The relationship of executive compensation with the amount of donations in large US nonprofit organizations

(1)

1

The relationship of executive compensation with the amount of donations in large US nonprofit organizations

Shirin Ramezani, 10445528

University of Amsterdam - Faculty of Economics and Business

Supervisor: Dr. T. Jochem Msc Finance, Corporate Finance August 2017

Keywords: nonprofit organization, direct donations, executive compensation, NTEE classification

Abstract

An examination of the impact of a change in executive compensation in nonprofit organizations on donations, indicates that the change in donations depend both on donor’s behavior and on the accuracy of information provided by the firm. The results of this study shows also broad differences across different types of nonprofit organization. The nature of service that a nonprofit organization provides could impact this relationship. Finally, the results show a positive impact of an increase in executive compensation on government grants.

(2)

2

Statement of Originality

This document is written by Student, Shirin Ramezani, who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text

and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

compensation system such as the level of an executive compensation is a topic of interest in corporate governance. But, most of the research done is round and about the for-profit organizations and less is done related to the nonprofit sector. The median change in Chief Executive Officer (CEO) compensation from 2010 to 2011 and 2011 to 2012 was 2.5 percent and 2.6 percent respectively, showing an upward trend (Charity Navigator, 2014). Moreover, total donations have been increasing since 2010. For example, private donations have

increased for about 18 percent, adjusted for inflation from 2009 to 2010.

Organizations aim to attract, motivate and retain competent employees. Therefore,

corporations usually motivate the talented workforce such as CEOs by means of attractive compensation systems. On the other hand, there are no shareholders or owners of nonprofits, thus, there is a lack of oversight on the motives and actions of the executives in this sector. Besides, the executives are not held accountable to the owners of the organizations. Nonprofit organizations have a charitable motive which is often ethnically driven, while, in for-profit companies the main motive is raising profit and maximizing shareholders value (Kreander et al., 2009).

According to the data provided by the Bureau of Labor Statistics (BLS) in 2012, the employment in nonprofit organizations accounts for 10.3 percent of all private sector

employment. Moreover, nonprofits contributed a share of 5.4 percent to the Gross Domestic Product (GDP) in 2012 (“US Bureau of Economic Analysis”). Given the labor intensity, the size and importance of nonprofit organizations in the US economy, this paper aims primarily to study the alignment of executive compensation and donations in large US nonprofit organizations.

Using a sample of 19,243 nonprofit firm-year observations between 2008 and 2013. I begin the analysis by examining the impact of change in executive compensation between year t-1 and t on the change in Direct Donations between year t and t+1. Direct Donations is the difference between total private contributions and government grants. I control for other factors such as nonprofit size, age, other revenue streams and different expenses ratio as efficiency proxies. To overcome reverse causality, I use the change of a year lagged independent variables. That means the change from year t-1 to the event year t which

(5)

5

consequently gives the donors the opportunity and time to consider the information provided on Form 990 before deciding to donate. The preliminary model shows a negative but not significant impact of a change in executive compensation between t-1 and t on the change of direct donation between t and t+1.

Next, I modify the model and use the natural logarithm of the direct donations as dependent variable. Still, I find no donors response to an increase in executive compensation. Following, Yetman and Yetman (2013) that uses zero fundraising expenses as a signal of low quality accounting reports, I expand my analysis by assuming that after dropping companies with zero fundraising expenses, the sample attracts more critical donors. Consequently, I find that a one percent increase in nonprofit executive compensation is associated with a 2 percent decrease in direct donations. Furthermore, I study the relation on the four largest nonprofit industries. I observe that this finding varies with the type of the nonprofits and the service they provide. The critical donors of nonprofits that belong to a more commercial industry, which provides private goods, show a positive and significant response to an increase in executive compensation. While an increase in executive compensation in charity

organizations with public services is associated negatively (but not significantly) with direct donations.

In addition to the investigation of the impact of a change in compensation on donation, I investigate the governments’ response to a change in nonprofit executive compensation. The results do not show any impact. However, after controlling for the possible impact of a one year lagged change in centered direct donations, the results show than a one percent increase in a year lagged executive compensation associates with a 5 percent increase in government grants for an organization with an average change in a year lagged direct donations.

The main purpose of this paper is to investigate the impact of a change in nonprofit executive compensation on direct contributions at the organizational level. To accomplish this task, the remainder of this paper is organized as follows. First, I review and discuss the existing literature on nonprofit organizations. Then, the hypotheses and methodology are presented. Next, the sample selection process and summary statistics of the data is described. Finally, results, conclusion and discussion follow.

(6)

6

Background and Literature Review

Prior studies regarding nonprofit organizations will be discussed in different subsections based on the main message of the literature.

Labor economics in nonprofit sector

Nonprofit employees tend to be intrinsically motivated and are characterized by the participation of voluntary workers who do their job for the greater good, while in contrast, the for-profit employees are mainly provided with financial incentives (Mirvis and Hackett, 1983; Newton, 2015). The theoretical “labor-donations” model of Preston (1989) postulates the altruistic nature of nonprofit workers’ motivation and suggests that nonprofit workers accept lower wages in exchange of a commitment to follow the mission of organizations that generates positive social externalities rather than firms that distribute monetary benefits. The term nonprofit organizations, however, should not be interpreted as organizations that cannot make profit. In fact, they may do, but they are not allowed to distribute their profits to anyone who has control over the firm (Hansmann, 1980, 1996). In other words, the key feature of nonprofit organizations is their non-distribution constraint (Hansmann, 1980) which implies that the profits in these firms should be retained and devoted to service rather than

distributing as dividends to individuals who run the firms. Moreover, according to Hansmann (1980) generating consumer and donor trust is a comparative advantage of nonprofits over for-profit organizations. According to Holmstorm and Milgrom (1991) because of the reduction in the financial incentives that is compensated by the increase in donations, the nonprofit managers remain attracted to fulfil their jobs.

Mervis and Hackett (1983) shows lower average compensation but higher mean job satisfaction for nonprofit workers in comparison with for-profit employees. Similarly, in a research review on labor economics and the nonprofit sector, Steinberg (1990) concludes that workers would supply their labor at lower salaries to socially conscious firms that are also willing to pay higher wages at the time of favorable market conditions. He also reveals that these higher pay packages could have reverse effects in attracting a different sort of people who may be less intrinsically motivated. Handy and Katz (1998) focus on a labor market

(7)

7

issue of nonprofits. They link the lower wages paid to nonprofit managers to the differences that exist between for- and non-profit organizations such as their policy of generating consumer trust and managers’ personality differences. They suggest that the lower pay packages of nonprofit managers compared to similarly qualified managers of for-profits promotes self-selection and would attract committed managers who are willing to work for the cause of a nonprofit. Preston and Sacks (2010), however, find that the average nonprofit employee receives approximately a similar pay package to a less-skilled employee in a for-profit firm.

Financial performance of nonprofits

Financial performance is a measure of financial health of a company and is critical to its long-term performance as well as its ability to follow its mission and retain stability. In case of nonprofits, their financial condition depends on many factors such as the level of

donations, stability and diversity of revenues, the management quality, and the size of nonprofit capital (Tuckman and Chang, 1991). Therefore, donors are often concerned about the present financial position of nonprofits as well as their future position in following their mission. Tuckman and Chang (1991) provided four criteria that helped them to identify the most vulnerable nonprofit organizations: limited access to equity balances, low revenue concentration, low administrative costs and low or negative operating margins. Prior

literature has explored donors’ reaction to measures of efficiency such as the ratio of program expenses to total expenses (Balsam and Harris, 2014; Weisbrod and Domingues, 1986). Harris et al (2014) provide evidence that all else equal, good governance positively affects donations.

