Do Donors Discount Low Quality Accounting Information?

Transcription

1 Do Donors Discount Low Quality Accounting Information? Michelle H. Yetman* Associate Professor of Accounting University of California, Davis Robert J. Yetman** Associate Professor of Accounting University of California, Davis March 25, 2012 Abstract: Prior research finds that donors reward nonprofits that report larger program ratios with more donations, and that managers overstate these ratios, ostensibly to attract donations. We examine how donors react to inflated program ratios and consider several means by which nonprofits can inflate their ratios. We find evidence suggesting that the average donor discounts ratios that are inflated only by our simplest and obvious means, whereby nonprofits report zero fundraising expenses when fundraising costs were likely incurred. We then examine the effect of nonprofit financial data availability on the average donor s ability to unravel inflated program ratios by using the significant shift in nonprofit accounting data availability that occurred in We find that before 1997, donors did not discount inflated ratios, suggesting that the significant costs of accessing data prior to 1997 resulted in donors being misled by inflated ratios. After nonprofit accounting information became easily available in 1997, donors began to discount inflated program ratios. Finally, we examine whether the discount applied to low quality financial reports varies across donor type and find that more sophisticated donors (measured as the percent of donations that are subject to restrictions) apply incrementally larger discounts to inflated program ratios and are unable to unravel more complex means of program ratio inflation. Keywords: nonprofit organizations, program ratios, financial reporting quality, regulation, cost shifting, agency problems JEL Classification: G1 G18 G3 G38 L3 L30 L31 M4 M41 M43 M48 The authors would like to thank Erica Harris, Chris Jones; Christo Karuna; Michael Maher; Mary Michel; Carl Olson; Christine Petrovits, Dan Tinkelman; and workshop participants at the 2009 Annual, 2009 Midyear Government and Nonprofit Section, and 2009 Western Meetings of the American Accounting Association; the 2009 Conference on Financial Economics and Accounting at Rutgers University; the University of California at Davis, the University of Houston, and Santa Clara University for their valuable comments and advice. Previous titles of this paper include Economic Consequences of Expense Misreporting in Nonprofit Organizations: Are Donors Fooled? and Are Donors Misled by Low Quality Financial Reports? * Gallagher Hall; One Shields Avenue; Davis, CA 95616; phone (530) ; mhyetman@ucdavis.edu. ** Gallagher Hall; One Shields Avenue; Davis, CA 95616; phone (530) ; rjyetman@ucdavis.edu

2 I. INTRODUCTION A central question facing accounting researchers and standard setters is the extent to which users of financial reports are able to disentangle and differentially respond to low quality financial information. If stakeholders are unable to disentangle the effects of low quality accounting information, then they are potentially making resource allocation decisions they otherwise might not have made. This question has received considerable attention in the forprofit setting, although this is not the case in the nonprofit setting. 1 Prior research shows that donors are sensitive to reported program ratios (i.e., the ratio of program expenses to total expenses), giving more donations to nonprofits with higher ratios (Weisbrod and Dominguez 1986; Posnett and Sandler 1989; Tinkelman 1999; Greenlee and Brown 1999; Okten and Weisbrod 2000; Parsons 2003; Tinkelman 2004; Tinkelman and Mankaney 2007; Parsons 2007). Prior research also shows that program ratios are frequently subject to intentional manipulation as well as unintentional errors (Tinkelman 1998; Wing et al. 2004; Hager and Greenlee 2004; Roberts 2005; Khumawala et al. 2005; Jones and Roberts 2006; Krishnan et al. 2006; Keating et al. 2008). The primary purpose of our paper is to examine whether donors to charities discount low quality program ratios and to determine whether these discounts vary across the financial data disclosure environment as well as donor sophistication. The first issue we investigate is whether the average donor discounts low quality program ratios. Tinkelman (1998) and Khumawala et al. (2005) examine charity donors responses to the reporting of joint costs from combined program and fundraising activities (hereafter, "joint costs"), which can lead to inflated program ratios, and do not find that the average donor 1 The volume of accounting research on this topic in the for-profit setting is vast. Beyer et al. (2010) provides a recent literature review. 1

