky and Rees-Jones (2017) find that very large increases in tax rates increase consumer attentiveness to taxes. Feldman et al. (2018) find to the contrary, with higher rates actually lowering attentiveness, albeit over less-dramatic variation in simulated tax rates. They suggest that this effect may be produced by confirmation bias: consumers selectively ignore tax when it is inconsistent with their pre-tax preferences (Galle, 2013a describes this theory in more detail). In sum, we should not expect the price-elasticity of donations to c(4) organizations to be constant across the distribution of tax prices, but instead that donations will have a nonlinear relationship to tax price. Although the rival behavioral theories make contending predictions about how elasticity will vary with tax price, my setting likely does not allow for clean identification of one or the other. Contending factors within a purely rational framework—substitution effects and misreporting—already may tend to produce elasticities that are positive over some range of the distribution and negative over others, confounding any effort to test the behavioral hypotheses. In any event, these likely nonlinearities motivate my use of higher-order terms of tax price in the regression analysis, as detailed more below. 3. Data Data on the fiscal behavior of social welfare organizations are drawn from individual Form 990 tax returns filed annually by exempt organizations and compiled by the National Center on Charitable Statistics in their SOI-Other files. I collect data for tax years ranging from 1991 to 2007.9 My main outcome variable of interest is “direct public support,” or donations received from the public and not via charities or other donation aggregators. Each annual SOI-Other file is a stratified sample of the population of tax returns filed by noncharitable exempt organizations, with overweighting of firms with greater assets; firms with more than |${\$}$|10 million in assets are always included. Stratification naturally raises concerns about whether the criteria for weighting may be correlated with outcomes of interest. My results are robust to sample weighting. The alternative source of information is the Core-Other files, also available via NCCS, which provide a more limited set of variables for the full population of firms. Unfortunately, the Core files fail to disaggregate individual contributions from government grants and “indirect” support, or contributions from other firms, but I employ them for robustness testing. I clean the data following protocols described in Galle (2016). These steps result in the omission of 608 firm-year observations, leaving a total of 21,040 firm-years for c(4) organizations. I also restate negative expense items at their absolute value, resulting in 148 changes. Most of the regression results limit the sample to firms that ever report receiving donations, yielding 9,186 firm-years. For some regression analysis, I attempt to match social welfare organizations with related c(3) firms. Guided by a fuzzy matching algorithm based on firm names and zip codes, I hand match c(4) entities in the SOI-Other database to related charitable organizations in the IRS Business Master File listing of all approved charities. I code a social welfare firm as paired with a matching charity if either entity reports the other as “related” on Schedule R of its tax return. I also code firms as related if they share a common address and identify one or more common individuals in their listings of highly compensated employees or board members. Finally, I code two firms as related if either makes reference to the other on its official web page.10 Ultimately, I have 314 firm-year c(4) observations in the SOI-Other file with matched c(3) data. For analysis using the Core-Other files, I repeated this process using the much larger set of Core firms, yielding 1,558 matched firm-years. I measure the after-tax cost of donating to charity using three alternative but similar measures. In the reported regressions, I employ the state-year price facing taxpayers in each state’s top tax bracket, as computed by Bakija and Heim (2011) using the calculator described in Bakija (2009). This variable, which I call the “after-tax price” or (following the standard term in the literature) “tax price of giving,” represents the net cost of an additional dollar of charitable contributions in a given state and year, taking account of the combined impact of state and federal taxes. A larger charitable contribution deduction results in a lower after-tax price. As the Supplementary Appendix Section A.1 describes in more detail, most of the within-state tax price variation in the sample is caused by changing federal law and its interactions with cross-sectional variations in state rules. For instance, increases in federal marginal rates reduce the value of a state charitable contribution deduction, but for a set of “control” states with no income tax or no deduction, those alterations have no impact. Figure 1 plots the log after-tax price of giving in these two groups of states over time. The vertical line at 2002 represents the January 1, 2003 effective date of “JGTRRA,” the Jobs Growth and Tax Relief Reconciliation Act of 2003. Figure 1. Open in new tabDownload slide Log After-Tax Price of Giving in States With and Without Income Tax. Notes: Prices shown are mean cost of donating |${\$}$|1 to charity, net of state and federal tax, for simulated top-bracket taxpayers in each state. Prices derived from Bakija and Heim (2011). Figure 1. Open in new tabDownload slide Log After-Tax Price of Giving in States With and Without Income Tax. Notes: Prices shown are mean cost of donating |${\$}$|1 to charity, net of state and federal tax, for simulated top-bracket taxpayers in each state. Prices derived from Bakija and Heim (2011). For robustness, I also use the dollar-weighted mean tax price paid by all taxpayers in each state-year, as computed by Duquette (2016). A third method is described in Supplementary Appendix Section A.1. I additionally control for capital gains tax rates. State and federal top effective capital gains rates are taken from NBER taxsim calculations, as summarized in Feenberg and Coutts (1993). In essence, all of these sets of tax data draw a random sample of exogenously itemizing 1984 taxpayers, deflate their data to a subsequent tax year, and compute the marginal effect on tax liability of an additional dollar of contribution or capital gains, respectively, under combined state and federal tax rules applicable in that year.11 I compute average yearly S&P 500 value using monthly S&P means from Robert Schiller’s web site, with each firm-year observation representing the mean of the 12 monthly observations falling within the firm’s fiscal year. All dollar values are deflated to 2007 dollars using the chained-CPI index for the year and month ending the firm’s fiscal year. I obtain state demographic and fiscal variables from the U.S. Census. For variables reported on a calendar-year basis, including year fixed effects, I match firm fiscal years to calendar year based on the calendar year in which a majority of the fiscal-year months fall. Fiscal years ending in June are assigned to the prior calendar year. Table 1 provides simple summary data of the variables I employ in the regression analysis, with separate means for firms that do or do not report receiving donations during the sample period. Table 1. Summary Statistics . All Firms . Firms with Donations . Variables . Mean . SD . Mean . SD . Firm variables Direct public support 628.5 13.2|$^{*}$| 1.44|$^{*}$| 19.9|$^{*}$| Indirect support 165.5 4.3|$^{*}$| 171.5 4.37|$^{*}$| Grants 984.8 22.0|$^{*}$| 1.25|$^{*}$| 26.0|$^{*}$| Program Svc. revenues 29.8|$^{*}$| 203|$^{*}$| 8.47|$^{*}$| 61.3|$^{*}$| Interest revenue 309.1 1.95|$^{*}$| 106.9 984.3 Dividend revenue 485.2 8.3|$^{*}$| 296.1 2.78|$^{*}$| Total revenues 34.9|$^{*}$| 215|$^{*}$| 13.3|$^{*}$| 74.8|$^{*}$| Officer compensation 153.4 1.2|$^{*}$| 89.8 524.6 Program Svc. expenditures 30.9|$^{*}$| 194|$^{*}$| 11.0|$^{*}$| 66.5|$^{*}$| Management expenses 2.6|$^{*}$| 20.7|$^{*}$| 1.19|$^{*}$| 7.82|$^{*}$| Fundraising expenses 102.6 1.4|$^{*}$| 232.8 2.07|$^{*}$| Outside fundraising exp. 8.2 143.5 18.6 216.7 Gross assets 48.0|$^{*}$| 547|$^{*}$| 15.2|$^{*}$| 69.3|$^{*}$| Liabilities 33.8|$^{*}$| 520|$^{*}$| 6.05|$^{*}$| 35.7|$^{*}$| Fundraising share .016 .080 .031 10.6 S&P 12-month avg. 929.11 377.0 961.6 365.3 State variables State tax price |$-$|0.086 0.042 |$-$|0.086 0.040 Population 11.2|$^{*}$| 9.8|$^{*}$| 10.4|$^{*}$| 8.7|$^{*}$| Share under 26 0.36 0.02 0.35 0.02 Share over 64 0.13 0.019 0.13 0.020 Net capital gains rate 26.5 5.40 26.0 5.26 . All Firms . Firms with Donations . Variables . Mean . SD . Mean . SD . Firm variables Direct public support 628.5 13.2|$^{*}$| 1.44|$^{*}$| 19.9|$^{*}$| Indirect support 165.5 4.3|$^{*}$| 171.5 4.37|$^{*}$| Grants 984.8 22.0|$^{*}$| 1.25|$^{*}$| 26.0|$^{*}$| Program Svc. revenues 29.8|$^{*}$| 203|$^{*}$| 8.47|$^{*}$| 61.3|$^{*}$| Interest revenue 309.1 1.95|$^{*}$| 106.9 984.3 Dividend revenue 485.2 8.3|$^{*}$| 296.1 2.78|$^{*}$| Total revenues 34.9|$^{*}$| 215|$^{*}$| 13.3|$^{*}$| 74.8|$^{*}$| Officer compensation 153.4 1.2|$^{*}$| 89.8 524.6 Program Svc. expenditures 30.9|$^{*}$| 194|$^{*}$| 11.0|$^{*}$| 66.5|$^{*}$| Management expenses 2.6|$^{*}$| 20.7|$^{*}$| 1.19|$^{*}$| 7.82|$^{*}$| Fundraising expenses 102.6 1.4|$^{*}$| 232.8 2.07|$^{*}$| Outside fundraising exp. 8.2 143.5 18.6 216.7 Gross assets 48.0|$^{*}$| 547|$^{*}$| 15.2|$^{*}$| 69.3|$^{*}$| Liabilities 33.8|$^{*}$| 520|$^{*}$| 6.05|$^{*}$| 35.7|$^{*}$| Fundraising share .016 .080 .031 10.6 S&P 12-month avg. 929.11 377.0 961.6 365.3 State variables State tax price |$-$|0.086 0.042 |$-$|0.086 0.040 Population 11.2|$^{*}$| 9.8|$^{*}$| 10.4|$^{*}$| 8.7|$^{*}$| Share under 26 0.36 0.02 0.35 0.02 Share over 64 0.13 0.019 0.13 0.020 Net capital gains rate 26.5 5.40 26.0 5.26 Notes: Dollar values in thousands of 2007 dollars. *millions Open in new tab Table 1. Summary Statistics . All Firms . Firms with Donations . Variables . Mean . SD . Mean . SD . Firm variables Direct public support 628.5 13.2|$^{*}$| 1.44|$^{*}$| 