Compensation related literature

The IRS (§501(c)(3)(d)(ii)) requires that nonprofit organizations “must not be organized or operated for the benefit of private interests…”. Furthermore, IRS specifies that the nonprofit executive compensation should be “fair and reasonable” (§ 4958). Otherwise, for an excess in executive compensation there would be penalties. As mentioned previously, according to the donation labor theory the managers and professionals are willing to accept lower wages than what they may receive in the for-profit sector because they prefer altruistic benefits to monetary rewards. This is initially donating a part of their compensation to the nonprofits (Hansmann, 1980; Preston, 1989; Rose-Ackerman, 1986). Therefore, excess compensation

(8)

8

specially in public charities that rely mainly on donations often evoke strong emotions by donors or media.

The existing literature is mainly about the wage differential between the for-profit and nonprofit sectors. The studies show that there is a relationship between the nonprofit executive compensation and the size of the organization, organization type and the level of dependence on public contributions. Frumkin and Keating (2010) find evidence that excess in CEO compensation is present where there exist “free cash flows”, which is consistent with the “non-distribution constraint” of the nonprofits. They also provide evidence that this excess in compensation level is not necessarily related to CEO performance. Kingsbergen and Tolsma (2013) uses a factorial survey design on Dutch nonprofit organizations and find a negative relationship between potential donation and overhead costs, although donors will be more tolerant and are willing still to contribute if they notice the capacities of paid staff members. In contrast, Balsma and Harris (2014) conclude that donations from large sophisticated donors are negatively associated with compensation levels. Finally, the efficiency wage hypothesis proposes that excessive compensation packages are a partial solution to the difficulty of monitoring the managers (Handy and Katz, 1998).

Donation related literature

Kingsbergen and Tolsma (2013) perceive the age of the nonprofit organization to be a proxy for its experience and reputation, thus, the older the nonprofit, the more competent to earn the trust of donors. (Kingsbergen and Tolsma, 2013; Weisbrod and Domingues,1986). Core, Guay, and Verdi (2006) document evidence that excess endowments in not-for profit firms lead to higher agency costs and consequently greater CEO pay. Additionally, Preston (1988) claims that when donor monitoring is weak, the excess donations could be used to fund perquisites. Yermack (2015) studies the role of donation restriction in financing major US art museums as an example of a nonprofit organization. He claims that “donor governance”, restricted donations as a type of corporate governance, reduces the agency problems by shifting the museum’s administration cost into programming expenses. Oster (1998) in a study of understanding the determinants of nonprofit executive compensation shows that as the percentage of income received from private donations increases, the executive

compensation decreases. Frumkin and Keating (2002) also investigates the determinants of nonprofit executive compensation and suggest that lower administrative costs and fundraising expenses are benefits of relying on a more concentrated revenue base. Finally, the only paper

(9)

9

that I know which investigates the impact of executive compensation on donations is the Balsma and Harris (2014) study. They investigate the impact of executive compensation on the growth of direct donations and show that large sophisticated donors would negatively respond to an increase in executive compensation.

Nonprofit organizations could increase executive compensation in the hope of attracting more capable and talented employees, who can consequently increase revenue of any kind, such as direct donations. Although, some scholars such as Bowman (2006) believe that spending lesson administrative costs leads to higher efficiency. Besides, while one type of donor will appreciate an organization’s building up reserves, according to Handy & Webb (2003) if an organization saves up instead of spending its surplus, donors will question the usefulness of their donations. All else equal, contributions will be negatively affected when a nonprofit’s assets increase (Marudas, 2004). I am therefore, interested to study the relation between direct donations and executive compensation.

This research contributes to the existing nonprofits literatures on executive compensation and donor response in several ways. First, I provide the first large sample evidence of the

relations between executive compensation and direct donations in nonprofit organizations after the revision of the Form 990 in 2008. Second, I demonstrate a potential mechanism that ranks each NTEE classification and as a result the corresponding organizations based on the nature of service they provide. Third, throughout the study I specifically control for

organizational and time fixed effects in a panel data set. Finally, I investigate the impact of a change in executive compensation on government funding while controlling for the change in donation.

Hypothesis and Methodology

This paper aims to investigate the relationship between donations directly received from the public, Direct Donations, and changes in the compensation of highly paid chief executive officers (CEO’s). Media and newspapers regularly yield evidence on the impact of executive compensation on donations and the performance of nonprofit organizations. For example, Goodwill Omaha, a nonprofit organization in providing jobs for people with severe

disabilities, seeks to restore community trust after disclosure of its executive compensation. The scrutiny on this charity not only led to a significant drop in donations, but also to the resignation of Frank McGree, its CEO (Associated Press State & Local, 2017). However,

(10)

10

there is hardly any literature that investigates the effect of nonprofits’ executive

compensation on nonprofits revenue streams, namely donations. Balsma and Harris (2014) is one of the few studies that show that large and sophisticated donors would contribute less as a result of an increase in the compensation of a nonprofit organization’s CEO. Donors not only take employee compensation but also financial performance and governance quality into account before contributing.

This study addresses the impact of an increase in executive compensation on Direct

Donations. CEO compensation is a large part of administrative costs and Direct Donations is a source of revenue for nonprofit organizations. I expect that an increase in the level of nonprofit executive compensation leads to a drop in the level of Direct Donations. This hypothesis is based on the assumption that critical donors make their decisions to contribute while taking into account that their donations are used effectively to support the mission of the company rather than increasing executive compensation. In a case where managers increase their own perquisites at the expense of stakeholders, I would expect that critical donors reduce their contributions to a nonprofit organization. Therefore, the first hypothesis would be formulated in null form as follows:

Hypothesis 1: All else equal, Direct Donations are negatively related to an increase in nonprofit highly-paid executives’ compensation.

I am not only interested to explain donations but also donations as a response variable to a donor’s behavior. Because, financial performance is critical for a nonprofit organization to continue its operation and carry out its mission. Therefore, well-informed donors are concerned about both the current and future operation of an organization. I assume that donors who take the time and consider the information disclosed in Form 990 react more responsively to executive compensation changes. Yetman and Yetman (2013) uses zero fundraising expenses as a signal of low quality accounting reports. Prior empirical research shows that approximately half of all nonprofit organizations that obtain donations, and roughly 25 percent of all nonprofits with more than one million dollars in public

contributions, report zero fundraising expenses (Wing et al., 2004; Yetman and Yetman, 2013). My sample consists of large US nonprofit organizations with large amounts of donations. Hence, I expect these nonprofits could not raise substantial amounts of money without spending some on fundraising. So, inconsistent reporting of fundraising expenses

(11)

11

could indeed be used as a proxy of misrepresentation of information and/or less accurate reporting on Form 990.

I use the distinction that Yetman and Yetman (2013) use between firms that report zero fundraising on the one hand and non-zero fundraising on the other, to discern what I assume to be the result of the behavior of “critical donors”. A critical donor is a donor that first evaluates the available information and bases his or her decision to donate on a more or less rigorous assessment of that information. A non-critical donor may donate based on a whim, an emotional or ad-hoc decision, or for example as a matter of habit. Donors of both types are in all the groups that donate to a particular firm, resulting in what we might call the “average donor”. Firms that do not report their income and expenses accurately (just as those that do) may attract this average donor: both critical and non-critical donors. But I assume

firms that report non-zero fundraising expenses are the ones that will be recognized by a

critical donor as the firms that report accurately and precisely. I suppose these firms will

therefore attract more critical donors and the effect of donor’s type of decision will be better visible in the population of firms with non-zero fundraising expenses.