3 discounts potentially low quality disclosures. 2 Although an important first step, the types of joint costs examined are used by less than one-half of one percent of nonprofits, leaving open the issue of whether or not the average donor is able to disentangle low quality accounting information. In contrast, we use four measures of low quality reporting, all of which affect the reported program ratio. These include whether the nonprofit reported zero fundraising expenses when fundraising expenses were likely incurred, the extent to which the nonprofit understated its fundraising expenses, whether the nonprofit recorded a high level of abnormal accruals; and whether the nonprofit engaged in a joint cost activity. To examine whether the average donor discounts low quality program ratios, we regress donations on the program service ratio, the program service ratio interacted with each of our reporting quality measures, and control variables using a large sample of data from We find that the average donor discounts the program ratios of zero fundraising reporters by approximately 50 percent, providing the first evidence that the average donor discounts low quality program ratios. We do not find that the average donor discounts the program ratios of nonprofits that inflate their program ratios by underreporting fundraising expenses, recording opportunistic abnormal accruals, or allocating joint costs. 2 Joint costs refer to costs incurred when a nonprofit organization combines the fundraising activity with another activity, typically a program activity such as public education. The discretion involved in allocating joint costs between the two activities can have a material effect on reported program expenses. Industry observers have long suspected that nonprofit organizations manipulate allocations to disguise fund-raising costs as program costs (Khalaf 1992; Froelich and Knoepfle 1996), and prior empirical research has shown that nonprofits opportunistically allocate costs in such a way as to improve their program ratio (Jones and Roberts 2006). Tinkelman (1998) and Khumawala et al. (2005) examine charity donors responses to joint cost allocations. Tinkelman (1998) uses proprietary data from a sample of 191 New York large national nonprofit organizations with public education programs from He reclassifies all joint costs as fund-raising, and then empirically examined the relation between donations and the recalculated program service ratio. He found that the adjustment in the ratio is not related to contributions from individual donors. Khumawala et al. (2005) use an experimental design, where they provide their subjects information on joint cost information (eliminating the search costs), and find that overall donors do not appear to be concerned about joint cost allocations and accept the reported program ratios as presented. 2

4 The second issue we examine is the effect of nonprofit financial data availability on the average donor s ability to unravel low quality program ratios. A historical shift in nonprofit financial data availability occurred over the period 1997 to Prior to 1997 it was difficult (and in most cases nearly impossible) for the average donor to obtain nonprofit financial information. Starting in 1999 the Taxpayer Bill of Rights 2 (TBOR2) required nonprofits to provide copies of their financial statements to anyone upon request. At the same time, Guidestar (a nonprofit data provider) obtained copies of nonprofit financial data and placed it on the Internet with free public access. A shift of this magnitude in the accounting disclosure environment rarely takes place, providing us with a unique opportunity to determine whether disclosure provides value to donors. 3 To our knowledge, we are the first researchers to examine the effects of TBOR2 and the related Guidestar data release on donor use of accounting information. We find that prior to 1997 the average donor did not discount the program ratios of nonprofits that reported zero fundraising. Then, starting in 1999 and continuing on to the end of the sample period in 2007 we find that the average donor discounted the program ratios of nonprofits that reported zero fundraising by an average of 60 percent. We also document a sharp increase in overall donor sensitivity to program ratios. Prior to 1997 we find an average coefficient for the program ratio of 1.67, while after 1999 the average coefficient rises to 3.16 (a statistically significant difference). 3 In addition, using this shift in the accounting disclosure environment provides a strong robustness test for our first analysis as it provides us with a difference-in-differences analysis, which allows us to control for selection biases resulting from time-invariant characteristics between low and high quality reporters. For instance, low quality reporters may fundamentally differ from high quality reporters across dimensions other than reporting quality, and it is possible that donors to low quality reporters utilize the program service ratio to a smaller degree in their donations decisions relative to high quality reporters for reasons other than reporting quality adjustments. Although we know of no reasons that this might be the case, we do view the difference-in-differences experimental design as valuable to rule out this alternative explanation. 3

5 The third and final issue we examine is whether discounts vary across donor sophistication. The extent to which certain types of users are disadvantaged by the accounting disclosure process is a question of significant importance to accountants and regulators. 4 Tinkelman (1998) and Khumawala et al. (2005) have both examined this issue in the nonprofit setting, although they reach exactly opposite conclusions. Tinkelman (1998) finds that larger (more sophisticated) donors are more likely to discount joint cost disclosures relative to smaller (less sophisticated) donors, whereas Khumawala et al. (2005) finds that expert (more sophisticated) donors are less likely to discount joint cost disclosures relative to novice (less sophisticated) donors. 5 The disagreement in these prior results, along with their focus on joint costs (which apply to very few nonprofits) leaves the issue of differences across donor sophistication largely unanswered. We measure variation across donor sophistication by estimating the percentage of fund balances that are either permanently or temporarily restricted. By attaching restrictions on their donations, donors are signaling a degree of oversight on the nonprofit, which suggests that they also may be motivated to judge the quality of the program ratio. We refer to donors who attach restrictions on their donations as "sophisticated" donors, but are careful to note that "sophistication" reflects both the incentive to incur the search costs involved in judging the quality of the program ratio and the ability to judge the quality of the program ratio. We find that 4 As an example, one of the primary missions of the United States Securities and Exchange Commission is to protect investors ( 5 Tinkelman (1998) uses a proprietary data set, which allowed him to further categorize direct public support across individuals, corporations, foundations, and legacies from deceased individuals. He argues that since individuals tend to give very small amounts relative to the other three types, they have less incentive to research organizational efficiency. He finds evidence that individual donors were more likely to accept reported program ratios, despite the presence of joint activities, relative to the other three types of donors. He is unable to disentangle whether the differences are due to differential concerns about the validity of the program ratio or the differential search costs on joint cost data. Khumawala et al. (2005) use an experimental design that includes both novice (students) and expert (preparers and foundation executives) donors. Despite the existence of joint cost allocation, although both experienced and novice donors gave a larger portion of their bequest to the charity with the higher reported program expense, the experienced donors gave significantly more of the donation to the charity with the highest program ratio relative to the novice donors. 4

