Abstract. Following the perceived discussion on nonprofit compensation schemes, this
paper documents dimensions of donations, compensation and corporate governance in U.S. educational institutions, a sector significantly dependent on donations. After confirmation of the documented negative impact of compensation on the amount of donations, I show that nonprofits that possess five indicators of good corporate governance, are financially rewarded by donors accordingly. For these firms, the results present that the donations – compensation sensitivity is lowered towards zero. All results are most pronounced for compensation of CEOs. Although these conclusions are not supported by a difference-in-differences model, they are robust to additional control and multicollinearity issues. Dimensions of Donations, Compensation and Corporate Governance in the Nonprofit Sector: Evidence from U.S. Educational Institutions University of Amsterdam Master Thesis Amsterdam Business School MSc Finance Author: Grijzenhout, A.B. Corporate Finance Supervisor: dr. Jochem, T. July 2017
2 STATEMENT OF ORIGINALITY This document is written by Anton Berend (Teun) Grijzenhout, 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
3 TABLE OF CONTENTS I. Introduction 4 II. Literature Review and Hypotheses Development 6 A. Literature 6 B. Regulation 9 C. Hypotheses 10 III. Data 11 IV. Methodology 13 A. Total Donations and Three Compensation Measures 14 B. Total Donations, Compensation and Five Governance Measures 15 C. Difference-in-Differences Analysis 17 V. Results 19 A. Total Donations and Three Compensation Measures 20 B. Total Donations, Compensation and Aggregate Governance Measure 21 C. Decomposition of the Aggregate Measure 25 D. Difference-in-Differences Analysis 28 VI. Robustness Checks 31 VII. Concluding Remarks and Limitations 32 References 34 Appendix 37
4
I. Introduction
Nonprofit organizations are nowadays significant contributors to the USA economy, representing 5.4% of the nation’s gross domestic product in 2012 (National Council of Nonprofits, 2017). Naturally, an increase in financial importance brings forward an increase in issues associated with the sector. Misuse of charitable funds and excessive compensation schemes are common sources of worry for donors contributing to these organizations (Harris et al., 2015). Corporate governance provisions are designed to deal with these issues, and are increasingly used by nonprofits, both voluntarily as well as due to state- and federal level regulation.
This paper examines whether U.S. nonprofit organizations, hereinafter also referred to as NPOs, that maintain higher quality of corporate governance experience a rise in received donations, accordingly, and whether this governance quality impacts the sensitivity of the compensation – donations relationship. Evidence comes from IRS Form 990 filings of educational institutions over the years 2008 to 2013. NPOs must file this Form 990 in order to maintain their federal tax-exempt status, as presented by section code 501(c)(3). The specific sector of educational institutions is selected due to its significant dependency on donations within the annual revenue stream: over 20% on average over the sample years. On top of that, it has enjoyed far less exposure from investigations than, for instance, hospitals. The timespan is selected due to data availability reasons. Only as of 2008, nonprofits are obliged to include a list of governance declarations in their filings, whereas post-2013 data has not yet been made public by the Internal Revenue Service.
The question whether donors value nonprofits that are better governed more than others, matters from an economic point of view. Implications of this conduct are specifically relevant for nonprofit managers and directors, charity-rating agencies and researchers. Directors can decide whether or not to include the governance factors included in this research, especially regarding the impact they have on the sensitivity of donations to executive compensation. Compensation of executives has a controversial nature in the nonprofit sector, merely due to the trade-off between attracting talent and the wish of donors for contributing their granted money to the core mission, rather than to high compensation of executives (Balsam & Harris, 2014). This trade-off is further sharpened by the competition for talent with for-profit organizations, which are able to grant larger and more diverse compensation, such as stock options (Preston, 1989).
5
Within the literature on agency problems, corporate governance, and nonprofit donation determinants, this research is, to my knowledge, the first to empirically investigate factors explaining donations in the nonprofit educational sector. The fact that governance data is only recently available, adds to the originality of the conduct. Furthermore, it is one of the first studies to investigate the interaction effects of compensation and governance as determinants of donations to nonprofits. Finally, I suggest a new theory on the way donors amend the height of their donations. Econometric results support that following the inclusion of good governance indicators, the amount of donations rises, and the sensitivity of donations to compensation is lowered towards zero. This is most fiercely expressed for CEO compensation, but with lower statistical significance for all officer compensation, and total reported compensation as well.