Furthermore, the information on fundraising expenses is reported on the first page of Form 990. I assume that a critical donor before any contribution considers the information

disclosed on the Form. If a critical donor notices fundraising expenses reported as ‘zero’, the donor would not trust the remaining information and could even ignore it. However, if the reported fundraising expense is nonzero, he or she could potentially delve deeper into Form 990 and take other parts such as compensation information into account. Therefore, if the average potential donor would not respond to an increase in executive compensation, the critical donor could react more responsively by reducing his contributions. So, I form the following research question:

RQ: Do critical donors react differently compared to the average donors?

So far, the impact of the disclosure of executive compensation on Direct Donations was based on all of the firm-year observations. But, I am interested to investigate this correlation based on the mission or type of service each nonprofit organization provide. Therefore, I partition the sample according to the four largest National Taxonomy of Exempt Entities

(12)

12

(NTEE) classifications1. Fisher, Wilsker, and Young (2011) also categorize the nonprofits according to the nature of the service provided to study the impact of the type of service on the level of revenues. They classify the service provided into three groups: private, public, and mixed. Fisher et al. (2011) claim that revenues obtained through program services are lowest for firms providing public goods and highest for the ones providing private benefits. “Private” services are those specifically meant for identifiable individuals. On the other hand, “public” goods are provided to a group of individuals regardless of their direct participation in nonprofit’s program. “Mixed” goods, is defined as services provided in between public and private goods.

It is reasonable to expect that the relation between executive compensation and donations would be different as a function of a nonprofits revenue structure. Nonprofit organizations gain revenue through three main streams; program service revenue, direct donations, and government grants. To address this issue, I describe each organization within an NTEE class as either a “charity” or “commercial”. Commercial organizations are the ones providing mainly private services and charity organizations provide more public goods. This distinction is primarily based on the comparison between the weighted proportion of each NTEE class average program service revenue to the pooled average program service revenue. Next, I compare an NTEE ratio of average direct donations to total revenue with the ratios of other main revues to total revenue of the same NTEE. An NTEE class that satisfies lower weighted average proportion of program service revenues to average program service revenue of all NTEE groups as well as showing a higher reliance on donations in comparison with its other revenue streams, is considered to be charity oriented and provides more public goods. The weighted ratios are to account for the size of each NTEE class or the difference in the number of observations in each group. Therefore, I multiply the proportion of program service

revenue to total revenue by the weight of each NTEE class (number of observations in each NTEE class to the total number of firm-year observations in the pooled sample) and compare the results after controlling for the size effect.

Program service revenue clarifies how well an organization accomplishes its stated exempt mission. If the proportion of an NTEE weighted average ratio of program service revenue to the average program service revenue of the pooled sample is higher than other NTEEs, then

1_{Internal Revenue Services (IRS) uses NTEE system to classify IRC 501(c)(3) nonprofit organizations into 26}

primary categories under 10 major groups based on their mission. I use the 4 largest out of the 10 major NTEE classification for my analysis.

(13)

13

the organizations of that category are more likely to be “commercial”. In contrast, if the weighted average of program service revenue of an NTEE class to average total program service revenue of pooled NTEE classes is lower than the other groups, then the nonprofit organizations in that class are more likely to be “charitable”. To make sure that an NTEE class is indeed providing public goods, or in other words to check commercial versus charity characteristics of an NTEE class, I also take the ratio of average direct donations of each NTEE class to its average total revenues into account.

Based on the results, I rank the four NTEE classes from the most commercial to the most charity type. Consequently, I expect that donors react more responsively to an increase in executive compensation in an NTEE that provides more public goods. In other words, an increase in executive compensation in a charity NTEE that provides public goods, more likely leads to a decrease in direct donations. These arguments develop the second hypothesis that is formulated in null form.

Hypothesis 2: All else equal, Direct Donations are negatively and strongly related to an increase in a charitable nonprofit executive’s compensation compared to commercial nonprofit organizations.

Lastly, I want to study the impact of a change in executive compensation and donors’ behavior on government grants. One could claim that governments often place a higher scrutiny on the information provided in Form 990 than critical donors. Moreover, extracting information for governments is less costly due to their larger economical scale. On the other hand, a higher executive compensation could be interpreted as indicating the presence of a more capable manager who could submit more requests for government funding. Besides a change in executive compensation, government grants could change due to changes in contributions. By the late 1960s, following the resource dependency theory, the character of nonprofits started changing significantly by replacing lost government grants and donations with earned and commercial revenues (Hansmann, 1980; Kerlin and Pollak, 2010). This suggests that Direct Donations declined by changes in the commercial composition of the organizations. However, Kerlin and Pollak (2010) claim that despite a large increase in commercial revenue between 1982 to 2002, there is little evidence that the increase was correlated with a decrease in government grants and private contributions.

(14)

14

I begin my analysis with a simple model, where I study the impact of a change in executive compensation and Direct Donations in the previous year on the natural logarithm of

government grants. Consequently, to make sure that the impact of executive compensation on government grants is not driven by the resource dependency theory or any change in donors behavior, I include the interaction term of the change in executive compensation and direct donations as a control variable in an adjusted model. In the adjusted model, I use the centered values both for the main variable of interest and for the interaction term. There are three reasons to use centered variables. First, both the change in executive compensation and Direct Donations between the year t-1 and t are continuous variables. Second, there are only 240 firm-year observations where the change of Direct Donations in the previous year is equal to zero, hence, the choice of using a dichotomous variable for donations is not a feasible option. Finally, the interpretation of the results would be more meaningful.

Therefore, I subtract the mean values from each variable. After the variables are centered, the main effect of an increase in executive compensation on government grants would be the effect of the previous year change in executive compensation on a nonprofit that has an average change in Direct Donations a year before disclosure of the Form.

Finally, due to the inclusion of centered variables and their interaction term, I assume that an increase in executive compensation on a nonprofit organization that has an average change in Direct Donations during previous year, was either meant to motivate the existing CEO or was intended to attract a more capable manager in submitting requests for government funds. These arguments form my third hypothesis.

Hypothesis 3: All else equal, the government grants today are positively related to an increase in nonprofit executive compensation of an organization with an average change in Direct Donations since previous year.

(15)

15

Empirical specification

Research model

Following the line of research studies that test the effect of the change in executive compensation on Direct Donations, I start with the Balsma and Harris (2014) model. My model is different with Balsma and Harris (2014) in several respects. For all parameters on the right hand side of the equation, I use the change in a one-year event window rather than the two years that are used in the Balsma and Harris paper. Even though this is a shorter time line, it still gives the donors enough time to react to the disclosure of Form 990. The fact that the control variables (except for Age in log form) are lagged for 1 year overcomes the potential concerns for reverse causality. Furthermore, to minimize the endogenity issue, I control for the time and firm fixed effects. Finally, the dependent variable is the natural logarithm of Direct Donations and the variable of interest is the change in CEO compensation last year. Hence, parameter estimates for testing the following model are interpreted as the percentage change (not absolute) in the dependent variable today which is associated with a one percent change in the independent variables in the previous year.