6 the discounts applied to low quality program ratios are increasing in our measure of donor sophistication when the signal of low quality is zero fundraising, low fundraising, or the presence of joint costs. Our results for joint costs are consistent with those in Tinkelman (1998). These results have several implications that are the first of their kind in the nonprofit setting. First, they suggest that the average donor is able to unravel only the most simplest and obvious types of low quality accounting disclosures. Our results suggest that more complex forms of low quality program ratios mislead the average donor. Second, TBOR2 and the related Guidestar disclosure enabled donors to see through and adjust for some forms of low quality disclosures (i.e., zero fundraising), providing the first evidence that these historical changes to the nonprofit financial disclosure environment had tangible benefits for financial statement users. Finally, sophisticated donors appear able to disentangle the effects of several types of low quality reporting, suggesting that they have advantages over less sophisticated donors in responding to low quality nonprofit financial information and allocating their donations efficiently. In the following section, we discuss the theory and hypotheses development. In section III we discuss the signals by which a donor could plausibly identify a low quality program ratio. Sections IV and V discuss our research methodology and our results. The final section concludes. II. THEORY AND HYPOTHESES The Role of Nonprofit Financial Information Many organizations, regardless of their ownership structure, face agency problems because writing and enforcing contracts is costly (Jensen and Meckling 1976, Fama and Jensen 1983). Although nonprofits are not owned in the traditional sense (i.e., they do not have residual claimants), they are accountable to stakeholders, including donors, lenders, customers, and 5

7 regulators. Nonprofit stakeholders, such as donors, can reduce agency losses through monitoring (Hansmann 1996). Financial reports play an important role in monitoring managerial actions as they provide a means for donors and other stakeholders to evaluate whether the nonprofit is using donations towards the charitable mission in an efficient manner. As a first-order condition, a nonprofit s objective function has been described as maximizing revenues and charitable expenses and minimizing non-charitable expenses (Hansmann 1980; Rose-Ackerman 1980, 1996). 6 In apparent recognition of this objective function, the Internal Revenue Service Form 990 (hereafter 990), the primary financial statement for all nonprofit organizations breaks down all expenses into one of three major categories: program, administrative, and fundraising. Consistent with the objective function, the relative amount of total expenses devoted to charitable purposes, the program ratio, is a common nonprofit performance benchmark, with a large amount of research showing that donors positively respond to higher ratios (Weisbrod and Dominguez 1986; Harvey and McCrohan 1988; Posnett and Sandler 1989; Callen 1994; Khumawala and Gordon 1997; Tinkelman 1999; Okten and Weisbrod 2000; Tinkelman 2004; Lagnado 2004; Andersen and Gevas 2006; Parsons 2007). 7 Because the program ratio is a product of an accounting system, it can be subject to both intentional manipulation as well as unintentional errors, both of which reduce the informational content of the ratio (Tinkelman 1998; Hager, Pollak, and Rooney 2001; Hager and Greenlee 2004; Roberts 2005; Khumawala et al. 2005; Jones and Roberts 2006; Krishnan et al. 2006; Keating et al. 2008). 6 This is an obvious simplification, as there may be times when a nonprofit appropriately increases administrative and/or fundraising expenses. However, in the long run, nonprofits seek to maximize the amount of resources expended on the charitable purpose. 7 We recognize that donors are likely to use many different financial disclosures. However, the program ratio has received the most research, and is commonly used by various charity watchdog ratings agencies as well. 6

8 Primary Research Hypothesis (H1) To examine why a donor would discount an inflated program ratio, we turn to the for-profit literature, which has well-developed analyses of stakeholder response to noisy or biased accounting information. Disclosure models in the for-profit setting consider an environment in which the firm makes a disclosure (or financial analysts make a recommendation based on firm disclosures) and profit-seeking investors react to the information they receive (see Verrecchia 2001 for a survey of this literature). When disclosures change investors beliefs about firms future profitability, investors rationally reallocate their investments across their investment set, whereby the response is a function of both the information and its precision. The more an investor believes the disclosure to be noisy or biased, the more an investor will rationally discount the disclosure (Dye and Sridhar 2007). It is important to note that this does not necessarily mean the investor will invest less in a firm with a noisy disclosure. Rather, the investor will discount the noisy signal, perhaps increasing their sensitivity to other less noisy signals with net investment in any particular firm going down, up, or staying the same. Prior research also addresses the issue of a rational manager who knows that investors might suspect a disclosure to be lower quality, in which case the manager s incentive to intentionally manipulate the disclosure is not clear. Research using costly state falsification models suggests that even in the presence of rational investors, managers will still withhold information (and incur some costs) because if they did not, investors would perceive the firm to be less valuable than it actually is (Dye 1988; Lacker and Weinberg 1989; Stein 1989; Fischer and Verrecchia 2000; Sankar and Subramanyam 2001; Dye and Sridhar 2007; Guttman et al. 2006; and Beyer et al. 2010). 7