19.9|$^{*}$| Indirect support 165.5 4.3|$^{*}$| 171.5 4.37|$^{*}$| Grants 984.8 22.0|$^{*}$| 1.25|$^{*}$| 26.0|$^{*}$| Program Svc. revenues 29.8|$^{*}$| 203|$^{*}$| 8.47|$^{*}$| 61.3|$^{*}$| Interest revenue 309.1 1.95|$^{*}$| 106.9 984.3 Dividend revenue 485.2 8.3|$^{*}$| 296.1 2.78|$^{*}$| Total revenues 34.9|$^{*}$| 215|$^{*}$| 13.3|$^{*}$| 74.8|$^{*}$| Officer compensation 153.4 1.2|$^{*}$| 89.8 524.6 Program Svc. expenditures 30.9|$^{*}$| 194|$^{*}$| 11.0|$^{*}$| 66.5|$^{*}$| Management expenses 2.6|$^{*}$| 20.7|$^{*}$| 1.19|$^{*}$| 7.82|$^{*}$| Fundraising expenses 102.6 1.4|$^{*}$| 232.8 2.07|$^{*}$| Outside fundraising exp. 8.2 143.5 18.6 216.7 Gross assets 48.0|$^{*}$| 547|$^{*}$| 15.2|$^{*}$| 69.3|$^{*}$| Liabilities 33.8|$^{*}$| 520|$^{*}$| 6.05|$^{*}$| 35.7|$^{*}$| Fundraising share .016 .080 .031 10.6 S&P 12-month avg. 929.11 377.0 961.6 365.3 State variables State tax price |$-$|0.086 0.042 |$-$|0.086 0.040 Population 11.2|$^{*}$| 9.8|$^{*}$| 10.4|$^{*}$| 8.7|$^{*}$| Share under 26 0.36 0.02 0.35 0.02 Share over 64 0.13 0.019 0.13 0.020 Net capital gains rate 26.5 5.40 26.0 5.26 . All Firms . Firms with Donations . Variables . Mean . SD . Mean . SD . Firm variables Direct public support 628.5 13.2|$^{*}$| 1.44|$^{*}$| 19.9|$^{*}$| Indirect support 165.5 4.3|$^{*}$| 171.5 4.37|$^{*}$| Grants 984.8 22.0|$^{*}$| 1.25|$^{*}$| 26.0|$^{*}$| Program Svc. revenues 29.8|$^{*}$| 203|$^{*}$| 8.47|$^{*}$| 61.3|$^{*}$| Interest revenue 309.1 1.95|$^{*}$| 106.9 984.3 Dividend revenue 485.2 8.3|$^{*}$| 296.1 2.78|$^{*}$| Total revenues 34.9|$^{*}$| 215|$^{*}$| 13.3|$^{*}$| 74.8|$^{*}$| Officer compensation 153.4 1.2|$^{*}$| 89.8 524.6 Program Svc. expenditures 30.9|$^{*}$| 194|$^{*}$| 11.0|$^{*}$| 66.5|$^{*}$| Management expenses 2.6|$^{*}$| 20.7|$^{*}$| 1.19|$^{*}$| 7.82|$^{*}$| Fundraising expenses 102.6 1.4|$^{*}$| 232.8 2.07|$^{*}$| Outside fundraising exp. 8.2 143.5 18.6 216.7 Gross assets 48.0|$^{*}$| 547|$^{*}$| 15.2|$^{*}$| 69.3|$^{*}$| Liabilities 33.8|$^{*}$| 520|$^{*}$| 6.05|$^{*}$| 35.7|$^{*}$| Fundraising share .016 .080 .031 10.6 S&P 12-month avg. 929.11 377.0 961.6 365.3 State variables State tax price |$-$|0.086 0.042 |$-$|0.086 0.040 Population 11.2|$^{*}$| 9.8|$^{*}$| 10.4|$^{*}$| 8.7|$^{*}$| Share under 26 0.36 0.02 0.35 0.02 Share over 64 0.13 0.019 0.13 0.020 Net capital gains rate 26.5 5.40 26.0 5.26 Notes: Dollar values in thousands of 2007 dollars. *millions Open in new tab 4. Identification Donation regressions in the main text are estimated using an equation of the form: $$\begin{align}\label{eq1} D_{it}&= \alpha _{i}+\beta _1 \textit{TaxPrice}_{st} +\beta _2 TaxPrice_{st} \ \ast\ \textit{TaxPrice}_{st} +\beta _3 CG_{st}\nonumber\\ &\quad +\beta _4 S\& P_{it} +\beta _5 S\& P_{it - 1} +\beta _6 X_{it} +\beta _7 W_{st} +\lambda _{t}+\phi _s \lambda _{t}+\varepsilon_{it}, \end{align}$$(1) where TaxPrice is the mean after-tax cost of donating to charity for top-bracket taxpayers in a given state-year, where |$s$| indexes states and |$t$| time, CG is the proxy for the net capital gain savings from excluding built-in gains, and S&P and S&P|$_{t-1}$| are the same-year and lagged values of the S&P variable, with |$i$| indexing firms.12|$X$| is a vector of firm controls, while |$W$| is a vector of state-level controls. |$\varphi _s \lambda _{t}$| is a set of state by year trends. Tax price and its square are my main variables of interest. In essence, I am looking at the correlation between within-firm changes in tax prices confronting in-state donors and the reported donations received by the firm. The results reported rely on the top-bracket tax price computed in Bakija and Heim (2011). That is, the tax price variable represents the mean cost of donating one dollar to charity, net of both federal and state charitable contribution deductions, for donors who face the highest tax rate in a given state-year.13 Alternatively, I also estimate results (not tabulated) using the dollar-weighted mean for all state-year taxpayers, as computed in Duquette (2016). Results are qualitatively similar using this estimate. As Yetman and Yetman (2012) do, I prefer the top-bracket tax price. Because there is little year-over-year variation in the tax price for most middle-bracket donors (Bakija and Heim, 2011), any effects are more difficult to identify using averages that include a large number of these donors. As Supplementary Appendix Section A.1 details, I also extend the Galle (2016) methodology for identifying the price-elasticity of giving to charities to donations to c(4) organizations. Galle (2016) uses a form of difference-in-differences approach to control nonparametrically for unobservables. Results using the tax price measure just described are almost identical to those obtained using the DD method. For simplicity, therefore, the main text omits the DD results. In a sense, however, even the vanilla tax price represents a kind of DD methodology. With year effects in the regression, states without an income tax should see little year-over-year variation in tax price, as the year effects will absorb the impact of federal tax changes. Firms in these states can thus be thought of as the control group.14 Identification accordingly depends on the assumption that there are no unobserved shocks correlated with both being located in a state with an income tax and with the timing of JGTRRA or the Pease phaseout, and which would not be accounted for with state by year trends. Donations to both c(3) and c(4) organizations allow the donor to exclude from tax any built-in gains from in-kind property. Bakija and Heim (2011, pp. 621–22) include in their price variable an estimate of the value of making gifts with unrealized built-in gain. I would like to identify separately the effects of permissible exclusion of built-in gain from impermissible “charitable” contribution deductions to 501(c)(4) organizations. I therefore include a separate control for these excluded capital gains, in order to isolate in the Bakija and Heim price the impact of the charitable contribution deduction.15 To measure capital gains rates, I attempt to replicate as closely as possible the calculation Bakija and Heim (2011) employ, but my data are organized at the firm level and not by individual donor. As they do, I use the combined state-federal marginal burden on capital gains for state taxpayers, as computed by NBER taxsim (described more in 3.0 above). I then modify this figure, multiplying it by the product, |$n_{bt}$| * |$s_{bt}$| * |$a$| * |$d$| * mcg|$_{bt + 1}$|⁠. As Bakija and Heim (2011) explain, their calculation of the value of untaxed appreciation derives from using the lead of the capital gains rate, mcg|$_{t+1}$|⁠, on the assumption that donors’ alternative to donation now is often a gift in a later year. They then multiply this value by |$n_{it}$|⁠, the share of gifts made in kind by like-bracket donors, then by |$s_{it}$|⁠, the portion of in-kind gifts made in stock for individuals in the same bracket, and finally by |$a$|⁠, the average historic gain-to-value ratio of donated stock. This figure is then discounted to present-year value at the rate |$d$|⁠. To replicate this method as closely as possible, I repeat these calculations, using top-bracket (⁠|$b)$| means for each year |$t$|⁠. I call the resulting variable “net capital gain rate.”16 The use of state-level tax variables may lead to some attenuation. Tax benefits are generally determined by the state of the donor, not the recipient organization. Therefore, a key assumption of my model is that most giving is within state, and that cross-state donations will tend to average out over time. A number of studies find that more than 75% of individual and foundation gifts are to within-state or otherwise spatially proximate organizations (Center on Philanthropy at Indiana University, 2008a, b; Glückler and Ries, 2012). Given that many c(4) organizations are community groups, that figure may overstate the geographic dispersion of many c(4) supporters.17 To test the extent to which attenuation contributes to the results, I also hand-code each donation-receiving c(4) firm for indicators of whether the firm provides regional or nation-wide services, and therefore is more likely to receive cross-state donations. Firms are coded based on geographic indicators in their titles, mission descriptions on their tax returns, or current web site. I code organizations as regional or national when these descriptors include terms such as national, nationwide, “across the United States,” global, and so on. Firms are also coded as national when it is clear from context that they serve members or beneficiaries across the country even if none of these exact geographic descriptors are used. Stock values and stock momentum can impact gifts to charities List and Peysakhovich 2010. To allow for the possibility that wealth effects and recent directional changes in stock values may have independent effects on giving, I control for lagged and same-year average value of the S&P 500. I further control for basic firm-level variables, such as investment revenue, officer compensation, liabilities, and lagged assets (since donations would affect same-year assets). In theory, if changes in the tax environment for charities affect giving to social welfare firms, those changes may also affect other firm outcomes, such as revenues from alternative sources, and especially fundraising expenditures (Supplementary Appendix Section A.2 provides evidence that tax price affects fundraising by c(4) firms). To avoid “bad controls” selection effects, therefore, I omit these outcome variables from reported regressions with donations on the left-hand side. It is possible that some secular trend might affect charitable giving among “treated” states—those with an income tax—but not untreated states. Such a trend could result in a spurious correlation between tax price and donations. I therefore control for state by year trends. Results are similar when excluding trends or including both linear and quadratic trends, although the specification in which I include controls for fundraising loses statistical significance with quadratic trends included. Additionally, I include calendar-year fixed effects and a set of