Ln ( 𝐷𝑖𝑟𝑒𝑐𝑡𝐷𝑜𝑛𝑎𝑡𝑖𝑜𝑛𝑠𝑖) = 𝛽0+ 𝛽1%∆𝐸𝑥𝑒𝑐𝑢𝑡𝑖𝑣𝑒 𝐶𝑜𝑚𝑝𝑒𝑛𝑠𝑎𝑡𝑖𝑜𝑛𝑖,𝑡−1 𝑡𝑜 𝑡+ 𝛽2%∆𝑃𝑟𝑜𝑔𝑟𝑎𝑚 𝐸𝑥𝑝𝑒𝑛𝑠𝑒 𝑅𝑎𝑡𝑖𝑜𝑖,𝑡−1 𝑡𝑜 𝑡+ 𝛽3%∆𝐹𝑢𝑛𝑑𝑟𝑎𝑖𝑠𝑖𝑛𝑔 𝐸𝑥𝑝𝑒𝑛𝑠𝑒𝑠 𝑅𝑎𝑡𝑖𝑜𝑖,𝑡−1 𝑡𝑜 𝑡+ 𝛽4%∆𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡−1 𝑡𝑜 𝑡+ 𝛽5%∆𝐺𝑜𝑣𝑒𝑟𝑛𝑚𝑒𝑛𝑡 𝐺𝑟𝑎𝑛𝑡𝑠𝑖,𝑡−1 𝑡𝑜 𝑡+ 𝛽6%∆𝑃𝑟𝑜𝑔𝑟𝑎𝑚 𝑆𝑒𝑟𝑣𝑖𝑐𝑒 𝑅𝑒𝑣𝑒𝑛𝑢𝑒𝑖,𝑡−1 𝑡𝑜 𝑡+ 𝛽7%∆𝑂𝑡ℎ𝑒𝑟 𝑅𝑒𝑣𝑒𝑛𝑢𝑒𝑠𝑖,𝑡−1 𝑡𝑜 𝑡+ 𝛽8 𝐿𝑛 (𝐴𝑔𝑒) + 𝛽₉₋₁₄𝑇𝑖𝑚𝑒 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝛽 𝐹𝑖𝑟𝑚 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝑢_𝑖

In this model, 𝑖 indicates nonprofit organization and 𝑡 indicates year. Direct Donations are total contributions, gifts, grants, and other similar amounts minus government grants (contributions). Executive compensation is the highest reported compensation, CEO

compensation, for each nonprofit organization in each specific year. According to Form 990, a nonprofit organization can allocate its costs into three groups: (1) Program Service

(16)

16

2015). Therefore, my model includes the following organizational factors that could affect donations:

 Program Expense Ratio; the percentage change in net program expenses to net total expenses. Program Expenses are costs incurred which are related to providing the nonprofit organizations programs according to its defined mission. These costs generally account for the majority of the overall expenses.

 The Fundraising Expenses Ratio; the change in net fundraising expenses to net total expenses in the last year.

Both Program Expense Ratio and Fundraising Expenses Ratio are used as a measure of efficiency. Moreover, the net values used here are after the deduction of the CEO’s compensation costs to mitigate the endogeneity problem. For the change in assets, I used total assets reported at the end of the year. Moreover, other revenue streams including Government Grants, Program Service Revenue and Other Revenues are used as control variables. The age of the company which is defined by the difference between the year of formation of the organization and the event year is included in log format as a proxy for reputation. Mature nonprofits are supposed to have a better reputation and consequently higher donations. Finally, 𝑢_𝑖𝑡 is the error term.

To make sure that natural log is not measuring a trend or percentage loss of Direct Donations, I include a one year lagged of Direct Donations as a control variable in the model. To test whether Direct Donations follow a trend or just vary randomly over time (nonstationary), I run an Augmented Dicky-Fuller (ADF) test. The ADF tests the null hypothesis that whether Direct Donations has a unit root or equivalently follows a random walk. One of the key features of a random walk is that the best predictor of the value of Direct Donations in the future is its value today. In case, I cannot reject the null hypothesis, instead of using the natural log of Direct Donations today I have to use its first difference to get the best predictions and results.

In other words, if Direct Donations and the change in CEO Compensation both follow a trend (nonstationary) then they can look related even if they are actually not, which would

consequently result in “spurious regressions”. If the results show that Direct Donations is stationary (mean reverting), next, I can check whether independent variables cause the dependent variable. The basic general idea is that if past values of the independent variables

(17)

17

are significant predictors of the current value of the dependent variable, even after including the past values of the dependent variable, then the independent variables exert a causal influence on the dependent variable. To check the causality effect, I use the Granger Causality test. Lastly, I check for the Variance Inflation Factor (VIF) which shows how the inclusion of each predictor could affect the change in the coefficient variance (or

consequently the standard errors) of other variables. Preliminary model

The equation presented above is the final version of the model after some corrections. To investigate the impact of an increase in nonprofit executive compensation on main sources of revenue, I began by a different model. The main difference of my initial model with the final one, is in the definition of the dependent variable. Initially, instead of using the natural logarithm of Direct Donations I began my research by using the percentage change in Direct Donation between year t and t+1 as dependent variable. This is what I call the change of Direct Donations next year or a year after the disclosure of the executive Compensation in the Form 990. As such, each firm-year observation needed information on a year before and a year after the event year. Thus, the sample ended up with a very short time line where four consecutive years 2009, 2010, 2011, and 2012 for each nonprofit was required.

Overall, my preliminary model needed some adjustments. Therefore, I modified the model to the one presented in the sample specification, where I investigate the relation between the percentage change in executive compensation and the natural logarithm of the Direct Donations rather than its change in the next year. Using natural log takes compounding in a systematic way into account and it is symmetric in terms of gains and losses, hence, it gives a much more meaningful and robust way to see the relationship between Direct Donations and compensation. Moreover, less firm-year observations had to be dropped. Because, only the previous year changes of independent variables are required. Consequently, the available time line increases to 5 years instead of 4.

Variable information

Direct Donations is defined as the total contributions, gifts and grants minus government grants. This measure, however, cannot capture the difference between large donations and donations of small individuals. Direct Donations is supposed to be more sensitive to higher

(18)

18

executive compensations than other sources of revenue because it is mainly provided to the nonprofits through public donors. Based on existing studies, I would expect a negative relationship between an increase in executive compensation and the natural logarithm of Direct Donation. Furthermore, I also investigate the impact of a change in executive compensation on government in natural log format.

According to prior literature, organizational performance is captured by the program expense ratio. The ratio of net program expenses to net total expenses is the Program Expense Ratio which captures the efficiency with which the organization supports its mission (Newton, 2015; Harris et al., 2014; Aggarwal et al., 2012; Baber et al, 2002). This ratio is widely used as a proxy to capture the efficiency with which a nonprofit organization’s expenses are spent on charities rather than on administrative or fundraising issues. My expectation based on previous studies would be that a more efficient nonprofit organization is better able to generate donations, therefore, I would expect a positive coefficient on this variable. Fundraising expenses ratio is used as a proxy for the effort that an organization puts into attracting donations. Therefore, I would also expect a positive coefficient on this variable (Baber, Roberts and Visvanathan, 2001; Newton, 2015; Yan and Sloan, 2014). Total assets is used as a measure to control for scale effects. I would expect the larger and the stronger the organization the higher would be the donations, hence, a positive coefficient for this variable. I also control for the age of the nonprofit organization. I expect a positive coefficient for the log of age, because the older organizations are supposed to be better known to the public, have better networks and could attract more donations as a result.