9 Based on this body of prior research we presume that donors, who are in some sense investors of nonprofits, will rationally discount program ratios that they suspect of being noisy or biased. This highlights our definition of a low quality program ratio, which is a program ratio whose reported value differs from the true (unobservable) value due to either intentional manipulation (bias) or to unintentional error (noise). Our theoretical analysis predicts that donors will rationally discount program ratios that they suspect of being low quality, which leads to our primary research hypothesis, in the alternative: H1: Donors discount program ratios they suspect to be lower quality. Data Availability Hypothesis (H2) Unlike the for-profit setting, where financial data has been publicly available for almost 80 years, nonprofit financial data has been available for only the past 12 years. Over the three-year period from 1997 to 1999 two important events occurred that affected the availability of nonprofit financial information to the public. First, Congress passed TBOR2 in 1997, which required nonprofits to mail copies of their 990s to anyone upon written request. Prior to TBOR2 nonprofits were not required to supply copies of their financial data to the public. Second, a nonprofit data consolidator known as Guidestar obtained copies of 990s from a set of large public charities and posted them on the Internet with free public access beginning in These two events mark the first time in history that nonprofit financial information became widely available to the public at low or no cost. A shift this significant in the accounting disclosure environment rarely occurs, and provides us with a unique experimental setting. To our knowledge no prior study has empirically examined the effects of TBOR2 and the concurrent 8 From a practical perspective, both TBOR2 and the Guidestar database have phase-in effects over the period 1997 to TBOR2 was not effective until 1999, although the initial law passed in 1997 and nonprofits were urged to voluntarily follow it. Similarly, the 1997 Guidestar data release was somewhat limited, and was expanded in 1998 and again in Thus, the effects of both TBOR2 and the Guidestar databases started in 1997, and were in full force by

10 Guidestar database release, nor has any study used this historical event as a part of their analysis. A donor must have access to a nonprofit s financial information in order to identify and discount a low quality reported program ratio. Based on this we hypothesize that the average donor was not able to identify and discount low quality program ratios prior to 1997, but that TBOR2 along with Guidestar disclosures enabled the average donor to identify and respond to low quality program ratios starting in It is not clear whether the average donor would have responded to the TBOR2 phase-in period or the more limited Guidestar data releases in 1997 and 1998, leaving donor response in 1997 and 1998 to be an empirical question. Our second hypothesis, stated in the alternative, is: H2: TBOR2 and the Guidestar database release enabled donors to identify and discount low quality program ratios. Donor Sophistication Hypothesis (H3) Much as investor sophistication varies in the for-profit sector, donor sophistication varies in the nonprofit setting. We refer to donor "sophistication" as the incentive to incur the search costs involved in judging the quality of the program ratio and the ability to judge the quality of the program ratio. Donors' use of financial information in making donations will depend on the circumstances and motivation surrounding their giving (Gordon and Khumawala 1999). Although some donations from individuals are large, most are very small and are motivated by a cursory decision process in response to mass direct-mail or telemarketing campaigns (Schervish and Havens 1994; Morgan et al. 1977). Tinkelman (1998) argues that the "incremental benefit to small donors of performing research on charities would therefore be relatively low." Experimental research suggests that many individual donors do not consider financial information in making their contribution decisions (Buchheit and Parsons 2006). In contrast, 9

11 other donors, such as corporations, foundations, grantors of legacies, as well as certain types of individual donors, such as major donors, follow a more careful decision making process and could be expected to use financial data to compare potential recipients (Gordon and Khumawala 1999; Tinkelman 1998). Based on this, our third hypothesis, stated in the alternative, is: H3: Sophisticated donors attach larger discounts to low quality program ratios relative to the average donor. III. SIGNALS OF LOW QUALITY PROGRAM RATIOS In order to discount lower quality program ratios, a donor must first have some signal about the ratio s quality. We consider four signals of low quality program ratios. Zero Fundraising It is a well-documented empirical phenomenon that roughly one-half of all nonprofits that receive contributions and roughly one-forth of all nonprofits reporting over a million dollars in contributions report zero fundraising expenses (GAO 2002; Hager 2003; Hager and Greenlee 2004; Wing et al. 2004; Krishnan et al. 2006). Underreported fundraising expenses most frequently wind up reported as program expenses, thereby inflating the program ratio (Krishnan et al. 2006). Zero fundraising could be due to intentional manipulation, unintentional errors due to a lack of managerial or accounting sophistication, or truthful disclosure (Tinkelman 1999; Keating et al. 2008). However, Krishnan et al. (2006) found that nearly 60 percent of charitable nonprofits that reported zero fundraising expenses actually incurred some form of fundraising expense. Lacking the ability to observe the true amount of fundraising expense, a rational donor who observes zero reported fundraising would form an expectation over the reported program ratio, categorizing it as lower or higher quality. From a donor s perspective, the question becomes, is zero fundraising a reliable signal of low quality? 10

