mopdf.com

and Roberts find that nonprofits use joint costs to mitigate changes in the program ratio.Krishnan et al. (2006) find that many nonprofits report zero fundraising expenses when they infact are engaging in fundraising activities. By understating their fundraising expenses (all theway to zero), these nonprofits are increasing the proportion of expenses classified as charitable.More recent research by Keating et al. (2008) finds that nonprofits that engage in telemarketingcampaigns systematically understate their fundraising costs, which has the effect of overstatingtheir charitable costs.2.3 Donors’ Use of Low Quality Financial InformationTo summarize the existing body of research, it is relatively well established that donorsrespond to reported nonprofit financial information, rewarding nonprofits with higher ratios ofcharitable to total expenses with more donations. A growing body of research shows thatmanagers are apparently aware of donor sensitivity to this ratio and respond by manipulating theamounts of expenses reported as charitable. The question that prior research leaves largelyunanswered is the extent to which donors disentangle managerial manipulation when makingtheir donations allocation decisions. My primary research hypothesis (in null form) is that donorsdo not disentangle nonprofit financial statement management but rather naively use theinformation as reported regardless of its quality.Tinkelman (1996, 1999) examines whether the relation between donations received by anonprofit and the PSR is affected by the reliability of the PSR (as measured by whether thenonprofit reported any fundraising expenses) using a small dataset of New York regulatoryfilings. There are several key differences between Tinkelman’s work and this study. First,Tinkelman uses a limited sample of New York nonprofit organizations across a limited timeperiod, whereas this study uses a broad sample of nonprofit organizations across a long time6

period. His sample of New York nonprofits may not be generalizable. In addition, the availableof nonprofit information has changed substantially since the early 1990s (his sample time frame)and it is possible that the results from that time period do not hold in more current time periods.Second, Tinkelman does not screen on reasons that a firm could plausibly report zero fundraisingexpenses (Krishnan et al. 2006), which introduces bias into his reliability measure. Third,Tinkelman uses New York regulatory filings to measure whether a firm reports fundraisingexpenses, whereas this study uses the publicly available IRS 990. It is key to use the publiclyavailable IRS 990 to address the question of whether donors are fooled by expense misreportingbecause nonprofits have both a stronger ability and incentive to manage their IRS 990 relative totheir regulatory reports and because empirical evidence has documented differences in thereporting of amounts in their expense categories across their publically available and regulatoryreports (Krishnan and Yetman 2009). Fourth, Tinkelman uses reported fundraising to control forthe effect of fundraising effort on donations. This is a biased proxy as many firms that reportzero fundraising are actually underreporting it (Krishnan et al. 2006). This bias has the effect ofpotentially inducing his results. I improve upon the proxy for fundraising effort by estimatingfundraising effort for organizations reporting zero fundraising.2.3.1 Time-Series VariationI also examine whether the disentanglement of low quality financial reporting has improvedas the ease of obtaining nonprofit financial information has increased. Both new IRS regulationsand the development of Guidstar have made it easier and more cost effective to obtain IRS 990sover the past ten years. Before 1999, IRS regulations required nonprofits to allow the public toinspect the IRS 990 in person. Without having access to an IRS 990, donors would have to relyon published reports of nonprofit efficiency ratios, such as Wise Giving Alliance or donations7