Government grants, program service revenue and other revenues represent the other types of revenue in a nonprofit organization. To control for any “crowding-in” and “crowding-out” effects, I include them as control variables. In the case of “crowding-in” effects, donors perceive the other sources of revenue positively and therefore, would donate more (Yetman and Yetman, 2013). However, “crowding-out” effects would have a negative effect on the donations as donors may feel that nonprofit organizations are supported otherwise financially (Weisbrod and Dominiguez, 1986). According to Andreoni and Payne (2003) an increase in government grants reduces fundraising events and consequently result in a crowd-in effect. Tinkelman (2010) shows that the estimated effects of crowding out vary considerably across researches. Because of these two effects, prediction of the coefficients’ signs is unclear.

(19)

19

Moreover, Horne, Johnson, and van Slyke (2005) explain these effects as a result of asymmetric information between donors and nonprofit organizations.

A year lagged change of total assets is included to control for the scale effect. Kingsbergen and Tolsma (2013) claim that large scale organizations benefit from the advantages of economies of scale. I expect a positive association between size (total assets) and

contributions due to an increase in donors’ awareness of an organization. Finally, I control for the age of an organization. The age is the difference between the event year and the formation year of the nonprofit. The more mature a company, the higher the amount of donations, therefore, I predict a positive coefficient for this parameter.

The revenue of the nonprofits is correlated with its type, industry and year, hence, I

incorporate into my model the higher dimensional fixed effect regressions in order to control for different time-invariant effects. All the control variables (except for age) are measured as a percentage change over a one year lagged period (Parsons and Trussel, 2008; Petrovits et al., 2011). That is a year before the Form 990 is disclosed. Moreover, the panel dataset is clustered based on Employee Identification Number (EIN) throughout the subsequent

analyses, because the compensation levels are bound to be serially correlated across years for a given nonprofit organization.

Critical versus average donor

Besides all the information available through the Form 990, there is no clear information available on how much donors actually know, and the extent to which they use their knowledge before contribution. Horne et al. (2005) bring evidence that donors have no information about government grants before donating. The source of this lack of information could be either of these two possibilities; the information is not provided or the information is provided but nobody takes note of it. Buchheit and Parsons (2006) studies the role of

accounting information in donors’ donation process. They conclude that even though the increase in information does not associate with an increase in actual giving, but, it increases the percentage of potential donors. According to Khumawala and Gordon (1997) donors are mainly interested in the allocation of expenses, rather than complex information on financial statements. Furthermore, given that researching on an organization is costly either financially or time wise, donors usually lack information on nonprofit firms (Buchheit and Parsons,

(20)

20

2006). These arguments are less of an issue on my study due to the compulsory disclosure of Form 990 through IRS filings 2.

I am not only interested to explain donations but also donations as a response variable to a donor’s behavior. Because, financial performance is critical for a nonprofit organization to continue its operation and carry out its mission. Therefore, well-informed donors are concerned about both the current and future operation of an organization. I assume that donors who take the time and consider the information disclosed in Form 990 react more responsively to executive compensation changes. Yetman and Yetman (2013) uses zero fundraising expenses as a signal of low quality accounting reports. Prior empirical research shows that approximately half of all nonprofit organizations that obtain donations, and roughly 25 percent of all nonprofits with more than one million dollars in public

contributions, report zero fundraising expenses (Wing et al., 2004; Yetman and Yetman, 2013). My sample consists of large US nonprofit organizations with large amounts of donations. Hence, I expect these nonprofits could not raise substantial amounts of money without spending some on fundraising. So, inconsistent reporting of fundraising expenses could indeed be used as a proxy of misrepresentation of information and/or less accurate reporting on Form 990.

As mentioned in methodology, Yetman and Yetman’s (2013) uses zero fundraising expenses as a signal of low quality accounting reports. Fundraising expenses include conducting campaigns and fundraising events (§28, SFAS 117) that would consequently raise public contributions. I use this assessment to identify what I assume to be the result of the behavior of “critical donors”. A critical donor is a donor that before deciding to contribute delve deep into the Form 990 and assess rigorously the information while a non-critical donor may donate based on ad-hoc decision or as a matter of habit. Donors of both types belong to any groups that donate to a particular firm, resulting in what one could call the “average donor”. To investigate the impact of a change in executive compensation on donations, the impact could be explained in either of two ways. First, I assume firms that report non-zero

fundraising expenses are the ones that will be recognized by a critical donor as the firms that report accurately and precisely. I suppose these firms will therefore attract more critical

2_{The nonprofit organizations whose tax year began in 2008 with assets over $2.5 million were required by IRS}

to file in the revised Form 990 that was available to public. This transparency was meant to restore public confidence in the governance of the exempt organization sector, particularly charities.

(21)

21

donors. Hence, the effect of donor’s behavior or type of decision will be better visible in the population of firms with non-zero fundraising expenses. Second, exclusion of

zero-fundraising expenses gives access to firm-year observations with more reliable and accurate information. Thus, I expect the impact of a change in executive compensation to be more precise after this correction. Brooks (2004, 2006) and Steinberg (1986) also question the validity of available information. Irvin (2005) mentions that accountability of nonprofit reports maintains public trust. Therefore, I modify my sample selection and drop the observations of which the fundraising expenses are reported zero.

Sample selection and hypothesis development

This paper uses a sample of nonprofit organizations that are tax exempt under the Internal Revenue Code Section 501(c)(3) as “public charities”. Public charities compose two-thirds of all registered nonprofits defined by the Internal Revenue Code. The number of public charities grew by about 20 percent between 2003 and 2013 while in the same period all the other registered nonprofit organizations grew by just about 3 percent (Urban Nonprofit sector in Brief, 2015). NTEE classifies the exempt IRC 501(c)(3) organizations into 26 primary categories under 10 major groups based on their mission. In this study, I keep the first four largest of the 10 major classification and set the remaining nonprofits in the fifth group. The NTEE major group classification includes; (1) arts, culture, and humanities, (2) education, (3) health care, (4) human services, and (5) other types of organization to which donors can contribute tax-deductible donations.

Data for this study are based on the years 2008 to 2013 filings of the Internal Revenue Service (IRS) “Form 990”. In 2008 Form 990 – annual information return – underwent a significant revision where new sections such as “Governance, Management and Disclosure” were added to the Form. Moreover, since then nonprofit organizations with revenues in excess of $25,000 were required to disclose their Form 990 through IRS filings. A large sample of Form 990 is provided publicly and for free as Internal Revenue Service Statistics of Income (SOI) files. According to Yetman and Yetman (2013), SOI provides over 90 percent of all nonprofit revenues even though it does not provide every nonprofit for every year. The compensation information is also obtained via the IRS website and after

(22)

22

The primary dependent variable is the natural logarithm of Direct Donations in the present year. Direct Donations is the difference between total contributions and government grants. The main variable of interest is the change in CEO Compensation in the previous year. Moreover, all the other independent variables are measured as the change of one year lagged except the age of the organization that is in log format. To ensure that donors have enough time to react to the information provided on Form 990, this paper uses a one-year window. Thus, as mentioned earlier, the variable of interest and all the control variables are calculated as the percentage change in the period t-1 to t (the previous year). Therefore, the five

consecutive years 2009 to 2013 for each firm-year observations are required. The year 2009 being the first “event” year with 2008 as year t-1.