12 Several pieces of evidence suggests that it would be reasonable for the average donor to presume that zero fundraising is incorrect, and thus associate zero fundraising with lower quality program ratios. As highlighted in Wing et al. (2004), "the widespread reporting of zero fundraising costs has been a focus of both Congress and the media." As a result of these congressional and media concerns, the Government Accountability Office conducted a study of nonprofit reporting and found several reporting weaknesses, including finding that over one-half of all nonprofits that earn donations fail to report any fundraising expenses (GAO 2002). In addition, charity watchdogs, professional organizations, and regulators have also warned donors. For example, the chief operating officer of the Better Business Bureau s Wise Giving Alliance (the nation s largest charity watchdog agency) stated The plain truth is it costs money to raise money. All organizations raising contributions and grants will incur some level of fundraising expense. This will be true no matter what size organization is involved or how long it has been around. (Greater Washington Society of CPAs 2011). Research shows that donors tend to follow the guidance of these watchdogs (Gordon et al. 2009; Chen 2009; Sloan 2008). Also, Mr. James Greenfield, of the Association of Fundraising Professionals (the leading trade group for professional fundraisers), stated Most fundraisers would scoff at the notion of a typical charity saying it has no fundraising costs. Some of those organizations doubtless had legitimate reasons for reporting no fundraising costs. But most charities are either misinformed about the reporting requirements or are misleading the public and potential donors about the effectiveness and efficiency of their fundraising programs. (Advancing Philanthropy, September/October 2000). Finally, in direct response to the zero fundraising phenomenon, the American Institute of Certified Public Accountants (AICPA) issued Technical Practice Aid , which states in 11

13 part that it would be unusual for a nonprofit to raise substantial donations with zero fundraising effort. In summary, given prior empirical evidence and the opinions of prominent industry participants and professional organizations it seems plausible that the average donor would presume zero fundraising to be an on average signal of a low quality program ratio. Understated Fundraising A nonprofit may not go as far as reporting zero fundraising expenses when actual fundraising expenses are incurred, but still may underreport the amount spent on fundraising. For instance, Wing et al. (2004) find that nearly 80 percent of nonprofits that receive grants from foundations and the government fail to report any of those proposal writing costs as fundraising. Consistent with this, prior research finds that many nonprofits report non-zero, but nonetheless understated fundraising expenses (Krishnan et al and Keating et al. 2008). As previously discussed, these underreported fundraising expenses most frequently wind up reported as program expenses, thereby inflating the program ratio (Krishnan et al. 2006). Just as a donor who observes zero fundraising expenses presumes it to be incorrect, a donor who observes an abnormally low amount of fundraising expenses (relative to the amount of donations received) would rationally presume it to be incorrect. Abnormal Accruals In the for-profit setting it is common practice to use an abnormal accruals model as a measure of low quality reported earnings (e.g., Jones 1991; Dechow et al. 1995). Trussel (2003) produces a summary program ratio manipulation metric that employs many features of these abnormal accrual models. Using the method of Trussel (2003) we can estimate the probability 12

14 that a particular nonprofit s program ratio has been manipulated, with higher probabilities representing lower quality program ratios. Presence of Joint Costs When a nonprofit organization combines the fundraising activity with another activity, typically a program activity such as public education, they are required to allocate the joint costs between the two activities. Although joint costs apply to very few nonprofits, they can be substantial to those nonprofits that do report them, representing about a fourth of all expenses. The inherent discretion involved in allocating joint costs between fundraising and program activities can have a material effect on the program ratio. Industry observers have long suspected that nonprofit organizations manipulate allocations to disguise fund-raising costs as program costs (Khalaf 1992; Froelich and Knoepfle 1996), and prior empirical research has shown that nonprofits opportunistically allocate costs in such a way as to improve their program ratio (Jones and Roberts 2006). Lacking the ability to observe the true cost allocations and understanding the inherent managerial discretion involved in such allocations, a rational donor who observes the presence of cost allocations might categorize the program service ratios as low quality. Consistent with this, Daniel Borochoff, head of Charity Watch, a nonprofit monitor, "says that because some charities might use joint-cost allocations to give the false impression they are spending less on fundraising and more on good works, his organization removes those expenses from its data analysis" (Blum 2012). 13

15 IV. RESEARCH DESIGN Data We collect 990 data from the Internal Revenue Service Statistics of Income (SOI) files, which are made available by the National Center for Charitable Statistics (NCCS) with assistance from the Internal Revenue Service (IRS). The SOI is the largest publicly available nonprofit database and spans the years (with the exception of 1984 as the SOI Division of the IRS was not given sufficient funding in 1984 to collect the sample). The SOI data includes a sample of Internal Revenue Code 501(c)(3) organizations but excludes private foundations. The universe of 501(c)(3) organizations that file a 990 is approximately 180,000 organizations per year, while the average SOI sample includes approximately 14,000 observations. Roughly 30 percent of the SOI sample includes nonprofits with assets greater than $50 million, with the remaining 70 percent of observations chosen based on a stratified random sample, with the express intent that the sample can be used for reliable statistical analyses. 9 Given the selection method, the SOI database captures over 90 percent of all nonprofit revenues. We subject our data to several screens intended to focus the analysis on nonprofits whose donations are more likely to be influenced by reported program ratios as well as to remove observations with erroneous or unusable data. The effects of our data screens on the sample are shown in Appendix A. Our sample begins with all 221,719 observations contained in the SOI data files from 1992 to The most recent year of available data is We begin with 1992 as it provides us with eight years of data up to the 1999 disclosure change and eight years of data 9 The intent of the IRS sampling methods is to produce a sample that can produce reliable statistical results that not only capture the economic bulk of the nonprofit sector but can also produce generalizable results than can be extrapolated to the population. See and for discussions of the IRS sampling techniques and methods. There are other 990 datasets available from the NCCS such as the IRS digitized data, although the digitized data is only available from 1998 to 2003, which would prevent us from examining the 1999 disclosure change. 14