solicitation information received from the nonprofit. Thus, donors may have used the PSR whenmaking donations decisions before 1999, but it was much more difficult and costly to obtain thedetails regarding expense reporting and therefore was more difficult to identify low qualityreporting.In June 1999, the regulations changed and required organizations to respond to a writtenrequest for their IRS 990, making it much easier for the public to obtain IRS 990 (although aformal request was still necessary and the nonprofit was permitted to charge fees for coping andmailing). Guidestar was created in 1994, and by 1996 they collected information (e.g., data onrevenues and expenses) on over 40,000 charities and launched their website(www.guidestar.org). In 1998 they expanded their database to include all 501(c)(3) publiccharities in the IRS Business Master File and added a digitized IRS 990 to charities' reports. By1999, they report IRS 990s for more than 200,000 public charities. This leads to my secondresearch hypothesis (in alternative form), which is that donors’ ability to disentangle nonprofitfinancial reporting management improved in the time periods in which the ease of obtainingfinancial reports increased (i.e., 1996, 1998, 1999).3. Empirical Model and Variable DefinitionsTo test my hypotheses, I build on prior empirical economic models on the determinants ofdonations (Weisbrod and Dominguez 1986, Okten and Weisbrod 2000). The model is as follows:Private Donationsi,t β0 β1 PSRi,t-1 β2 Zero Fundraising Indicatori,t-1 β3 PSRi,t * Zero Fundraising Indicatori,t-1 β4 Fundraising Efforti,t-1 β5 Agei,t β6 Agei,t * Fundraising Efforti,t-1 β7 Government Grantsi,t-1 β8 Feeder Donationsi,t-1 β9 Sales Revenuesi,t-1 β10 Assetsi,t ε.8(1)

Consistent with prior research, the model is estimated using natural logs and expressesprivate donations as a function of lagged information (with the exception of age and assets),consistent with the idea that donors are expected to respond to information from the prior year.Private Donations (line 1a of the IRS 990) is the dollar amount of donations from individualsand corporations, as well as grants from foundations. PSR is the proportion of charitableexpenses (line 13 on the IRS 990) as a fraction of total expenses (line 17 on the IRS 990); thus β1captures donor sensitivity to an organization’s relative efficiency of operations. I expect thecoefficient on the ratio to be positive, in line with prior research, suggesting that donors rewardnonprofits that are more efficient. The primary variable of interest is the interaction of this ratiowith the reporting quality variable, Zero Fundraising Indicator, which is equal to one if thenonprofit reported zero fundraising expenses (line 15 on the IRS 990), and zero otherwise. Ifdonors at least partially see through low quality financial reports and adjust their sensitivity to theratio, the interaction term will be negative and significantly different from zero. Alternatively, ifdonors are unable to see through (or see through but don’t care about) low quality nonprofitfinancial reports, the interaction will not be statistically different from zero.There are several possible candidates for the reporting quality metric, including misreportedtelemarketing expenses (Keating et al. 2008), misreported joint costs (Jones and Roberts 2006),and misreported fundraising expenses (Krishnan et al. 2006). The reporting quality metric I useis from Krishnan et al., who find that when a nonprofit receives substantial private donations(over 10,000 in my study), yet reports zero fundraising expenses, it is highly likely that theorganization has understated its fundraising expenses and, thus, overstated its charitableexpenses.4 I favor this reporting quality metric over other possible candidates for two reasons.4The central notion of this prior analysis is that earning donations is a production function and fundraising effort is aprimary input. Although some donations can plausibly be raised with zero fundraising effort, in order to consistently9

First, it is the only metric that applies to the majority of nonprofits. Information on telemarketingexpenses as examined by Keating et al. is available for approximately one percent of allnonprofits, and the joint cost metric examined by Jones and Roberts applies to less than onepercent of all nonprofits.Second, the Krishnan et al. (2006) quality metric is arguably the easiest for a donor toobserve prior to making their donation allocation decisions. To observe misreportedtelemarketing expenses as examined by Keating et al. (2008), a donor would need to access stateregulatory filings and compare them to the organization’s IRS 990. To observe misreported jointcosts as examined by Jones and Roberts (2006), a donor would need to employ a statisticalexpectations model to determine expected joint costs and compare that figure to actual jointcosts. By contrast, the metric examined by Krishnan et al. is simply whether or not the nonprofitreports any fundraising expenses. To observe zero reported fundraising expenses, a donor needsto examine only a single line on the front page of the IRS 990 (i.e., line 15).5 If it is zero and thenonprofit reports earning substantial donations and has no other plausible reason to be reportingzero fundraising expenses (e.g., it received donations through a feeder organization), thenonprofit is likely understating its fundraising expenses and correspondingly overstating itscharitable expenses. Thus, a large benefit of using this metric is that it is relatively easy todisentangle. To the extent donors do not disentangle zero reported fundraising, it is unlikely thatthey would disentangle any of the other, arguably more complex, types of financial statementmanipulation documented in the literature. In other words, the results in this study set a lowerbound on donors’ ability to disentangle nonprofit financial statement management.earn larger amounts of donations, a nonprofit must undertake some fundraising effort, and this effort should revealitself as an expense on the IRS 990.5There are many ways to obtain an organization’s IRS 990. First, every organization must supply a copy if asked.Second, the IRS will supply a copy for a small fee if asked. Third, all IRS 990s of all organizations can be found atwww.guidestar.org, a free Internet-based service.10