Table 1: Distribution of observations across industries and years

Panel A: Industry

distribution

NTEE Major Classification Frequency %

Arts, Culture, and Humanities 1,743 9.06%

Education 8,582 44.60% Health 2,551 13.26% Human Services 2,874 14.94% Others 3,493 18.15% Total 19,243 100.00%

Panel B: Year distribution

year Frequency % 2008 3,164 16.44% 2009 2,905 15.10% 2010 3,123 16.23% 2011 3,166 16.45% 2012 3,336 17.34% 2013 3,549 18.44% Total 19,243 100.00%

The files on the IRS website are provided in ASCII flat formats. After transferring the flat files to .dta format, there are 296,156 firm-year observations available from SOI files that satisfy the public charity coding of 501 (c) (3) and have assets of $ 50 million or more. This

(23)

23

number of observations relates to 7,912 unique nonprofit organizations. According to the raw sample with 9 NTEE classifications, “Education” is the largest industry class (33.93 percent), followed by “Health Care” (33.55 percent) and “Human Services” (13.03 percent). The smallest group (0.38 percent) is related to the “Mutual/Membership Benefit” sector.

Moreover, the raw sample is fairly evenly distributed between the sample years from 2008 to 2013.

To prepare the final sample, first, I keep only firm-year observations with the highest

executive compensation derived from Part VII of Form 990 as reportable compensation from the organization. Thus, 264,395 firm-year observations will be omitted. This large drop of firm-year observations is due to compensation packages available to different employees. Hence, there are still 7,912 unique nonprofits available. Moreover, 1,471 observations for which no or negative compensation is reported are deleted. Then, following prior literature (Balsma and Harris, 2014) to minimize the measurement error, 156 firm-year observations with zero or negative Direct Donations are also eliminated. Furthermore, 10,889 observations with zero fundraising expenses are deleted. This yields a final sample of 19,243 firm-year observations for approximately 4,408 unique nonprofit organizations. Furthermore, instead of keeping all the 9 NTEE classifications, I only keep the first 4 largest groups and merge the remaining groups into the fifth class which is called “Others”. Table 1 presents the industry and year distribution. Panel A of table 1 shows that “Education” is still the largest industry class (44.6 percent), followed by “Others” (18.15 percent) and “Human Services” (14.94 percent) and. According to panel B of table 1 the final sample is fairly evenly distributed between the sample years from 2008 to 2013. In the following statistical tests, the final step is to generate the least biased estimates with respect to exogenous unobserved factors that could impact both dependent and independent variables. As mentioned earlier the change of

independent variables one year lagged reduces reverse causality. I also account for

differences across nonprofit organizations, industries, and time by including either firm fixed effects, industry fixed effect or a combination of either of them with year fixed effects. As such, the results are expected to be less biased.

Descriptive Statistics

Table 2 provides the descriptive statistics of the variables used as dependent and independent variables in the subsequent empirical analysis. The detailed definition of all variables is

(24)

24

provided in Appendix I. The mean (median) of Direct Donations is $ 24.2 million ($ 6.13 million) and the mean (median) of the natural logarithm of Direct Donations is 15.63 (15.63). Comparing the Direct Donations across the first four major industries reveals that the highest values for Direct Donations belong to the “Health” sector (mean = $ 25.1 million, median = $ 5.21 million) and the lowest values to the “Arts, culture, and Humanities” sector (mean = $ 18.4 million, median = $ 8.86 million). The mean (median) of CEO Compensation is $ 363,451 ($ 273,491).

Table 2: Descriptive statistics

Note: This table presents the characteristics of the sample of nonprofit organizations. Descriptive statistics are computed for a maximum of 19,243 firm-year observations between 2008 and 2013. Variable definitions are provided in Appendix.

There are in total 776 (4 percent) firm-year observations with a CEO compensation of 6 digits and more, where the highest compensations fall in the “Health” sector. Moreover, about 79 percent of these large CEO compensations are associated with nonprofits in the “Education” and “Health” sector. The mean (median) of previous year change in CEO

Compensation is 11 percent (3 percent). The mean (median) of the Government Grants in the entire sample is equal to $ 12.2 million ($ 97,174). Moreover, the mean (median) of the Program Service Revenue of the whole sample is $ 61.5 million ($ 15.4 million) which is larger than both the revenues received through Direct Donations and Government Grants. The change in Program Expense Ratio and Fundraising Expenses Ratio as the main measures of organizational performance have mean (median) of 2 and 25 percent (0.1 and -2 percent) respectively.

N Mean Median Std. Deviation Minimum Maximum

Direct Donations 19,243 24,200,000 6,130,445 77,400,000 12 2,020,000,000

Log (Direct Donations) 19,243 15.63 15.63 1.69 2.48 21.42

CEO Compensation 19,243 363,451 273,491 483,443 0 18,700,000

%∆ CEO Compensation 13,956 0.11 0.03 2.03 -1.00 175.00

Government Grants 19,243 12,200,000 97,174 87,900,000 -553322 4,920,000,000

%∆ Government Grants 8,188 2.38 -0.01 163.66 -26.73 14782.81

Program Service Revenuee 19,243 61,500,000 15,400,000 214,000,000 -891785.00 7,550,000,000

%∆ Program Service Revenuee 12,362 0.30 0.04 8.33 -15.32 556.95

Program Expense Ratio 19,243 0.84 0.86 0.13 -0.57 5.05

%∆ Program Expense Ratio 14,792 0.02 0.00 0.63 -1.61 72.77

Fundraising Expenses Ratio 19,243 0.04 0.02 0.06 -0.10 1.00

%∆ Fundraising Expenses 14,747 0.25 -0.02 5.26 -44.58 388.57 Total Assets 19,243 336,000,000 111,000,000 1,510,000,000 50,000,000 72,800,000,000 %∆ Total Assets 14,835 0.07 0.05 0.15 -0.85 5.91 Other Revenuee 19,243 2,723,639 318,478 17,700,000 -651000000 755,000,000 %∆ Other Revenuee 13,646 3.35 -0.01 187.94 -3634.93 17146.59 Age 19,080 75.21 60 85.97 0 749

(25)

25

Comparisons of expenses across the four largest NTEE industry groups is not tabulated here. But, the results show that the “Health” sector has the highest average ($ 128 million) program service expenses. The “Arts, culture, and humanities” industry has the lowest average ($ 31.4 million) program service expenses. The highest average ($ 3.32 million) fundraising expenses is related to the “Health” sector and the lowest average ($1.42 million) belongs to “Human services”.