state demographic variables, comprising GDP, unemployment rate, median income, population, and share of the population over 64 and under 26. Variables are estimated in logs, with the exception of the outcome variable and of course indicators. Many firms in the sample report zero donations or zero fundraising for some or all of their firm-years. When either of these variables appears on the left-hand side I do not log them. Instead, I estimate marginal effects using a fixed-effects panel negative binomial regression. Negative binomial estimation is often used for count variables, but since it is an exponential function, when the left-hand variable is estimated in levels and the right-hand variables are in logs, it can be interpreted similarly to, and substituted for, traditional log-log estimates (Woolridge, 2010; see Almunia et al., 2020 for application to charitable giving). It therefore avoids the need to add some arbitrary, and potential biasing, amount to zero outcomes before logging. The implementation of the fixed-effects panel negative binomial estimator in Stata omits firms for which the outcome variable is always zero. That approach makes theoretical sense in my sample. Many social welfare organizations, such as health insurance cooperatives, do not have a business model that involves donations. Including these firms would presumably bias any result towards zero. For similar reasons, Andreoni and Payne (2013) omit firms with zero reported gifts, grants, or fundraising from their sample (Duquette, 2016 similarly omits firms without donations). In regressions with fundraising on the left-hand side (Supplementary Appendix Section A.2), I omit firms that never report gifts as well as firms that never report fundraising. I cluster standard errors at the state level because that is the level of treatment. Due to technical limitations in Stata, I am obliged to implement clustering using wild bootstrap standard errors (Cameron et al., 2008). 4.1. Linear or Quadratic Estimates? Once more, theory suggests that the relationship between tax price and donations to 501(c)(4) organizations may not be linear. Graphical analysis confirms that quadratic estimates may better capture the relationship between tax price and donations.18Figure 2 plots 40 equal-width bins of log donations by their mean tax price, limiting the sample to c(4) firms that ever report receiving donations. A quadratic best-fit line through the bin means forms a parabola. Figure 2. Open in new tabDownload slide Scatterplot of Forty Equal-Width Bins of Residualized Log Donations vs. Log Tax Price Notes: Solid line = quadratic fit line. Bins defined by log donations and log tax price, residualized by controlling for firm and state fixed effects. Sample limited to 501(c)(4) organizations ever reporting nonzero donations. Figure 2. Open in new tabDownload slide Scatterplot of Forty Equal-Width Bins of Residualized Log Donations vs. Log Tax Price Notes: Solid line = quadratic fit line. Bins defined by log donations and log tax price, residualized by controlling for firm and state fixed effects. Sample limited to 501(c)(4) organizations ever reporting nonzero donations. As a result, in regressions for which donations are the outcome variable, I include both tax price and tax price squared as predictors. 5. Results I investigate the determinants of two sets of social welfare firm outcomes. Section 5.1 examines the predictors of donations, and Section 5.2 reports robustness testing using the larger Core-Other files. 5.1. Determinants of Direct Public Support I first consider the determinants of giving to social welfare organizations among firms that ever report receiving donations from the public. For my baseline estimates, I omit other “outcome” variables that might also be influenced by the tax price of giving, such as fundraising. These estimates thus correspond to what Galle (2016) dubs the “reduced-form” elasticity, |$E_{pD}^R $|⁠. It is also of interest, however, to separately identify the extent to which any effects may be caused by fundraising. The reduced-form elasticity can be decomposed into its structural components, as in equation 2: $$\begin{align}\label{eq2} E_{pD}^R = E_{pD}^S + E_{pF} \cdot E_{FD} \end{align}$$(2) where the right-hand side values are the “structural” components of donations. That is, |$E_{pD}^S $| is the price-elasticity of giving with fundraising held constant, or |$\frac{\partial D(P,F)}{\partial P}\cdot \frac{P}{D(P,F)}$| The value |$E_{pF} $| is the price-elasticity of fundraising, |$\frac{dF(P)}{dP}\cdot \frac{P}{F}$|⁠. And the last value, |$E_{FD} $|⁠, is the fundraising elasticity of donations, |$\frac{\partial D(P,F)}{\partial F} \cdot \frac{F}{D(P,F)}$|⁠.19 Columns 1 and 2 of Table 2 report specifications in which I directly estimate |$E_{pD}^R $|⁠, and in which I separately estimate |$E_{pD}^S$| and |$E_{FD} $|⁠, respectively. Again, in both cases I include both linear and quadratic terms for the tax price variables.20 Table 2. Effects of Tax Price on Donations to Social Welfare Organizations Variables . (1) . (2) . (3) . (4) . State tax price -10.84*** -6.513** -8.492** -10.50*** (3.368) (2.865) (3.525) (2.973) State tax price square -12.82*** -8.455** -10.04** -12.07*** (3.787) (3.302) (4.044) (3.424) Log net capital gains -0.459 -0.0162 -1.024* -1.216*** (0.887) (0.402) (0.523) (0.423) Log S&P 500 -0.0326 -0.231 -0.175 -0.0196 (0.365) (0.332) (0.458) (0.361) Lag of Log S&P 500 0.236 0.0327 -0.484 0.0506 (0.356) (0.318) (0.437) (0.340) Log fundraising 0.353*** (0.0170) Log fundraising share -0.177*** (0.0224) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of Tax Price 3.34*** 2.84*** 2.61** 2.84*** (1.16) (0.99) (1.20) (1.03) |$\quad$||$-1$|SD of Tax Price 2.31** 2.16** 1.81* 1.88** (0.95) (0.78) (0.95) (0.81) |$\quad$| Median Tax Price 0.47 0.94 0.365 0.144 (0.73) (0.54) (0.67) (0.56) |$\quad$||$+$|1 SD of Tax Price -1.66* -0.46 -1.30 -1.86*** (0.93) (0.70) (0.87) (0.73) |$\quad$||$+$|2 SD of Tax Price -3.17*** -1.46 -2.49** -3.28*** (1.25) (1.00) (1.24) (1.03) Observations 8,723 8,676 6,539 8,448 Number of Firms 1139 1139 933 1100 Variables . (1) . (2) . (3) . (4) . State tax price -10.84*** -6.513** -8.492** -10.50*** (3.368) (2.865) (3.525) (2.973) State tax price square -12.82*** -8.455** -10.04** -12.07*** (3.787) (3.302) (4.044) (3.424) Log net capital gains -0.459 -0.0162 -1.024* -1.216*** (0.887) (0.402) (0.523) (0.423) Log S&P 500 -0.0326 -0.231 -0.175 -0.0196 (0.365) (0.332) (0.458) (0.361) Lag of Log S&P 500 0.236 0.0327 -0.484 0.0506 (0.356) (0.318) (0.437) (0.340) Log fundraising 0.353*** (0.0170) Log fundraising share -0.177*** (0.0224) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of Tax Price 3.34*** 2.84*** 2.61** 2.84*** (1.16) (0.99) (1.20) (1.03) |$\quad$||$-1$|SD of Tax Price 2.31** 2.16** 1.81* 1.88** (0.95) (0.78) (0.95) (0.81) |$\quad$| Median Tax Price 0.47 0.94 0.365 0.144 (0.73) (0.54) (0.67) (0.56) |$\quad$||$+$|1 SD of Tax Price -1.66* -0.46 -1.30 -1.86*** (0.93) (0.70) (0.87) (0.73) |$\quad$||$+$|2 SD of Tax Price -3.17*** -1.46 -2.49** -3.28*** (1.25) (1.00) (1.24) (1.03) Observations 8,723 8,676 6,539 8,448 Number of Firms 1139 1139 933 1100 Notes: Fixed-effects panel regressions using negative binomial estimation. Wild bootstrap standard errors clustered by state in (parenthesis). Nonindicator independent variables logged. Column three omits firms with zero fundraising. Column four omits firms with a matched c(3) sister organization. All columns include controls for calendar year and state by year trends, firm assets, liabilities, program expenses, investment gain/loss, officer compensation, dividend income, and interest income, as well as state median income, gdp, population, population share under 26, and population share over 64. Predicted Marginal Effects rows report estimated combined impact of tax price and price squared at the identified points in the distribution of tax price; SD = standard deviation. **Statistically significant at 5% level. ***Statistically significant at 1% level. Open in new tab Table 2. Effects of Tax Price on Donations to Social Welfare Organizations Variables . (1) . (2) . (3) . (4) . State tax price -10.84*** -6.513** -8.492** -10.50*** (3.368) (2.865) (3.525) (2.973) State tax price square -12.82*** -8.455** -10.04** -12.07*** (3.787) (3.302) (4.044) (3.424) Log net capital gains -0.459 -0.0162 -1.024* -1.216*** (0.887) (0.402) (0.523) (0.423) Log S&P 500 -0.0326 -0.231 -0.175 -0.0196 (0.365) (0.332) (0.458) (0.361) Lag of Log S&P 500 0.236 0.0327 -0.484 0.0506 (0.356) (0.318) (0.437) (0.340) Log fundraising 0.353*** (0.0170) Log fundraising share -0.177*** (0.0224) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of Tax Price 3.34*** 2.84*** 2.61** 2.84*** (1.16) (0.99) (1.20) (1.03) |$\quad$||$-1$|SD of Tax Price 2.31** 2.16** 1.81* 1.88** (0.95) (0.78) (0.95) (0.81) |$\quad$| Median Tax Price 0.47 0.94 0.365 0.144 (0.73) (0.54) (0.67) (0.56) |$\quad$||$+$|1 SD of Tax Price -1.66* -0.46 -1.30 -1.86*** (0.93) (0.70) (0.87) (0.73) |$\quad$||$+$|2 SD of Tax Price -3.17*** -1.46 -2.49** -3.28*** (1.25) (1.00) (1.24) (1.03) Observations 8,723 8,676 6,539 8,448 Number of Firms 1139 1139 933 1100 Variables . (1) . (2) . (3) . (4) . State tax price -10.84*** -6.513** -8.492** -10.50*** (3.368) (2.865) (3.525) (2.973) State tax price square -12.82*** -8.455** -10.04** -12.07*** (3.787) (3.302) (4.044) (3.424) Log net capital gains -0.459 -0.0162 -1.024* -1.216*** (0.887) (0.402) (0.523) (0.423) Log S&P 500 -0.0326 -0.231 -0.175 -0.0196 (0.365) (0.332) (0.458) (0.361) Lag of Log S&P 500 0.236 0.0327 -0.484 0.0506 (0.356) (0.318) (0.437) (0.340) Log fundraising 0.353*** (0.0170) Log fundraising share -0.177*** (0.0224) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of Tax Price 3.34*** 2.84*** 2.61** 2.84*** (1.16) (0.99) (1.20) (1.03) |$\quad$||$-1$|SD of Tax Price 2.31** 2.16** 1.81* 1.88** (0.95) (0.78) (0.95) (0.81) |$\quad$| Median Tax Price 0.47 0.94 0.365 0.144 (0.73) (0.54) (0.67) (0.56) |$\quad$||$+$|1 SD of Tax Price -1.66* -0.46 -1.30 -1.86*** (0.93) (0.70) (0.87) (0.73) |$\quad$||$+$|2 SD of Tax Price -3.17*** -1.46 -2.49** -3.28*** (1.25) (1.00) (1.24) (1.03) Observations 8,723 8,676 6,539 8,448 Number