16 after. We remove observations with less than $1,000 of total expenses, program expenses, assets, and total revenues, missing age or missing National Taxonomy of Exempt Entities (NTEE) codes. 10 Next we remove observations whose donors are less likely to be responsive to the program ratio including educational and medical nonprofits, nonprofits whose donations make up less than 10 percent of their total revenues, and finally nonprofits with less than $100,000 in private donations (Gordon and Khumawala 1999; Tinkelman and Mankaney 2007). 11 As shown in Appendix A, the criterion that removed the most observations was requiring that donations be at least 10 percent of total revenues, which removed roughly 29 percent of the sample, followed by removing observations with low levels of donations (22 percent of the sample has zero donations, and an additional 2 percent of the sample has less than $100,000 in donations). Our final sample includes 34,984 observations. Some variables are not contained in the SOI databases causing us to use other NCCS databases such as the digitized data, which imposes additional data constraints. We discuss these additional data limitations in our empirical section. Empirical Models Our primary empirical model builds on a donations demand model used extensively by prior research (Weisbrod and Dominguez 1986; Posnett and Sandler 1989; Tinkelman 1998; Tinkelman 1999; Greenlee and Brown 1999; Okten and Weisbrod 2000; Tinkelman 2004; Tinkelman and Mankaney 2007). 10 The NTEE was established by the IRS as a means of categorizing nonprofits into 26 broad categories. Information related to the nomenclature can be found at 11 We remove educational and medical nonprofits as research indicates that individual donations to university and hospitals are often motivated from personal contact (Odendahl 1990; Ostrower 1994). As such, these individuals do not consider a charity s financial statements in making their contribution decisions (Gordon and Khumawala 1999). 15

17 Private Donations i,t = β 0 + β 1 Program Ratio i,t-1 + β 2 Low Quality i,t-1 * Program Ratio i,t-1 + β 3 Low Quality i,t-1 + β 4 Fundraising Expenses i,t-1 + β 5 Age i,t + β 6 Feeder Donations i,t-1 + β 7 Government Grants i,t-1 + β 8 Sales Revenues i,t-1 + β 9 Zero Officer Pay i,t-1 + β 10 Small Size i,t-1 + β 11 Unaudited i,t-1 + β 12 Cash Basis i,t-1 + δ j Single Digit NTEE Indicators i,t + γ k Year Indicators + ε i (1) Prior research using this model posits that Private Donations (which include donations from individuals, corporations and foundations as reported on line 1a of the 990) are primarily a function of the Program Ratio (program expenses divided by total expenses calculated using line 13 divided by line 17 of the 990), Fundraising Expenses (from line 15 of the 990), the reputation of the nonprofit (measured as its age), and other revenue sources. The other revenue sources, which include Feeder Donations (which include donations from consolidated fundraisers such as the United Way as reported on line 1b of the 990), Government Grants (from line 1c of the 990), and Sales Revenues (from line 2 of the 990), control for crowd out effects which occur when donors perceive a nonprofit s need for donations to be smaller when these other sources of revenues are higher. Finally, year and industry control variables are also commonly included. We add an accounting quality variable, Low Quality, and interact it with Program Ratio. We measure Low Quality in four different ways, which we will discuss later. Also, we add controls for reporting and managerial sophistication as they may be correlated with both reporting quality and donations. Zero Officer Pay is equal to one if the nonprofit reports having no paid officers (from line 25 of the 990), suggesting that its officers are volunteers, and zero otherwise. Small Size is equal to one if the nonprofit s total assets (from line 59 of the 990) are $1 million or less, 16