Table 3: The revenue streams comparisons mechanism within and between each NTEE class

Note: The values of revenue streams are the average dollar values within each NTEE class. Weight is equal to the ratio of each NTEE number of observations divided by the total number of observations. The "between NTEE class" for each NTEE class calculated as the relative average value of each NTEE revenue stream divided by the average value of pooled sample, multiplied by its weight. For example, the "Between NTEE class, weighted proportion of prg_rev of "Education" sector is equal to { (Weight of Education)* [ (pg_rev of Education)/(prg_rev of pooled sample) ] }. The "within NTEE class" ratios are the relative value of each NTEE class revenue stream divided by the total revenue of the same NTEE class. For example, the within NTEE class ratio of average of DD to total revenues of "Education" sector is equal to { (DD of Education)/(total revenue of Education) }.

prsg_rev: Program Service Revenue DD: Direct Donations

gov_grnt: Government Grants

Table 3 shows the revenue stream comparisons within and between each NTEE class. The table provides the mean dollar value of revenue sources in each NTEE class and the pooled sample. According to the table, I would specify whether an NTEE class is providing public services or private goods. As mentioned earlier, if the revenues obtained through program services are highest, then, the firm is providing private goods and if it is lowest, the nonprofit is providing public goods (Fisher et al., 2011). To check which NTEE class is likely

providing private goods, first, I compare the weighted portions of program service revenue of

Arts, Culture

and Humanities Education Health Human Services Pooled sample

Program Service Revenues 10,900,000 87,200,000 111,000,000 36,200,000 245,300,000 Direct Donations 18,400,000 18,700,000 25,100,000 14,400,000 76,600,000 Government Grants 6,968,573 15,100,000 14,700,000 5,934,606 42,703,179 Total Revenues 42,700,000 134,000,000 159,000,000 59,900,000 395,600,000

Number of observations 1,743 8,582 2,551 2,874 15,750

Weight 0.111 0.545 0.162 0.182

Between NTEE class, weighted proportion of prg_rev to average total prg_rev 0.005 0.194 0.073 0.027 Between NTEE class, weighted proportion of DD to average total DD 0.027 0.133 0.053 0.034 Between NTEE class, weighted proportion of gov_grnt to average total gov_grnt 0.018 0.193 0.056 0.025

Within NT EE class, ratio of average prg to total revenues 0.255 0.651 0.698 0.604 0.620

Within NTEE class, ratio of average DD to total revenues 0.431 0.140 0.158 0.240 0.194

Within NTEE class, ratio of average gov_grnt to total revenues 0.163 0.113 0.092 0.099 0.108

(26)

26

each NTEE to the program service revenue of the pooled sample. Therefore, I look at the “between NTEE class, weighted portion of program service revenues” and compare them with each other. The highest ratio (0.19) is related to “Education” and the lowest (0.01) to “Arts”. Thus, I expect that “Education” is likely providing private services and “Arts” is mainly introducing public goods. The program service revenue comparison ranks the four major industries from the most commercial to the most charity as follow: “Education”, “Health”, “Human”, and “Arts”.

This is however, a rather simple way to specify the type of service each NTEE class provides. To support my idea, in the second step, I also look at the ratio of average Direct Donations to average total revenues within each group. The NTEE classes with a higher reliance on Direct Donations compared to other sources of revenues are the ones likely providing public goods. Comparing the shares of Direct Donations to total revenues of each NTEE class with the within ratios of the same class’s other revenues present that “Arts” has the highest (0.43) and “Education” has the lowest (0.14) figures. According to second scheme of comparison, the industries will be sorted as “Education”, “Human”, “Health”, and “Arts”, where, “Education” is the most commercial sector and “Arts” is the most charity one.

The two ends of the rankings based on the comparison mechanisms above confirm each other. However, “Health” and “Human services” sectors show different results. The final prediction is that “Education” most likely consists of “commercial” organizations and “Arts” sector is most likely composed of charity organizations.

Table 4 provides the correlation between the natural logarithm of Direct Donations and the independent variables. The table shows a positive correlation (0.03) between the change in CEO compensation and the change in total assets which suggests that larger organizations would offer higher compensation to their managers and professionals. Additionally, the measure of the size (total assets) is positively correlated with all the other predictors.

(27)

27 Ta b le 4 : T h e c o rr e la ti o n ma tr ix 1 2 3 4 5 6 7 8 9 1. L o g (D ir e ct D o n at io n s) 1. 00 2. % ∆ CE O Comp e n sa ti o n 0. 03 1. 00 3. % ∆ G o ve rn me n t G ra n ts -0 .0 1 0. 00 1. 00 4. % ∆ P ro gr am Se rv ic e R e ve n u e e 0. 02 0. 01 0. 16 1. 00 5. % ∆ P ro gr am Ex p e n se R at io 0. 00 -0 .0 1 0. 00 0. 14 1. 00 6. % ∆ Fu n d ra is in g Ex p e n se s R at io -0 .0 1 0. 00 0. 00 0. 00 -0 .0 1 1. 00 7. % ∆ To ta l A ss e ts 0. 10 0. 01 0. 03 0. 03 0. 02 0. 00 1. 00 8. % ∆ Oth e r R e ve n u e e -0 .0 1 0. 00 0. 00 0. 00 0. 00 0. 00 -0 .0 3 1. 00 9. L o g (A ge ) 0. 12 0. 03 -0 .0 3 -0 .0 6 -0 .0 6 -0 .0 4 -0 .0 1 -0 .0 1 1. 00 V ar ia b le d e fi n it io n s ar e p ro vi d e d in A p p e n d ix .

(28)

28

Furthermore, the highest correlation (0.16) is related to the change in Program Service

Revenue and the change in Government Grants. The rest of the correlations are below 0.16 in absolute value. Furthermore, nonprofit age is highly correlated (0.12) with the Direct

Donations, meaning an older nonprofit could build up a reputation that can positively affect the amount of donations. Overall, the low magnitude of the correlations as well as the absence of a significant correlation in the matrix warrants no concern for multicollinearity. Consequently, I check for the Variance Inflation Factor (VIF). Even though the results are not tabulated here, the VIFs of all the independent variables are almost equal to 1 which also further supports the idea that multicollinearity is not an issue in the model.

Empirical results

Preliminary results

Table 5 shows the preliminary results of regressing the change in Direct Donations a year after the Form 990 publication on the change of executive compensation a year before. According to table 4, an increase in CEO Compensation leads to a decrease in Direct Donations in all the 5 different regressions. However, the coefficients of the variable of interest in none of the regressions is significant. The only variable that seems to have a significant effect on the change in Direct Donations is the change in Other Revenues. This impact however, is almost zero. Moreover, the change in Total Assets or the size of an organization is negatively associated with the change in Direct Donations a year after disclosure of information on Form 990. As can be seen, the adjusted 𝑅2 are fairly low. This however, is not a problem as it could be justified by the results of Kothari and Zimmerman (1995) study. They claim that in return studies in a market setting where percentage change variables are used, lower 𝑅2 statistics are more common than in the price models where level variables are used.

(29)

29

Table 4: The relation between the change in Direct Donations and the change in CEO Compensation.

Dependent variable: %∆Direct Donations𝑖,𝑡+1 (1) (2) (3) (4) (5) %∆𝐸𝑥𝑒𝑐𝑢𝑡𝑖𝑣𝑒𝐶𝑜𝑚𝑝𝑒𝑛𝑠𝑎𝑡𝑖𝑜𝑛 -0.14 -0.12 -0.13 -0.13 -0.11 (0.120) (0.107) (0.118) (0.116) (0.106) %∆𝑃𝑟𝑜𝑔𝑟𝑎𝑚 𝐸𝑥𝑝𝑒𝑛𝑠𝑒 𝑅𝑎𝑡𝑖𝑜 -0.01 -0.02 -0.03 -0.02 -0.01 (0.012) (0.017) (0.037) (0.034) (0.013) %∆𝐹𝑢𝑛𝑑𝑟𝑎𝑖𝑠𝑖𝑛𝑔 𝐸𝑥𝑝𝑒𝑛𝑠𝑒𝑠 𝑅𝑎𝑡𝑖𝑜 0.18 0.19 -1.05 -1.04 0.18 (0.134) (0.134) (0.700) (0.700) (0.138) %∆𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠 -0.92 -1.19 -0.88 -1.24* -1.34 (1.327) (1.317) (0.939) (0.675) (1.359) %∆𝐺𝑜𝑣𝑒𝑟𝑛𝑚𝑒𝑛𝑡 𝐺𝑟𝑎𝑛𝑡𝑠 -0.00 -0.00 -0.00 -0.00 -0.00 (0.002) (0.002) (0.003) (0.003) (0.002) %∆𝑃𝑟𝑜𝑔𝑟𝑎𝑚 𝑆𝑒𝑟𝑣𝑖𝑐𝑒 𝑅𝑒𝑣𝑒𝑛𝑢𝑒 -0.02 -0.02 0.00 -0.00 -0.02 (0.022) (0.023) (0.023) (0.022) (0.022) %∆𝑂𝑡ℎ𝑒𝑟 𝑅𝑒𝑣𝑒𝑛𝑢𝑒𝑠 0.00*** 0.00*** 0.00*** 0.00*** 0.00*** (0.000) (0.000) (0.000) (0.000) (0.000) 𝐿𝑜𝑔 (𝐴𝑔𝑒) -0.37* -0.37* 0.04 0.11 -0.28 (0.192) (0.195) (0.268) (0.236) (0.196) Constant 2.22** (1.048) Observations 5,403 5,403 5,207 5,207 5,403 Adjusted R-squared -0.001 -0.001 -0.002 -0.002 0.001