of Firms 1139 1139 933 1100 Notes: Fixed-effects panel regressions using negative binomial estimation. Wild bootstrap standard errors clustered by state in (parenthesis). Nonindicator independent variables logged. Column three omits firms with zero fundraising. Column four omits firms with a matched c(3) sister organization. All columns include controls for calendar year and state by year trends, firm assets, liabilities, program expenses, investment gain/loss, officer compensation, dividend income, and interest income, as well as state median income, gdp, population, population share under 26, and population share over 64. Predicted Marginal Effects rows report estimated combined impact of tax price and price squared at the identified points in the distribution of tax price; SD = standard deviation. **Statistically significant at 5% level. ***Statistically significant at 1% level. Open in new tab In Table 2, the bottom panel labeled “Predicted Net Marginal Effect of Tax Price” reports post-estimation results for the combined impact of tax price, accounting for both its linear and quadratic terms, as calculated using the margins command in Stata 15. To give a sense of the change in coefficient across the distribution, for each column I report the combined effect at the median tax price, as well as at one and two standard deviations above and below the median. These span the range of an after-tax cost of 58 cents for a dollar of giving at the low end to 74 cents at the top end. In line with my predictions, the impact of tax price on giving varies as the tax price changes. For the baseline case, reported in Column 1 of Table 2, the net correlation of tax price with donations is negative and highly significant for prices above the median, where the government’s subsidy per dollar of giving is smallest. At the median, the impact is negative but significant only at the 5% level. Two deviations below median, where government per-dollar subsidies are the largest, tax price is positively and highly significantly correlated with giving—that is, in this range, donations to c(4)’s are declining as the subsidy for giving to c(3)’s increases. Figure 3 offers a more detailed summary of this variation. In this figure, I plot predicted net coefficients by tax price decile. It is evident that the marginal effects of tax price are negative and significant for observations on the right-hand side of the figure, where the after-tax cost of giving is highest. Figure 3. Open in new tabDownload slide Net Marginal Effect of Tax Price on Donations to 501(c)(4) Firms, by Decile of Tax Price Notes: Summarizes results of regression in which the outcome variable is log donations. Predicted marginal effect coefficient is combined linear and quadratic effect of log tax price of charitable contributions. Fixed-effects panel regression using negative binomial estimation. Standard errors clustered by state. Includes controls for capital gains, same-year and lagged S&P 500, firm assets, liabilities, expenses, officer compensation, dividend, and other investment income, as well as calendar year and state median income, state by year trends, gdp, population, population share under 26, and population share over 64. Figure 3. Open in new tabDownload slide Net Marginal Effect of Tax Price on Donations to 501(c)(4) Firms, by Decile of Tax Price Notes: Summarizes results of regression in which the outcome variable is log donations. Predicted marginal effect coefficient is combined linear and quadratic effect of log tax price of charitable contributions. Fixed-effects panel regression using negative binomial estimation. Standard errors clustered by state. Includes controls for capital gains, same-year and lagged S&P 500, firm assets, liabilities, expenses, officer compensation, dividend, and other investment income, as well as calendar year and state median income, state by year trends, gdp, population, population share under 26, and population share over 64. In short, I find evidence that for roughly one-quarter to one-third of all 501(c)(4) firm-year observations, tax subsidies for contributions to charity are correlated with increased contributions to c(4)s, even though these donations are not legally eligible for the subsidy. These effects are relatively large in magnitude. For example, in the deciles where tax price begins to show statistical significance on the negative side, the elasticity of donations received to changes in tax price is about -1, comparable to the price-elasticity of giving to charity reported in other firm-side studies (Yetman and Yetman, 2012; Galle, 2016). Among firms in jurisdictions where the relative subsidy is largest, however, tax price correlates with smaller donations, consistent with a prediction that gifts to charities and social welfare organizations are substitutes.21 Figure 4 further explores this result graphically. To isolate tax price from capital gains and the stock market, I first regress log donations on log capital gains and the S&P 500, and capture the residuals. I then plot these “predicted” gifts against tax price in treatment and control states (i.e. states with no income tax). Although the results are not easy to interpret with the naked eye, it does seem that the relationship between price and giving is negative for treatment states (where the after-tax price of a donation is lower) and positive in control states (where the price is higher). Figure 4. Open in new tabDownload slide Plot of Predicted Gifts Against After-Tax Price of Giving Notes: Predicted gifts are residuals obtained after regressing actual log gifts against log capital gains and log S&P 500. Treated states are states with an income tax, control states without. Figure 4. Open in new tabDownload slide Plot of Predicted Gifts Against After-Tax Price of Giving Notes: Predicted gifts are residuals obtained after regressing actual log gifts against log capital gains and log S&P 500. Treated states are states with an income tax, control states without. Columns 2 through 4 of Table 2 shed some additional light on these surprising results. When I control for fundraising as well as the share of firm expenditures devoted to fundraising, coefficients are still positive and significant for the lower half of the tax price distribution. Coefficients for the upper half of the distribution are still negative but no longer significant. As expected, fundraising increases donations, while the share of firm expenses devoted to fundraising depresses them.22 One potential interpretation of these results is that fundraising by c(4)’s does not affect crowding out but does in part explain the positive relationship between high after-tax costs of charitable giving and donations to c(4)’s. However, in several alternative specifications, such as with instruments for fundraising or regressions omitting state-by-year trends, coefficients on tax price remain negative and significant to the right of median. Another angle on the relative importance of fundraising is to examine the effects of tax price only among firm-years that report zero fundraising expenditures.23 These results are given in Column 3. Omitting firm-years with nonzero fundraising has only modest impact on the estimated coefficients, suggesting that fundraising is not the main driver of my results. Yet another possible explanation for the influence of tax price on social welfare giving is that the results are driven by charities with c(4) affiliates. Changes in tax price would naturally impact gifts to the c(3) entity, but its managers might steer some donations to the c(4).24 I have anecdotal evidence of this phenomenon from anonymous sources who have worked closely with c(3)–c(4) affiliated groups. Alternately, since related firms often have very similar names, some donors might simply confuse the two. Lastly, reputational spillovers from one firm might influence donations to the other (Minkoff, 2002). I find no direct support for these possibilities. As described in Section 3, I hand match c(4) firms in my sample with c(3) entities in the IRS Business Master File. In Column 4, I report results only for firms with no matched c(3) related firm. If anything, the coefficients for the combined tax price variables are a bit larger in absolute magnitude than those for Column 1 at both ends of the distribution. To be sure, it is likely that my matching technique has some number of false negatives, so that the Column 4 results still include some members of affiliated groups. But the fact that the effect of tax price increases when I omit 314 firm-year observations from related firms tends to suggest that these related groups are not driving the result.25 Overall, the data lend support to my hypothesis of a complex relationship between subsidies for charity and giving to noncharitable firms. As I described earlier, however, these results do not allow us to cleanly identify the exact drivers of that complexity. The shift from positive to negative coefficients as the tax price of charitable giving rises has several plausible explanations. If the results were driven entirely by salience and taxpayer confusion, they would align with the experimental findings by Taubinsky and Rees-Jones (2017). As in their study, I find evidence consistent with the theory that naïve donors are a larger share of the donor population when per-dollar subsidies are relatively smaller. When donors have less at stake, they may be more apt to make mistakes. It might be argued that this theory is more appealing in a cross-sectional setting than in one, such as mine, where identification relies on variation within states. Sophisticated donors arguably should not become naïve when subsidies shrink. As Chetty et al. (2009) find, however, individuals may be aware of tax rules without necessarily being willing to invest the cognitive effort of accounting for them. But I cannot clearly disentangle mistaken responses from rational behavior. My results are also potentially explainable by the contending influences of substitution effects and intentional misclaiming. It could be that substitution away from c(4) giving prevails when the tax price of charitable giving is very low, while misclaiming is predominant when tax price is high. In addition, there is some limited evidence to suggest that fundraising might be relevant over some portions of the distribution of tax price, although there is also evidence to the contrary. I investigate the relationship between tax price and fundraising more completely in Supplementary AppendixSection A.2. 