18 and zero otherwise. 12 Unaudited is equal to one if the nonprofit did not undergo a financial statement audit, and zero otherwise. 13 Cash Basis is equal to one if the nonprofit used the cash method of accounting as evidenced by no accruals or deferrals reported on its balance sheet on the 990, and zero otherwise. Consistent with prior research we use the natural logs of Private Donations, Fundraising Expenses, Feeder Donations, Government Grants, and Sales Revenues in the empirical analyses. Prior research also used lagged values for many of the independent variables under the assumption that donations in the current period are a function of the variables in the prior period, and we also follow this approach. 14 Primary Research Hypothesis (H1) Our first research hypothesis is that the average donor discounts program ratios they suspect to be of low quality. Our primary variable of interest is the interaction of Program Ratio and Low Quality. If donors discount program ratios they suspect of being lower quality, the coefficient on the interaction, β 2, will be statistically negative. The Program Ratio main effect, β 1, will measure the average donor s response to high quality program ratios, with the sum of β 1 + β 2 representing the average donors response to low quality program ratios. A statistically positive value on the sum of β 1 + β 2 suggests that the average donor continues to rely somewhat on lower quality program ratios, whereas a value statistically indistinguishable from zero suggests that the average donor completely ignore low quality program ratios. 12 Because we include a small size indicator as a measure of nonprofit sophistication, we do not also include a measure of total assets. However, we do include total assets in a robustness test later in the paper. 13 Audit requirements come from many sources. Nonprofits that spend more than $300,000 of government grants in any year must undergo an A-133 audit. Many states require audits as well, as do many feeder organizations. See Keating et al. (2008), Kitching (2009), and Yetman and Yetman (2011) for more details on nonprofit audit requirements. 14 We adjust the standard errors using the method of White (1980), with an additional adjustment for repeated firm observations (i.e., clustered" standard errors from Peterson 2009). In addition, in all analyses, we screen for influential observations using Cooks D, Welsch Distance, and studentized residuals (Belsley et al. 1980). 17

19 We use four measures of Low Quality. Zero Fundraising is equal to one if the nonprofit reported zero fundraising expenses from line 15 on the 990, and zero otherwise). Underreported Fundraising is the difference between reported fundraising and expected fundraising, scaled by total donations, and is calculated only for those observations reporting non zero values for fundraising. We follow the approach in Yetman and Yetman (2011) to estimate expected fundraising by using a fundraising expense prediction model. 15 Abnormal Accruals is equal to one if the estimated probability of program ratio manipulation is higher than the median value, and zero otherwise. 16 Joint Costs is equal to one if the nonprofit engaged in a joint educational and fundraising campaign as reported on Part II of the 990, and zero otherwise, which is consistent with that used in Tinkelman (1998). Joint cost data is from the NCCS digitized files, which start in 1998 and end in 2003, reducing the sample size of our analysis using Joint Costs. Our empirical measures of Low Quality will undoubtedly classify some program ratios as high quality when it in fact they are low quality, and vice versa. These errors will introduce noise into our reporting quality measures, which will inflate the standard error of their estimates and bias us against statistically significant results. Data Availability Hypothesis (H2) To examine the effects of TBOR2 and the Guidestar data release, we estimate Model 1 by year and test for differences in the discount applied to low quality program ratios across time. As discussed, we predict that prior to 1997 donors did not discount program ratios, in which case β 2 15 The estimation model is: Fundraising Expenses i,t = β 0 + β 1 Private Donations i,t+1 + β 2 Government Grants i,t + β 3 Feeder Donations i,t + β 4 Sales Revenues i,t + β 5 Age i,t + β 6 Assets i,t + ε. All variables are defined in Table We use the method from Trussel (2003) to calculate probability of accounting manipulation, which is 1/[1+e (Z) ], where Z is equal to * MARGIN * DEFEXP * GROWTH * DEPPROG * DEFREV * PROGCHG. MARGIN is equal to (total revenues total expenses) / total revenues, DEFEXP is equal to (prepaid expenses + other assets) / total assets, GROWTH is the one period percentage change in total revenues, DEPPROG is equal to depreciation / (depreciation + total fixed assets), DEFREV is equal to deferred revenues / total assets, and PROGCHG is equal to one period percentage change in the program ratio. 18

20 would not be different from zero. Starting in 1999 we predict that donors discount low quality program ratios, in which case β 2 would be negative. It is difficult to ex-ante predict whether donors will to some degree respond to low quality program ratios after 990 data becomes available (i.e., β 1 + β 2 > 0), or will ignore them (i.e., β 1 + β 2 = 0). It is also difficult to predict whether donors had sufficient 990 information, or sufficient time to digest and respond to 990 information over the TBOR2 and Guidestar phase-in period from 1997 to Donor Sophistication Hypothesis (H3) To examine whether there is a positive association between donor sophistication and the discount applied to low quality program ratios, we expand Model 1 to include a measure of donor sophistication, which we interact with Program Ratio and Low Quality * Program Ratio. We measure donor sophistication as the percentage of donations that are restricted, as donors who attach restrictions on their donations are more likely to be at least somewhat sophisticated. Restrictions can take many forms but often prevent the nonprofit from spending the corpus (i.e., the principal) and frequently restrict how or when the funds may be spent (Weisbrod and DeScioli 2010). Our variable, Restricted Donations is the ratio of restricted donations, measured as the change in temporary and permanent restricted fund balances, to Private Donations (from the sum of lines 68 and 69, scaled by line 1a of the 990). Our primary variable of interest in this analysis is the coefficient on the triple interaction (Low Quality * Program Ratio * Restricted Donations), which captures the additional discount sophisticated donors apply to low quality program ratios. The coefficient on Program Ratio * Restricted Donations captures the additional weight that sophisticated donors apply to program ratios (without considering quality). Data on restricted fund balances is from the NCCS Core files and was not picked up by the NCCS until 1998 (but does run through the end of our sample period of 2007), reducing the sample size. 19