Fundraising Effort is the reported dollar amount of fundraising expenses (line 15 of the IRS990) for those that report non-zero fundraising, or estimated values of fundraising for those thatreport zero fundraising. This variable serves to control for the direct effects of fundraising effort,which is that of informing potential donors of the charitable need, much as advertising informspotential buyers of a product or service (Weisbrod and Dominguez 1986, Okten and Weisbrod2000). For firms that report non-zero fundraising on their IRS 900, in my primary analyses Iestimate their fundraising effort by using the statistical method of calibration (i.e., reverseregression). I begin by using the firms that report non-zero fundraising to observe therelationship between donations and fundraising by estimating industry specific regressions offundraising on donations. Then, using the industry specific regression estimates, I estimatefundraising effort for the sample of firms that report zero fundraising. In robustness tests, Iutilize alternative methods to estimate fundraising effort, as well as estimate the model withoutcontrolling for fundraising effort to ensure the results are not sensitive to my primary method forestimating fundraising effort.Age is the age of the nonprofit in years and is intended to be a proxy for reputation capital(Weisbrod and Dominguez 1986, Okten and Weisbrod 2000). The next three variables controlfor other sources of revenues that could crowd out (or crowd in) private donations (Posnett andSandler 1989). Government Grants is the dollar amount of grants received from federal or stateagencies (line 1b on the IRS 990), and Feeder Donations is the dollar amount of donations fromfederated fundraising organizations such as the United Way (line 1b on the IRS 990), SalesRevenues is the dollar amount of revenues received from the sales of products and services (line2 of the IRS 990), which could also crowd out (or crowd in) private donations (need cite).Donors could view sales revenues favorably as a form of self-help or, unfavorably, as a11

distraction from the primary exempt mission. Assets is year-end total assets and is included tocontrol for size. All models include single digit industry and year indicator variables.6Because the analysis uses a pooled cross section, the possibility of non-independence ofobservations arises. In particular, it is very likely that an organization’s financial reportingbehavior and the amount of donations received is correlated across time, in which case ordinaryleast squares will produce downwardly biased standard errors (although the coefficient estimateswill remain unbiased). A direct way to address this issue is to use the full panel of data, but toadjust the standard errors using the method of White (1980), with an additional adjustment forrepeated firm observations (i.e., “clustered" standard errors).7 In addition, in all analyses, I screenfor influential observations using Cooks D and Welsch Distance (Belsley et al. 1980).4. Data4.1 SampleThe IRS 990 is the primary source of publicly available nonprofit financial information. Allnonprofits with revenues over 25,000 must file the IRS 990 annually. Congressional reportssuggest that the IRS 990 serves as the primary source of publicly available nonprofit financialinformation (Joint Committee on Taxation 2000). To ensure the wide dissemination of IRS 990information, the IRS Statistics of Income division sponsors the Urban Institute to collect andmake freely available IRS 990 data for virtually all nonprofits. This data can be found on the6The National Taxonomy of Exempt Entities was established by the Internal Revenue Service as a means ofcategorizing nonprofits into 26 broad categories. Information related to the nomenclature can be found eterson (2009) provides an extensive review and analysis of the various methods used to address correlationsacross time and/or firms and provides the unambiguous recommendation that if a firm effect is suspected to bepresent (i.e., there is correlation across time within firms), the standard errors should be clustered by firm. Froot(1989), Rogers (1993), and Peterson (2009) show that clustered standard errors are unbiased and produce correctlysized confidence intervals in the presence of either temporary or permanent firm effects. Furthermore, clusteredstandard errors are robust to heteroskedasticity. The fixed effects estimator is not tractable as the zero fundraisingindicator variable has little within-firm variation.12