5.2. Robustness Analysis 5.2.1. Alternative specifications. Donations to certain veterans organizations may be deductible even if the organization is not recognized as a 501(c)(3) charity (Aprill, 2018). To rule out the possibility that these organizations drive the results, I repeat the regressions (not tabulated) omitting any firm self-reporting itself as a veterans’ organization, or one in which the words “veteran” or “veterans” appear in the entity name. The impact of state tax price and capital gains are slightly larger and more precisely estimated when these 221 firm-years are omitted. As described earlier, attenuation may also be a concern given that I cannot observe the tax rules governing donors. Results (untabulated) are essentially identical when I omit the 674 firm-years coded as regional or national in focus.26 Although in my view estimation using negative binomial regression is the most appropriate response to the presence of firms with zero donations,27 some readers may also be interested in OLS results. I therefore repeat the analysis employing OLS, but using a two-stage Heckman selection model. The first stage is a probit regression where the dependent variable is whether an organization ever reports receiving donations, and the second stage is an OLS estimate, conditioning on the inverse Mills ratio obtained in the first stage. This results in coefficients (untabulated) that are similar to those reported, but with rather larger confidence intervals. 5.2.2. Replication using the core-other files. In order to obtain greater statistical power, I attempt to repeat my regression analysis using a larger set of tax returns. This alternative database, the Core-Other files, contains a much more limited set of variables from each tax return than the SOI files, but includes returns from every filing firm. Most problematically, the Core file does not separately report individual contributions, but instead has a single field, “total support,” that aggregates individual donations, government grants, and contributions from other exempt organizations. Another difficulty for the Core files is that they have been given less editorial attention by IRS staff. A large fraction of firm-year observations have missing or obviously incorrect fiscal years, and many lack address information. I am able to recover or correct some fiscal years by linking reported beginning-of-year and end-of-year asset and liability balances. I also impute missing middle-value zip codes. However, approximately 70,000 observations are omitted because one or both of these errors were uncorrectable. Of the remaining 426,000 or so firm-year observations, only 275,675 ever report nonzero values for total support. In general, results for this subset of 275,675 are broadly similar to the results reported in Sections 5.1 and the Supplementary Appendices. As might be expected given that total support includes not only donations but also two other sources of revenue that might potentially move in the opposite direction from donations, the coefficients in donation regressions are smaller. Table 2 Columns 1 and 2 report full-sample regressions in which “total support” is the outcome variable. The tax price of giving to charity still has a negative and significant sign for a substantial portion of the distribution, with elasticities near or exceeding |$-$|1.0. The effect when controlling for fundraising (Column 2) is essentially the same. I interpret this as additional evidence that fundraising does not explain all of donor responses to tax price. Likewise, coefficients are basically unchanged in the sub-samples with zero fundraising (Table 3 Column 3) or with no linked c(3) organization (Table 3 Column 4). Table 3. Effects of Tax Price on Total Support Among Core-Other 501(c)(4) Organizations Variables . (1) . (2) . (3) . (4) . Log tax price -6.104*** -5.931*** -5.072*** -6.333*** (0.468) (0.465) (0.515) (0.471) Log tax price square -7.066*** -6.880*** -5.914*** -7.365*** (0.538) (0.534) (0.591) (0.541) Log net capital gains 0.325*** 0.342*** 0.217*** 0.303*** (0.0650) (0.0646) (0.0719) (0.0653) Log fundraising exp 0.126*** (0.00148) Log S&P 500 -0.256*** -0.235*** -0.147** -0.258*** (0.0582) (0.0578) (0.0635) (0.0585) Lag of log S&P 0.0315 0.0278 -0.0454 0.0310 (0.0564) (0.0557) (0.0623) (0.0567) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of tax price 1.711*** 1.678*** 1.469*** 1.812*** 0.161 0.160 0.177 0.162 |$\quad$||$-1$|SD of tax price 1.146*** 1.1279*** 0.996*** 1.223*** 0.128 0.127 0.140 0.1283 |$\quad$| Median tax price 0.128 0.137 0.144 0.162* 0.0897 0.0894 0.0984 0.090 |$\quad$||$+$|1 SD of tax price -1.0448*** -1.005*** -0.838*** -1.060*** 0.116 0.115 0.128 0.116 |$\quad$||$+$|2 SD of tax price -1.879*** -1.817*** -1.535*** -1.929*** 0.164 0.163 0.181 0.165 Observations 275,675 275,675 241,965 273,413 Number of firms 30,519 30,519 28,357 30,289 Variables . (1) . (2) . (3) . (4) . Log tax price -6.104*** -5.931*** -5.072*** -6.333*** (0.468) (0.465) (0.515) (0.471) Log tax price square -7.066*** -6.880*** -5.914*** -7.365*** (0.538) (0.534) (0.591) (0.541) Log net capital gains 0.325*** 0.342*** 0.217*** 0.303*** (0.0650) (0.0646) (0.0719) (0.0653) Log fundraising exp 0.126*** (0.00148) Log S&P 500 -0.256*** -0.235*** -0.147** -0.258*** (0.0582) (0.0578) (0.0635) (0.0585) Lag of log S&P 0.0315 0.0278 -0.0454 0.0310 (0.0564) (0.0557) (0.0623) (0.0567) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of tax price 1.711*** 1.678*** 1.469*** 1.812*** 0.161 0.160 0.177 0.162 |$\quad$||$-1$|SD of tax price 1.146*** 1.1279*** 0.996*** 1.223*** 0.128 0.127 0.140 0.1283 |$\quad$| Median tax price 0.128 0.137 0.144 0.162* 0.0897 0.0894 0.0984 0.090 |$\quad$||$+$|1 SD of tax price -1.0448*** -1.005*** -0.838*** -1.060*** 0.116 0.115 0.128 0.116 |$\quad$||$+$|2 SD of tax price -1.879*** -1.817*** -1.535*** -1.929*** 0.164 0.163 0.181 0.165 Observations 275,675 275,675 241,965 273,413 Number of firms 30,519 30,519 28,357 30,289 Notes: Fixed-effects panel regressions using negative binomial estimation. Wild bootstrap standard errors clustered by state in (parenthesis). Nonindicator independent variables logged. Column three omits firms with zero fundraising. Column four omits firms with a matched c(3) sister organization. All columns include controls for calendar year and state-by-year trends, as well as gross assets, liabilities, program service revenue, investment income, investment losses, officer compensation, state median income, gdp, population, population share under 26, and population share over 64. Predicted Marginal Effects rows report estimated combined impact of tax price and price squared at the identified points in the distribution of tax price; SD= standard deviation. *: statistically significant at 10% level. **: statistically significant at 5% level. ***: statistically significant at 1% level. Open in new tab Table 3. Effects of Tax Price on Total Support Among Core-Other 501(c)(4) Organizations Variables . (1) . (2) . (3) . (4) . Log tax price -6.104*** -5.931*** -5.072*** -6.333*** (0.468) (0.465) (0.515) (0.471) Log tax price square -7.066*** -6.880*** -5.914*** -7.365*** (0.538) (0.534) (0.591) (0.541) Log net capital gains 0.325*** 0.342*** 0.217*** 0.303*** (0.0650) (0.0646) (0.0719) (0.0653) Log fundraising exp 0.126*** (0.00148) Log S&P 500 -0.256*** -0.235*** -0.147** -0.258*** (0.0582) (0.0578) (0.0635) (0.0585) Lag of log S&P 0.0315 0.0278 -0.0454 0.0310 (0.0564) (0.0557) (0.0623) (0.0567) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of tax price 1.711*** 1.678*** 1.469*** 1.812*** 0.161 0.160 0.177 0.162 |$\quad$||$-1$|SD of tax price 1.146*** 1.1279*** 0.996*** 1.223*** 0.128 0.127 0.140 0.1283 |$\quad$| Median tax price 0.128 0.137 0.144 0.162* 0.0897 0.0894 0.0984 0.090 |$\quad$||$+$|1 SD of tax price -1.0448*** -1.005*** -0.838*** -1.060*** 0.116 0.115 0.128 0.116 |$\quad$||$+$|2 SD of tax price -1.879*** -1.817*** -1.535*** -1.929*** 0.164 0.163 0.181 0.165 Observations 275,675 275,675 241,965 273,413 Number of firms 30,519 30,519 28,357 30,289 Variables . (1) . (2) . (3) . (4) . Log tax price -6.104*** -5.931*** -5.072*** -6.333*** (0.468) (0.465) (0.515) (0.471) Log tax price square -7.066*** -6.880*** -5.914*** -7.365*** (0.538) (0.534) (0.591) (0.541) Log net capital gains 0.325*** 0.342*** 0.217*** 0.303*** (0.0650) (0.0646) (0.0719) (0.0653) Log fundraising exp 0.126*** (0.00148) Log S&P 500 -0.256*** -0.235*** -0.147** -0.258*** (0.0582) (0.0578) (0.0635) (0.0585) Lag of log S&P 0.0315 0.0278 -0.0454 0.0310 (0.0564) (0.0557) (0.0623) (0.0567) Firm-level controls Y Y Y Y State-level controls Y Y Y Y Year and state-by-year effects Y Y Y Y Predicted net marginal effects of tax price |$\quad$||$-2$|SD of tax price 1.711*** 1.678*** 1.469*** 1.812*** 0.161 0.160 0.177 0.162 |$\quad$||$-1$|SD of tax price 1.146*** 1.1279*** 0.996*** 1.223*** 0.128 0.127 0.140 0.1283 |$\quad$| Median tax price 0.128 0.137 0.144 0.162* 0.0897 0.0894 0.0984 0.090 |$\quad$||$+$|1 SD of tax price -1.0448*** -1.005*** -0.838*** -1.060*** 0.116 0.115 0.128 0.116 |$\quad$||$+$|2 SD of tax price -1.879*** -1.817*** -1.535*** -1.929*** 0.164 0.163 0.181 0.165 Observations 275,675 275,675 241,965 273,413 Number of firms 30,519 30,519 28,357 30,289 Notes: Fixed-effects panel regressions using negative binomial estimation. Wild bootstrap standard errors clustered by state in (parenthesis). Nonindicator independent variables logged. Column three omits firms with zero fundraising. Column four omits firms with a matched c(3) sister organization. All columns include controls for calendar year and state-by-year trends, as well as gross assets, liabilities, program service revenue, investment income, investment losses, officer compensation, state median income, gdp, population, population share under 26, and population share over 64. Predicted Marginal Effects rows report estimated combined