I make this contribution by following four hypotheses and associated research methodologies, using three measures of compensation as explanatory variables, i.e. CEO, All Officer, and Total Reported Compensation. First, an OLS regression confirms the expected negative correlation between the compensation measures and the amount of donations. Thereafter, interaction of the compensation measures and five provisions capturing the quality of corporate governance presents guidance on the second hypothesis, which states that good corporate governance should improve the amount of donations. This statement is tested including the provisions altogether in an aggregate measure, and with the provisions separately. Both tests are in panel data regressions using industry- and year fixed effects. The results present positive effects on the intercept, and insignificant or negative results on the slope, mostly pronounced when reviewing CEO compensation. This is suggestive evidence on a theory I pose on the donation process, implying donor insensitivity to compensation when firms a well governed. This is a cautious confirmation of the second hypothesis. To address concerns of endogeneity due to reversed causality, a difference-in-differences estimator is constructed employing two state level regulatory changes. The third hypothesis stating that these changes will significantly impact donations through their impact on governance levels, is not supported by the results. A direction for future research is given here by the suggestion to employ two other, more significant changes, which were out of my available data. At last, the fourth hypothesis denotes the results to be more salient for CEO compensation, than for All Officer or Total Reported Compensation. This
6 expectation is supported by all models, and reflected by nearly all coefficients in terms of statistical significance. The remainder of this research is ordered as follows: In Section II, the literature is reviewed followed by a description of the development of the hypotheses. Next, Section III describes the data and data collection process. Section IV explains carefully the methodology this paper follows. The results produced by the empirical models are discussed in section V. Consecutively, additional robustness checks of the results are reflected in section VI. Finally, Section VII concludes with some remarks, limitations, implications and an overall conclusion.
II. Literature Review and Hypotheses Development
In this section, I commence in part A with a navigation through the existing literature regarding the topics this thesis addresses, which are depicted as agency problems in general, corporate governance and compensation in for-profit and nonprofit organizations, dimensions of determinants of donations to nonprofits, and literature on educational governance. Consecutively, since I pursue employment of a regulatory change in state level regulation, in part B, I elaborate on the nonprofit legal framework, both at the federal as well as at the state level. Finally, I derive 4 hypotheses from the related literature and the legal framework in part C.
A. Literature
To start with, the base of the framework consists of work of Jensen and Meckling (1976) discussing agency problems resulting from the separation of ownership and control, which exist in both for- and non-profit organizations regardless of the ownership structure. Corporate governance mainly arose to deal with these kinds of problems and is in general expected to improve firm performance and prevent organizations from overpaying their executives, i.e. setting the right incentives (Core, Holthausen & Larcker, 1999). Among others Bebchuk, Cohen and Ferrell (2009) performed an investigation of which factors capturing good corporate governance matter the most to investors, based on the GIM index which include 24 of such factors. More recently, Conyon (2014) argues that general determinants of executive compensation in for-profit firms include firm size and board structure, but also firm
7
performance, the use of compensation consultants and the recently introduced phenomenon of say-on-pay.
Whereas board structure and firm size can be translated as well to nonprofit organizations, the latter determinants are alien to these entities due to the absence of shareholders. There are, nevertheless, other factors documented that are specifically important to NPOs. An early work mentioning nonprofit governance is a Fama and Jensen (1983) paper, where they point out that nonprofits are more likely to survive if they are characterized by a greater potential supply of donations and lower cost of separating decision management from decision control. Steinberg (1986) concludes that NPOs mainly objectivize fundraising, and that educational institutions are service maximizers. Work of Preston (1989) contributes to the comparison between nonprofit and for-profit compensation, stating that executives supply labor to NPOs at lower than market wages in return for the opportunity to provide positive social externalities, also known as “labor donations”. Analogous to executive compensation in for-profits, important determinants of CEO compensation in non-profits are again performance and size (among others Carroll et al., 2005 and Mesch and Rooney, 2008), gender, age, ideology and the composition of the revenues (Oster, 1998).
Naturally, the follow-up question arises on how the compensation process is governed in nonprofits, and whether this process deviates from the parallel for-profit determination. The national council of nonprofits (2017) presents a clear view on how compensation should be determined. From their fiduciary duty, the board of directors is responsible for setting pay levels based on the “rebuttable presumption”. This presumption is a three-step process, as described below. From table I, it appears that 89.9% of my sample used such a process for establishing their CEO compensation.
1. The board should arrange an independent body for establishing comparative reasonableness. This body is often a compensation committee.