Firm fixed effect NO NO YES YES NO

Year fixed effect NO YES NO YES YES

Industry fixed effect

Standard errors are clustered by EIN

NO NO NO YES NO YES NO YES YES YES Note: In this table industry fixed effect is based on the 5 largest major groupings. Robust standard errors are shown in parentheses. ***, **, and * indicate significance at the 0.01, 0.05, and 0.10 levels for a two tailed test. See Appendix for variable definitions.

Adjusted Model

Overall, the preliminary model provided no conclusive results and it needed an adjustment. Therefore, I use the model presented in the model specification. I run the natural logarithm of Direct Donations on executive compensation and table 6 provides the results. The first

column shows the results of a robust OLS regression. The second column, shows the

coefficient estimates where I control for unobserved time-invariant effects and consequently, in each of the columns 3 to 5, I control for a different effect. The first, second and fifth column provide a positive and significant association between an increase in CEO Compensation and the log of Direct Donations. This is not in accordance to my first hypothesis. The third and fourth column on the other hand show a negative correlation between CEO Compensation and Direct Donations, but the coefficients are not significant. In table 6, the highest adjusted 𝑅2 can be found in columns (3) and (4). In the third column I control for time fixed effects and in the fourth column, I control for both firm and time fixed effect. I am mainly interested in analysing the impact of CEO Compensation and other predictors on Direct Donations on the organizational level. Hence, in subsequent empirical

(30)

30

analyses, I would provide the results where I only control for the entity and time differences that may affect my results.

Table 5: The relation between natural log of Direct Donations and the change in CEO compensation

Dependent variable: Ln (Direct Donations) (1) (2) (3) (4) (5) %∆𝐸𝑥𝑒𝑐𝑢𝑡𝑖𝑣𝑒𝐶𝑜𝑚𝑝𝑒𝑛𝑠𝑎𝑡𝑖𝑜𝑛 0.06* 0.06* -0.02 -0.02 0.07* (0.036) (0.037) (0.012) (0.010) (0.043) %∆𝑃𝑟𝑜𝑔𝑟𝑎𝑚 𝐸𝑥𝑝𝑒𝑛𝑠𝑒 𝑅𝑎𝑡𝑖𝑜 0.01* 0.01** -0.00 -0.00 0.02*** (0.005) (0.005) (0.003) (0.003) (0.005) %∆𝐹𝑢𝑛𝑑𝑟𝑎𝑖𝑠𝑖𝑛𝑔 𝑅𝑎𝑡𝑖𝑜 -0.03** -0.03** 0.01 0.01 -0.03* (0.014) (0.014) (0.012) (0.011) (0.015) %∆𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠 1.31*** 1.34*** 1.13*** 1.20*** 1.38*** (0.219) (0.227) (0.146) (0.158) (0.220) %∆𝐺𝑜𝑣𝑒𝑟𝑛𝑚𝑒𝑛𝑡 𝐺𝑟𝑎𝑛𝑡𝑠 -0.00*** -0.00*** 0.00*** 0.00** -0.00*** (0.000) (0.000) (0.000) (0.000) (0.000) %∆𝑃𝑟𝑜𝑔𝑟𝑎𝑚 𝑆𝑒𝑟𝑣𝑖𝑐𝑒 𝑅𝑒𝑣𝑒𝑛𝑢𝑒 0.01* 0.01* 0.01 0.01 0.01 (0.008) (0.008) (0.004) (0.004) (0.009) %∆𝑂𝑡ℎ𝑒𝑟 𝑅𝑒𝑣𝑒𝑛𝑢𝑒𝑠 -0.00 -0.00 -0.00** -0.00*** -0.00** (0.000) (0.000) (0.000) (0.000) (0.000) 𝐿𝑛 (𝐴𝑔𝑒) 0.23*** 0.23*** 0.21* 0.02 0.38*** (0.049) (0.049) (0.120) (0.059) (0.051) Constant 14.73*** (0.219) Observations 6,996 6,996 6,771 6,771 6,996 Adjusted R-squared 0.025 0.026 0.900 0.902 0.092

Firm fixed effect NO NO YES YES NO

Year fixed effect NO YES NO YES YES

Industry fixed effect

Standard errors are clustered by EIN

NO YES NO YES NO YES NO YES YES YES

Note: In this table industry fixed effect is based on the 5 largest major groupings. Robust standard errors are shown in parentheses. ***, **, and * indicate significance at the 0.01, 0.05, and 0.10 levels for a two tailed test. See Appendix for variable definitions.

Unit root test

To check whether the Direct Donations follows a trend that just varies randomly over time, I did an Augmented Dicky-Fuller (ADF) test as described in ‘Method’, the results are not tabulated here. But, according to the results I reject the null hypothesis that Direct Donations follows a random walk (unit root, nonstationary). The stationary (mean-reversion) nature of Direct Donations is an important, even necessary condition to prove that independent variables cause the dependent variable. Hence, to test whether the past values of executive compensation are significant predictors of the current value of donations exert a causal influence on donations I have to run the Granger Causality test on my panel data. I could

(31)

31

however, not run the test. The reason is that there should be at least 7 periods of time to run this test, and I only have 5 periods. Consequently, I cannot provide evidence that an increase in executive compensation causes any change in Direct Donations. Therefore, the subsequent results show the correlations rather than the causal impact.

Critical donors’ reaction to disclosure of compensation information on Form 990

In this section, I exclude the zero fundraising expenses from my sample. Next, I regress the natural logarithm of Direct Donations on the change in CEO compensation in the previous year. Table 6 shows than a one percent increase in executive compensation is associated with a 2 percent decrease in Direct Donations. This result supports my first hypothesis. The significant result after excluding the zero fundraising expenses could be explained either in terms of donors’ behaviour or based on the accuracy of information disclosed in the Form 990. First, the large companies who do not report zero fundraising expenses are also probably more precise in reporting other information in the Form. Second, a critical donor considers the information provided in the Form 990 more seriously, hence, responds negatively to an increase in CEO compensation.

Turning to control variables, a one percent increase in nonprofit size (total assets) is

associated with a 118 percent increase in donations. Finally, my results support the crowd-in effect of the government grants. Donors respond positively to an increase in government support of nonprofit organizations. Although the coefficient of the change in government grants is very low but it is positive and significant