impact of tax price and price squared at the identified points in the distribution of tax price; SD= standard deviation. *: statistically significant at 10% level. **: statistically significant at 5% level. ***: statistically significant at 1% level. Open in new tab 6. Discussion Over some portion of the range of tax prices of giving, donors to c(4) organizations appear to respond to the charitable contribution deduction as though it applied to gifts to c(4)’s, at least with respect to their giving behavior, if not on their tax returns. The theoretical and policy implications of that fact depend in some measure on what underlies it. Most dramatically, if donors in fact are responding to the charitable contribution deduction by giving to c(4) organizations, that would upend many important assumptions about our current political system. Charitable organizations are prohibited from electioneering, and are limited to “insubstantial” lobbying, because political activities are supposed to “be conducted without public subvention” (Slee v. Commissioner, 42 F.2d 184, 185 (2d Cir. 1930)). The charitable contribution deduction provides an almost unlimited matching grant from taxpayers to donors. Uncapped subsidies are not usually optimal, even for public goods, when those goods are congestible. Lobbying expenditures cause congestion externalities as interest groups fight for legislative attention (see Galle, 2013b for a review of the supporting empirical literature). Political subsidies structured like the charitable contribution deduction would also have dramatic implications for the role of wealth in our political system (Hasen, 2008; Galle, 2013b). Deductions would tend to systematically favor wealthier interests in legislative and electoral contests, and to do so through a mechanism that leaves the source of expenditures completely opaque to the voting public (for evidence that funding sources matter to voters, see Grose and Wood, 2020). My results would not signal a complete abolition of the barrier between politics and public support but would represent incremental steps in that direction. It is unlikely that large gifts made with advice of counsel would erroneously be made to firms ineligible for deductible contributions, and similarly unlikely that single multi-million dollar gifts could be wrongfully deducted beneath the IRS’s notice. In the aggregate, though, the charitable contribution deduction might well be encouraging meaningful amounts of untraceable donations to c(4) political organizations. At the same time, crowdfunding has become a major feature of the modern political-contribution landscape (Karpf, 2012; Johnson, 2016). It is possible that on net subsidized c(4) contributions tend to aid grassroots or lower-income donors over those with greater resources. To the extent that is a desirable outcome, however, it likely would be preferable to do so through a mechanism explicitly designed to that end, such as through small-dollar matching grants (Hasen, 1996). Politics aside, it is possible that the social value of contributions to social welfare firms might be lower than for gifts to charity. Local public goods and lobbying activity may produce smaller positive externalities per average dollar than, say, international relief organizations or research institutions (e.g. because free riding incentives are smaller, see Leshem and Tabbach, 2017). If so, tax subsidies intended for charity may be less cost-effective to the extent they are claimed by c(4) organizations. Of course if this is true it may well also argue against the existing subsidy for contributions of built-in gain property to c(4) organizations. That provision appears not to have received serious scrutiny from policy makers and does not even appear in the Joint Committee on Taxation’s list of tax expenditures (for more extended criticism of the built-in gain provision, see Halperin, 2002; Colinvaux, 2013). What if donors are mistakenly giving to c(4) organizations on the belief that they will receive a subsidy, but then in fact not claiming the contributions on their tax returns? Faulhaber (2012) and Galle (2013a) argue that taxpayers who mistakenly overestimate the value of charitable contribution deductions in essence provide free public goods to the community at large. On the other side of the balance, these donors are likely not maximizing their own preferred consumption choices, and the net results for consumer welfare likely depend on how donors adjust to new information about their past mistakes (Chetty et al., 2009). The implications are more modest if most of the observed effects are caused by fundraising. I cannot clearly rule out a role for fundraising. In the smaller SOI-Other sample, controlling for fundraising leaves coefficients statistically insignificant in some portions of the tax price distribution, while in the larger Core-Other sample it does not. In Supplementary Appendix Section A.2, I report that national firms, many of them advocacy organizations, are the most apt to adjust their fundraising in response to tax policy. With donations on the left-hand side, SOI-Other specifications without these firms retain significant coefficients even when controlling for fundraising. Thus, it seems possible that fundraising explains some of the donation response to taxes at these national firms but not among the community-oriented organizations that comprise the bulk of the sample. 7. Conclusion I have presented evidence suggesting new and previously unknown linkages between the U.S. system of supports for charitable organizations and the behavior of social welfare firms. It appears that subsidies for charity simultaneously pressure and lift up c(4) organizations via a variety of channels. Policy makers may wish to consider these results in a number of contexts. Donor confusion raises significant issues. Donors who mistakenly contribute to social welfare organizations, without claiming the deduction, are providing the public with free benefits, albeit perhaps in a way that might raise consumer-protection concerns. On the other hand, subsidies of the level I find are potentially troubling for organizations that engage in partisan politics to the degree that social welfare firms now do. Regulators may also wish to consider reforms if donors are claiming the deduction, and if social welfare firm outputs are on average of lower net social benefit than charity. Possibilities include better informing donors or more closely scrutinizing their tax returns. Alternately, Congress could consider paring back the official subsidies c(4) firms now claim, such as the exclusion for donations of built-in gain property. Lastly, I find evidence that the net payoff of dollars diverted away from charity may be higher than prior literature suggests. Slemrod (1989) models the treasury efficiency of the charitable contribution deduction under the assumption that misclaimed donations are not socially productive. Chetty (2009) notes that avoidance behaviors can still have social value. Here, I find potential evidence of one such value-creating avoidance activity: while contributions to c(4) organizations may not be what Congress intended for the charitable contribution deduction, many likely do create positive externalities of some magnitude. The net treasury efficiency of tax subsidies for charity thus depends in some part on their incidental effects on social welfare firms. Supplementary material Supplementary material is available at American Law and Economics Review Journal online. Footnotes 1. One possible psychological explanation for this phenomenon is suggested by recent experimental results in which subjects cheat less when the stakes from cheating are larger (Rahwan et al., 2018). 2. Supplementary Appendix Section A.2 additionally reports regressions examining the extent to which the charitable contribution deduction affects fundraising and returns to fundraising in c(4) firms. 3. On the other hand, transfers also eliminate built-in losses. Well-advised donors do not contribute depreciated property, at least if that property would have generated a deduction when sold by the donor. 4. Gifts to charity of appreciated stock, and in some cases of other assets, can be deducted at their fair market value (see Ackerman and Auten, 2011 for more detailed discussion of the applicable rules). The FMV deduction in combination with the exclusion of tax on built-in gain can produce a negative effective tax rate. 5. The Schedule A instructions do not change in relevant respects between 1991 and 2006. 6. Slemrod (1989) reports that IRS audits in the early 1980’s detected mis-claimed charitable contributions of between 5% and 10% of total claimed contributions, with smaller fractions mostly at the top of the income distribution. This result could still be consistent with large shares of donations to c(4) organizations being mis-claimed. In the Core file that covers all c(4) returns, total annual reported “support” from all c(4) filers ranges from |${\$}$|2.9 to |${\$}$|5.1 billion (in 2007 dollars) during my sample period, 1991 to 2007. U.S. charitable contribution deductions in 2007 totaled about |${\$}$|193 billion (IRS, 2017). If Slemrod’s sample was representative, his figures would imply between |${\$}$|9.5 and |${\$}$|19 billion in inaccurately claimed donations. 7. That is, for any single donor the marginal effect on the output of a charitable organization will not be affected by the extent of their contribution to an unrelated social welfare firm. 8. In theory, a decline in the after-tax price of charitable contributions could also increase donations to competing c(4) entities through an income effect. That is, the subsidy may expand the household’s budget enough that it consumes more of all goods, including c(4) donations. But this is very implausible in our setting, where household giving averages less than 2% of household income (Giving USA 2018) and the effective subsidies to the marginal price of charity are only a fraction of that. 9. Major revisions to the ways in which donations are reported on IRS Form 990, effective in 2008, make comparisons of donation levels between the two editions somewhat problematic. 