Internet at www.guidestar.org or can be obtained in computer readable form from the NationalCenter for Charitable Statistics for a fee at www.nccs.urban.org. The IRS 990 contains typicalfinancial statements, including a statement of revenues and expenses and a balance sheet, as wellas a substantial amount of other information related to the nonprofit’s charitable purpose andactivities.8The data I use for the analysis is from the Internal Revenue Statistics of Income files. Thesample includes only public charities (and excludes private foundations) exempt under InternalRevenue Code 501(c)(3). Although there are well over 10 types of tax-exempt organizations,501(c)(3) organizations comprise the economic bulk and are differentiated from other types ofnonprofits both because they are tax exempt and because their donors receive tax deductions fortheir donations. The IRS data is available from 1982 to 2005, with the exception of 1984. SinceI need lagged values in my analysis, I begin with 1985.The original full sample from 1985 to 2005 is reduced by requiring a minimum of 10,000in private donations as nonprofits that receive donations in excess of 10,000 are likely to haveundertaken at least some fundraising effort, thus reducing possible measurement error in the zerofundraising indicator variable.9 In addition, I remove observations that have alternativeexplanations for reporting zero fundraising as described Krishnan et al. (2006). Alternativeexplanations for reporting zero fundraising involve organizations with transactions betweenrelated parties (where one organization could reimburse the other for its administrative orfundraising expenses) or organizations with fundraising as their primary purpose (as theseorganizations could classify those expenses as charitable rather than fundraising). Thus, I remove8Keating and Frumkin (2003) provide an in-depth discussion of the IRS 990.Consistent with Krishnan et al. (2006), I exclude the other two types of donations (i.e., government grants andfeeder donations such as those from the United Way) as those can frequently be raised with little consistentfundraising effort. In addition, results are not sensitive to decreasing this minimum private donations threshold downto 1 or increasing it up to 1 million.913

organizations that are members of affiliated groups as reported on lines H and J of the IRS 990,auxiliary and fundraising organizations, whose primary purpose is fundraising (denoted by twodigit NTEE codes ending in a 11, 12 and 19), and philanthropic organizations for reasons similarto fundraising organizations (denoted by NTEE code “T”). Finally, I require nonmissing valuesfor all of the regression variables. The final sample includes 96,217 usable observations.4.2 Descriptive StatisticsTable 1 provides the descriptive statistics of the analysis variables partitioned acrossFundraising. All variables, except for Age and PSR, are scaled by 1,000. Of this final sample of96,217, approximately 39 percent report zero fundraising. Of those with zero fundraising,average private donations are approximately 1.2 million. For these zero-reporters, the averagePSR is 83 percent. On average, observations that report fundraising expenses report moredonations ( 6.6 million) and a lower PSR (79 percent). This difference in PSRs is at leastpartially due to under-reported fundraising expenses for those that report zero.Table 2 presents the Pearson correlation statistics for some of the analysis variables. Asexpected, the amount of private donations received has a large economic and statisticallysignificant positive correlation with the amount of fundraising expenses reported. Because ofthis, coupled with the fact that the PSR is negatively affected by the amount of fundraising effort,it is difficult to interpret the relationship between PSR and donations in a univariate sense.However, although the correlation between private donations and the PSR is negative, it issmaller for firms with lower quality reporting as measured by the zero fundraising indicator (0.4561), consistent with my hypothesis of donors being less sensitive to the program ratio whenthe nonprofit reports zero fundraising. This is consistent with donors at least partiallydisentangling nonprofit financial statement management. However, as evidenced by the negative14

sign on the correlation between PSR and private donations, a multivariate analysis is required tocontrol for variables correlated with both PSR and private donations.5. ResultsThe multivariate empirical results are contained in tables 3 through 6. Table 3 provides thefull sample pooled analysis, whereas tables 4, 5, and 6 partition

in nonprofit financial reporting have increased regulatory attention (Strom 2004). Concerns that donors are being misled by low quality financial information has led some state governments to . rewarding nonprofits with higher ratios of charitable to total expenses w