21 V. RESULTS Descriptive Statistics Table 1 provides the descriptive statistics. Private Donations, Fundraising Expenses, Feeder Donations, Government Grants, Sales Revenues, and SFAS 124 Adjustment (to be discussed later) are all scaled by 100,000 for presentation purposes, although as previously discussed we use natural log values for the empirical analysis. Roughly 21 percent of our observations report zero fundraising. 17 Underreported Fundraising, which is only available for the 27,539 observations that reported non-zero fundraising expense, has a mean that is very close to zero by construction; recall that these represent the error from an fundraising expense model. Underreported Fundraising ranges from at the 25 th percentile to 0.04 at the 75 th percentile. The average of Abnormal Accruals is 0.49 by construction; recall that we set this equal to 0 for observations with an estimated probability below the median value. Less than three percent of our sample engages in a joint cost activity, which is higher than the one half of one percent number for the population of nonprofits because we include only larger nonprofits in our analysis, who are much more likely report joint costs. Joint costs data is limited to the period 1998 to 2003, leaving us with 13,035 observations. On average, Restricted Donations comprise roughly 35 percent of total fund balances for our sample. Data on restricted donations is only available for the period 1998 and up, limiting the sample to 23,454 observations. Roughly 28 percent of our sample reports exactly zero officer s compensation, while nine percent has total assets of less than $1 million. Nine percent of our observations do not use accrual accounting, and 32 percent are not required to undergo an audit. 17 This is lower than the approximately 50 percent in the full sample as we limit our sample to observations that receive at least $100,000 in annual donations and also receive at least 10 percent of their total revenues from donations, which significantly reduces the proportion of the sample that reports zero fundraising. 20

22 The remaining variables are very much in line with prior research that investigated our donation s demand model. Although we do not present correlation statistics, we do test for the effects of multicollinearity and find that none of the variance inflation factors (with the exception of the interaction variables) are over 4.0 (Belsley et al suggest that values over 10.0 represent high multicollinearity). Primary Research Hypothesis (H1) Our first hypothesis is that the average donor will discount program ratios they suspect of being lower quality. As shown in Table 2 we find that donors discount the program ratios of nonprofits that report zero fundraising expenses by , which represents a roughly 50 percent discount (relative to the Program Ratio main effect coefficient of 2.671). The sum of these two coefficients is different from zero at the one percent level, suggesting that the average donor does not completely ignore the program ratios of zero fundraising reporters. We find that the average donor does not discount the program ratios of nonprofits that appear to be reporting low amounts of fundraising expenses, manipulating their program ratios using abnormal accruals, or that have joint costs. One explanation for these null results is that our measures contain a large amount of measurement error sufficient to preclude statistical significance. Another possible explanation is that donors are not able or choose not to use these signals as means of identifying low quality program ratios. As previously discussed, Joint Costs are rare, and possibly so rare that the average donor simply disregards them. With respect to Underreported Fundraising and Abnormal Accruals, both methods are relatively computationally sophisticated, and perhaps beyond the skill set of the average donor. If true, then our results suggest that the average donor is misled by low quality program ratios caused by reporting low (but positive) amounts of fundraising and by using abnormal accruals to inflate 21

23 program ratios. The average donor appears only able to unravel the simplest and most obvious forms of low quality program ratios (i.e., zero fundraising). Data Availability Hypothesis (H2) Our second hypothesis is that TBOR2 and the Guidestar data release improved the average donor s ability to detect and discount low quality program ratios. We present results for the Zero Fundraising measure of Low Quality, as it is the only one for which we find significant results for our interaction variable in Table 2. In untabulated analyses we found no across time results for any of the other three measures of Low Quality. Results in Table 3, which we consider to be striking, show that prior to 1997 donors did not discount the program ratios of nonprofits that reported zero fundraising. However, donors did respond to the reported program ratio, with an average coefficient on the Program Ratio main effect of 1.67 prior to Starting in 1999 and continuing on to the end of our sample period, donors discounted the program ratios of nonprofits that reported zero fundraising expenses by an average of -1.97, consistent with our second hypothesis. During this period the coefficient on the main effect Program Ratio averaged 3.16, suggesting that donors apply an average discount of 60 percent to the program ratios of zero fundraising reporters. In all years after 1999 the sum of the coefficients for the Program Ratio main effect and the interaction of Program Ratio and Zero Fundraising is statistically larger than zero, suggesting that the average donor does not completely ignore the program ratios of zero fundraising reporters in the post 1998 period. These findings are significant for several reasons. First, they suggest that TBOR2 and the related Guidestar disclosure enabled donors to see through and adjust for some forms of low quality nonprofit financial disclosures. These historical events drastically changed the nonprofit financial disclosure landscape, and we provide the first evidence that they had a tangible benefit 22