10. I did not encounter any instances in which a web site referred to another organization in order to explain that the two were unrelated and should not be confused. 11. An “exogenous itemizer” is a taxpayer who would have itemized their deductions under the standard income tax (not necessarily the AMT) even with zero charitable contribution deductions. Using this population avoids the endogeneity problem that arises if increased donations change the tax price, such as would result if additional donations would lead the donor to itemize. See Bakija & Heim (2011) for more discussion. 12. Since I control for calendar year, identification for the S&P variables depends on their impact in firms that follow a non-calendar fiscal year. 13. The tax price computation also includes adjustments for the effects of claiming a charitable contribution deduction on other determinants of taxable income. For example, because state and local taxes are federally deductible, a state contribution deduction increases federal taxes. See Bakija and Heim (2011) and Galle (2016) for more detail on these interactions. 14. Most of the tax-price variation in the sample occurs as a result of federal-law changes in 2003 and 2006. The change in 2006 was caused by the phaseout of the Pease limitation, a provision that tended to sharply reduce the value of charitable giving for some high-bracket taxpayers in states without an income tax. The 2003 changes were largely a product of complex interactions of state taxes with the federal rate structure, and in particular the AMT. See Supplementary Appendix Section A.1 for more discussion. 15. Although some state capital gains rates match the state rate on other sources of income, federal rules often interact with ordinary rates to a differing extent than capital gains rates. The coefficient of correlation between the state tax price and net capital gains rate variables is 0.80. 16. Results are robust to simply using the unadjusted NBER capital gains rate. 17. To offer a sense of the possible degree of attenuation, I implement a Monte Carlo exercise in which I use actual observed tax prices but adjust donations so that the price-elasticity of giving is |$-$|1. I then randomly replace the after-tax price for 25% of the observations with the same-year price of another state. After replacement, the estimated coefficient falls from |$-$|1 to |$-$|0.787. If I repeat the exercise replacing only 10% of the price observations, the estimated coefficient is |$-$|0.911. 18. RESET analysis also suggested that quadratic forms better fit the data. 19. |$E_{FD}$| can also be thought of as appearing in reduced form. Galle (2016) does not consider the potential separate impact of the share of firm expenditures devoted to fundraising. Donors may dislike firms with high fundraising shares. If so, |$E_{FD} $| measures the net effect of an additional dollar of fundraising on donations through two channels, one directly, and the other through fundraising share. 20. Results are similar when omitting any trends and when including both linear and quadratic state-by-year trends. In the latter case, some coefficients of tax price at below-median price lose significance. 21. An alternative explanation might be that tax price is correlated with the share of itemizers in a jurisdiction, and that itemizer share might in some way drive the outcome. This is implausible, as the coefficient of correlation between the two is only -0.147. Nonetheless, I run additional regressions in which I control for log share of itemizers. For the years (1997–2007) for which I have these data available, coefficients for the two sets of regressions are essentially identical. 22. Fundraising is endogenous to donations. Unfortunately, there are few convincing instruments for fundraising that would satisfy the exclusion restriction. To nonetheless give a sense of what IV results might look like, I instrument for fundraising using the control function approach of Woolridge (2013), and employ occupancy (as in Andreoni and Payne, 2011; Heutel, 2014), a measure of expected fundraising (as in Galle, 2016), and printing expenditures as instruments. In these regressions, the coefficient on fundraising falls in about the same range as that reported in Table 2, around 0.29, or 0.47 if I omit expected fundraising as an instrument. The coefficient on the state tax price in these specifications remains significant for most portions of the distribution of tax price, as in regressions without fundraising controls, but is larger in magnitude. 23. Since firms are likely to under-report fundraising effort or classify it in another category, firms with zero reported fundraising might still be expending resources on solicitations. Some firms might also rely on volunteer fundraising campaigns. 24. Kerlin and Reid (2010) report that in some advocacy organizations, the c(3) arm transfers large portions of its revenues to the c(4). Why, then, would firms also steer donations to the c(4), if gifts to the c(3) receive a much larger tax benefit? Recall that a c(3) cannot make grants earmarked for funding political activities if it would itself be prohibited from engaging in those same political activities. Of course, money is fungible. Still, donors who want to fund electioneering or large lobbying campaigns may prefer to give directly to the c(4) in order to ensure that the funds are spent on preferred projects. 25. I may also obtain more precise results when omitting linked firms because these political organizations are more apt to receive cross-state donations. Dropping them therefore reduces attenuation. 26. When I control for fundraising in sub-samples omitting either c(3)-linked or national firms, I obtain statistically significant and negative coefficients at the top of the distribution of tax price, unlike in regressions using the full sample. This could be due to the possibility that strategic fundraising responses to the charitable contribution deduction are mostly a feature of national advocacy organizations, a possibility I explore more fully in Supplementary Appendix Section A.2. 27. Unsurprisingly, results are very similar when alternately using cluster-robust poisson estimates. References Ackerman, Deena , and Auten Gerald. 2011 . “ Tax Expenditures for Noncash Charitable Contributions ,” 64 National Tax Journal 651 – 88 . Google Scholar Crossref Search ADS WorldCat Almunia, Miguel , Guceri Irem, Lockwood Benjamin, and Scharf Kimberley. 2020 . “ More Giving or More Givers? The Effects of Tax Incentives on Charitable Donations in the UK ,” 183 Journal of Public Economics 104 – 114 . Google Scholar Crossref Search ADS WorldCat Andreoni, James , and Payne A. Abigail. 2013 . “ Is Crowding Out Due Entirely to Fundraising? Evidence from a Panel of Charities ,” 95 Journal of Public Economics 334 – 43 . Google Scholar Crossref Search ADS WorldCat Andreoni, James , and Payne A. Abigail. 2003 . “ Do Government Grants to Private Charities Crowd Out Giving or Fundraising? ,” 93 American Economic Review 792 – 812 . Google Scholar Crossref Search ADS WorldCat Aprill, Ellen P . 2012 . “ Once and Future Gift Taxation of Transfers to Section 501(c)(4) Organizations: Current Law, Constitutional Issues, and Policy Considerations ,” 15 Journal of Legislation and Public Policy 289 – 327 . Google Scholar OpenURL Placeholder Text WorldCat Aprill, Ellen P . 2013 . “ Regulating the Political Speech of Noncharitable Exempt Organizations After Citizens United ,” 10 Election Law Journal 363 . Google Scholar Crossref Search ADS WorldCat Aprill, Ellen P . 2018 . “ Examining the Landscape of §501(c)(4) Social Welfare Organizations ,” 21 N.Y.U. Journal of Legislation and Public Policy 345 – 411 . Google Scholar OpenURL Placeholder Text WorldCat Ayres, Ian . 2017 . “ Voluntary Taxation and Beyond: The Promise of Social-Contract Voting Mechanisms ,” 19 American Law & Economics Review 1 – 48 . Google Scholar Crossref Search ADS WorldCat Cameron, A Colin. , Gelbach Jonah B., and Miller, Douglas L. 2008 . “ Bootstrap-Based Improvements for Inference with Clustered Errors ,” 90 Review of Economics and Statistics 414 – 427 . Google Scholar Crossref Search ADS WorldCat Chand, Daniel E . 2017 . “ Lobbying and Nonprofits: Money and Membership Matter—But Not for All ,” 98 Social Science Quarterly , 1297 – 1312 . Google Scholar Crossref Search ADS WorldCat Chetty, Rah . 2009 . “ Is the taxable income elasticity sufficient to calculate deadweight loss? The implications of evasion and avoidance ,” 1 American Economic Journal: Economic Policy 31 – 52 . Google Scholar Crossref Search ADS WorldCat Chetty, Raj , Looney Adam, and Kroft Korey. 2009 . “ Salience and Taxation: Theory and Evidence ,” 99 American Economic Review 1145 – 77 . Google Scholar Crossref Search ADS WorldCat Child, Curtis D. , and Gronbjerg Kirsten A.. 2007 . “ Nonprofit Advocacy Organizations: Their Characteristics and Activities ,” 88 Social Science Quarterly 259 – 81 . Google Scholar Crossref Search ADS WorldCat Colinvaux, Roger . 2013 . “ Charitable Contributions of Property: A Broken System Reimagined ,” 50 Harvard Journal on Legislation 263 – 329 . Google Scholar OpenURL Placeholder Text WorldCat Dimmery, Drew , and Peterson Andrew. 2016 . “ Shining the Light on Dark Money: Political Spending by Nonprofits ,” 2 Russell Sage Foundation Journal of the Social Sciences 51 – 68 . Google Scholar Crossref Search ADS WorldCat Dougherty, Terence . 2013 . “ Section 501(c)(4) Advocacy Organizations: Political Candidate-Related and Other Partisan Activities in Furtherance of Social Welfare ,” 36 Seattle Law Review 1377 – 1412 . Google Scholar OpenURL Placeholder Text WorldCat Duquette, Nicolas J . 2016 . “ Do Tax Incentives Affect Charitable Contributions? Evidence from Public Charities’ Reported Revenues ,” 137 Journal of Public Economics 51 – 69 . Google Scholar Crossref Search ADS WorldCat Eckel, Catherine , and Grossman Philip J.. 2017 . “ Comparing Rebate and Matching Subsidies Controlling for Donors’ Awareness: Evidence from the Field ,” 66 Journal of Behavioral and Experimental Economics 88 – 95 . Google Scholar Crossref Search ADS WorldCat Faulhaber, Lilian V . 2012 . “ The Hidden Limits of the Charitable Deduction: An Introduction to Hypersalience ,” 92 Boston University Law Review 1307 – 1348 . Google Scholar OpenURL Placeholder Text WorldCat Feenberg, Daniel , and Coutts Elisabeth. 1993 . “ An Introduction to the TAXSIM Model ,” 12 Journal of Policy Analysis and Management 189 – 94 . Google Scholar Crossref Search ADS WorldCat Feldman, Naomi E. , Goldin Jacob, and Homonoff Tatiana. 2018 . “ Raising the Stakes: Experimental Evidence on the Endogeneity of Taxpayer Mistakes .” 71 National Tax Journal 201 – 230 . Google Scholar Crossref Search ADS WorldCat Feldman, Naomi E. , Katuscak Peter, and Kawano Laura. 2016 . “ Taxpayer Confusion: Evidence from the Child Tax Credit ,” 106 American Economic Review 807 – 35 . Google Scholar Crossref Search ADS WorldCat Feldstein, Martin . 1999 . “ Tax Avoidance and the Deadweight Loss of the Income Tax ,” 81 Review of Economics and Statistics 674 – 80 . Google Scholar Crossref Search ADS WorldCat Gallagher, Kelly , and Muehlegger Eric. 2011 . “ Giving Green to Get Green: Incentives and Consumer Adoption of Hybrid Vehicle Technology ,” 61 Journal of Environmental Economics and Management 1 – 15 . Google Scholar Crossref Search ADS WorldCat Galle, Brian . 2013a . “ Carrots, Sticks, and Salience ,” 67 Tax Law Review 53 – 109 . Google Scholar OpenURL Placeholder Text WorldCat Galle, Brian . 2013b . “ Charities in Politics: A Reappraisal ,” 54 William & Mary Law Review 1561 – 1632 . Google Scholar OpenURL Placeholder Text WorldCat Galle, Brian . 2016 . “ How Do Nonprofit Firms Respond to Tax Policy? ,” 45 Public Finance Review 364 – 94 . Google Scholar Crossref Search ADS WorldCat Galston, Miriam . 2011 . “ When Statutory Regimes Collide: Will Citizens United and Wisconsin Right to Life Make Federal Tax Regulation of Campaign Activity Unconstitutional? ,” 13 University of Pennsylvania Journal of Constitutional Law 867 . Google Scholar OpenURL Placeholder Text WorldCat Goldin, Jacob , and Listokin Yair. 2013 . “ Tax Expenditure Salience ,” 16 American Law & Economics Review 144 – 76 . Google Scholar Crossref Search ADS WorldCat Grose, Christian , and Wood Abby K.. 2020 . “ Campaign Finance Transparency Affects Legislative Candidate Performance at the Polls ,” American Journal of Political Science . Google Scholar OpenURL Placeholder Text WorldCat Hackney, Philip . 2020 . “ Political Justice and Tax Policy: The Social Welfare Organization Case ,” 8 Texas A&M Law Review . Google Scholar OpenURL Placeholder Text WorldCat Halperin, Daniel . 2002 . “ A Charitable Contribution of Appreciated Property and the Realization of Built-In Gains ,” 56 Tax Law Review 1 – 38 . Google Scholar OpenURL Placeholder Text WorldCat Halperin, Daniel . 2011 . “ Is Income Tax Exemption for Charities a Subsidy? ,” 64 Tax Law Review 283 – 312 . Google Scholar OpenURL Placeholder Text WorldCat Halperin, Daniel . 2018 . “ The Tax Exemption Under Section 501(c)(4) ,” 21 N.Y.U. Journal of Legislation and Public Policy 519 – 35 . Google Scholar OpenURL Placeholder Text WorldCat Hasen, Richard L . 1996 . “ Clipping Coupons for Democracy: An Egalitarian/Public Choice Defense of Campaign Finance Vouchers ,” 84 California Law Review 1 – 58 . Google Scholar Crossref Search ADS WorldCat Hofmann, Mary Ann . 2007 . “ Tax-Motivated Expense Shifting by Tax-Exempt Associations ,” 29 Journal of the American Taxation Association 43 – 60 . Google Scholar Crossref Search ADS WorldCat Horton Smith, David . 1997 . “ The Rest of the Nonprofit Sector: Grassroots Associations as the Dark Matter Ignored in Prevailing ‘Flat Earth’ Maps of the Sector ,” 26 Nonprofit & Voluntary Sector Quarterly 114 – 31 . Google Scholar Crossref Search ADS WorldCat Hungerman, Daniel M. , and Rinz, Kevin “ Where Does Voucher Funding Go? How Large-Scale Subsidy Programs Affect Private-School Revenue, Enrollment, and Prices ,” 136 Journal of Public Economics 62 – 85 . Crossref Search ADS WorldCat Internal Revenue Service. 2006a . “ Instructions for Schedule A, Itemized Deductions .”, Available at: https://www.irs.gov/pub/irs-prior/i1040sa–2006.pdf (accessed September 17, 2020 ). Internal Revenue Service. 2006b . “ Instructions for Form 8283: Noncash Contributions ,” Available at: https://foundation.mtech.edu/give/i8283.pdf (accessed September 17, 2020 ). Internal Revenue Service. 2017 . “ Individual Income Tax Returns: Selected Income and Tax Items for Tax Years 1999 – 2016 ,” Available at: https://www.irs.gov/statistics/soi-tax-stats-historical-table-1 (accessed September 17, 2020 ). Johnson, Dennis . 2016 . Campaigning in the Twenty-First Century: Activism, Big Data, and Dark Money Second Edition . New York, NY : Routledge . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Karpf, David . 2012 . The MoveOn Effect: The Unexpected Transformation of American Political Advocacy . New York, NY : Oxford University Press . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Kerlin, Janelle A. , and Reid Elizabeth J.. 2010 . “ The Financing and Programming of Advocacy in Complex Nonprofit Structures ,” 39 Nonprofit and Voluntary Sector Quarterly 802 – 24 . Google Scholar OpenURL Placeholder Text WorldCat Kimberlin, Sara E . 2010 . “ Advocacy by Nonprofits: Roles and Practices of Core Advocacy Organizations and Direct Service Agencies ,” 9 Journal of Policy Practice 164 – 82 . Google Scholar Crossref Search ADS WorldCat Koulish, Jeremy . 2016 . “ From Camps to Campaign Funds: The History, Anatomy, and Activities of 501(c)(4) Organizations ,” Urban Institute Research Report , Available at: https://www.urban.org/research/publication/camps-campaign-funds-history-anatomy-and-activities-501c4-organizations. (accessed September 17, 2020 ). Google Scholar Leff, Benjamin Moses . 2009 . “ Sit Down and Count the Cost: A Framework for Constitutionally Enforcing the 501(c)(3) Campaign Intervention Ban ,” 28 Virginia Tax Review 673 – 731 . Google Scholar OpenURL Placeholder Text WorldCat Leshem, Shmuel , and Tabbach Avraham. 2017 . “ Solving the Volunteer’s Dilemma: The Efficiency of Rewards Versus Punishments ,” 18 American Law & Economics Review 1 – 32 . Google Scholar Crossref Search ADS WorldCat List, John , and Peysakhovich Yana. 2010 . “ Charitable Donations Are More Responsive to Stock Market Booms than Busts ,” 110 Economic Letters 166 – 69 . Google Scholar Crossref Search ADS WorldCat Maguire, Robert . 2014 . “ Dark Money Hits |${\$}$|50 Million, Most Still to Come ,” Center for Responsive Politics, August 28 . Available at: https://www .opensecrets.org/news/2014/08/dark-money-hits-50-million-most-still-to-come. Mancino, Douglas . 2005 . “ The Impact of Federal Tax Exemption Standards on Health Care Policy and Delivery ,” 15 Health Matrix 5 . Google Scholar PubMed OpenURL Placeholder Text WorldCat Mayer, Lloyd Hitoshi . 2018 . “ A (Partial) Defense of Section 501(c)(4)’s Catchall Nature ,” 21 N.Y.U. Journal of Legislation and Public Policy 439 – 80 . Google Scholar OpenURL Placeholder Text WorldCat Minkoff, Debra C . 2002 . “ The Emergence of Hybrid Organizational Forms: Combining Identity-Based Service Provision and Political Action ,” 31 Nonprofit and Voluntary Sector Quarterly 377 – 401 . Google Scholar Crossref Search ADS WorldCat Nicholson-Crotty, Jill . 2007 . “ Politics, Policy, and the Motivations for Advocacy in Nonprofit Reproductive Health and Family Planning Providers ,” 36 Nonprofit and Voluntary Sector Quarterly 5 – 21 . Google Scholar Crossref Search ADS WorldCat Nicholson-Crotty, Jill . 2009 . “ The Stages and Strategies of Advocacy Among Nonprofit Reproductive Health Providers ,” 38 Nonprofit and Voluntary Sector Quarterly 1044 – 53 . Google Scholar Crossref Search ADS WorldCat O’Neil, Cherie , Steinberg Richard S., and Thompson, G. Rodney “ Reassessing the Tax-Favored Status of the Charitable Deduction for Gifts of Appreciated Assets ,” 49 National Tax Journal 215 – 233 . OpenURL Placeholder Text WorldCat Rahwan, Zoe , Hauser Oliver P., Kochanowska Ewa, and Fasolo Barbara, “ High Stakes: A Little More Cheating, A Lot Less Charity ,” 152 Journal of Economic Behavior & Organization 276 – 95 . Crossref Search ADS WorldCat Roberts, Russell D . 1987 . “ Financing Public Goods ,” 95 Journal of Political Economy 420 – 37 . Google Scholar Crossref Search ADS WorldCat Slemrod, Joel . 1989 . “ Are estimated tax elasticities really just tax evasion elasticities? The case of charitable contributions ,” 71 Revue of Economics and Statistics 517 – 22 . Google Scholar Crossref Search ADS WorldCat Slemrod, Joel , and Yitzhaki Shlomo. 2002 . “ Tax Avoidance, Evasion, and Administration ,” in Alan J. Auerbach, and Martin Feldstein, eds. Handbook of Public Economics , Vol. 3. Google Scholar OpenURL Placeholder Text WorldCat Steinberg, Richard . 1991 . “ Does Government Spending Crowd Out Donations? Interpreting the Evidence ,” 62 Annals of Public and Cooperative Economics 591 - 617 . Google Scholar Crossref Search ADS WorldCat Steinberg, Richard . 1997 . “On the Regulation of Fund Raising,” in Burlingame D. ed., Critical Issues in Fund Raising . San Francisco, CA : Jossey-Bass , pp. 234 - 44 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Taubinsky, Dmitry , and Rees-Jones Alex. 2017 . “ Attention Variation and Welfare: Theory and Evidence from a Tax Salience Experiment ,” 85 Review of Economics & Statistics 2462 – 2496 . Google Scholar Crossref Search ADS WorldCat Tobin, Donald B . 2007 . “ Political Campaigning by Churches and Charities: Hazardous for 501(c)(3)s, Dangerous for Democracy ,” 95 Georgetown Law Journal 1313 . Google Scholar OpenURL Placeholder Text WorldCat U.S. Department of the Treasury. 2018 . “ Treasury Department and IRS Announce Significant Reform to Protect Personal Donor Information to Certain Tax-Exempt Organizations .” Available at: https://home.treasury.gov/news/press-releases/sm426 (accessed July 18, 2018 ). Yetman, Michelle H. , and Yetman, Robert 2013. “ How Does the Incentive Effect of the Charitable Deduction Vary Across Charities? ,” 88 The Accounting Review 1069 . Crossref Search ADS WorldCat © The Author 2020. Published by Oxford University Press on behalf of the American Law and Economics Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Journal

American Law and Economics ReviewOxford University Press

Published: Dec 1, 2020

References