2. The body is expected to review compensation scheme data from similar nonprofit employers, comparable in size and mission.
3. The body is expected to document the process, including who was involved in the conduct and the minutes the board of directors spent on the decision.
8
In addition, it is encouraged to adopt a written conflict of interest policy, as well as transparency in financial transactions.
This compensation process matters a great deal to NPOs, especially regarding the tradeoff between remaining competitive with for-profits in the attraction of talent and the desire of donors to grant their money to nonprofits refraining from excessive compensation schemes (Balsam and Harris, 2014). In line with Hansmann (1980), Frumkin and Keating (2010) elucidate that the non-distribution constraint that nonprofits face, imposes the main difference between nonprofit and for-profit pay levels. They state that in the U.S., the IRS is responsible for monitoring nonprofit compensation schemes, but that reversal of the tax-exempt status is rarely executed.
An important source for revenue for some nonprofit subsectors, such as the educational institution field, is the amount of grants and contributions that are offered by donors. As to my best knowledge, however, no research addressed the dimensions of donations, compensation and governance for the specific field of educational institutions. For the nonprofit sector as a whole, some evidence has been documented. Balsam and Harris (2014) find consistent evidence that sophisticated donors reduce their gifts to NPOs when they are featured by excessively high CEO compensation. On the other hand, Harris et al. (2015) document that donor rewards are positively correlated with six factors that capture good governance in nonprofits, including formal written policies, independent audits and audit committees, review and approval of executive compensation, board oversight, management characteristics and accessible financial information. This confirms the findings of Garner & Harrison (2013), being that above average compensation for executives is associated with poor firm performance as measured by the amount of donations received, especially in bad governed firms. Lichtenstein, Drumwright and Braig (2004), at last, added a paper describing the beneficial effect non-profit reputation has on the donations – compensation relationship.
As for dimensions of donations, governance and compensation, most of the nonprofit literature is centered around the subfield of healthcare organizations (Brickley, van Horn and Lawrence, 2002; Eldenburg et al., 2004). Literature on educational institutions, however, is limited and not very recent. Besides a lack of empirical studies, a few papers commented on governance of educational institutions. Several authors (Dearlove, 1997; Kezar and Eckel, 2004) point out the need for a
9
movement towards a more corporate structure of governance, whereas before, this structure was characterized more by shared and participative governance. Broadbent (2007) stresses the importance for educational institutions of a system of performance measurement tracked by independent directors. On the other hand, he argues that these entities experience a strong form of non-measurable governance. Bleiklie and Kogan (2007), finally, report that governance of universities follows a supranational trend. At first, decision-making processes were guided by the vision of the university as a republic of scholars. Now, the consensus has moved towards the university as a stakeholder organization, where institutional autonomy is considered as the basis for managerial decision-making, satisfying interests of all major stakeholders. I believe a comparison with the ongoing discussion on the shareholder- and stakeholder approach in for-profit firms is justified here.
As of firm structure, educational institutions, especially the larger ones, are typically structured featuring a university president (CEO), a board of directors, and a team of faculty deans. Some entities, mostly universities, also have a form of student delegation participating in the decision-making process (Bleiklie and Kogan, 2007).
B. Regulation
Harris et al. (2015) mention endogeneity concerns as both governance and donations as well as compensation and donations might suffer from simultaneous causality. To address these concerns, changes in the regulatory framework for nonprofits need to be identified. As experienced in many other sectors in the U.S., regulation of nonprofit organizations is characterized by overlap of federal- and state-level regulation, explains Hitoshi Mayer (2015). At the state level the attorney general acts as the main regulator, whereas the Internal Revenue Service (IRS) does so at the federal level, he compares. The attorney general imposes fiduciary duties on board members and officers, ensures that nonprofit assets stay in possession of charities, and offers nonprofits exemption from state income and other taxes. The IRS, on the other hand, exerts its controlling duty merely by deciding whether or not a nonprofit is eligible for a tax-exempt status. The role of controlling private parties is limited to cooperative activities with government entities.
Some major state-level regulatory changes in the last fifteen years are the Nonprofit Integrity Act (California, 2004), the District of Columbia Nonprofit Corporation Act (D.C.,
10
2012) and the Nonprofit Revitalization Act (New York, 2013). In Massachusetts, a 2011 law titled “An Act Further Regulating Public Charities” introduces harsher penalties for nonprofits which fail to comply with the state level filing requirements. In section IV.C, I introduce a difference-in-differences analysis employing the introduction of the 2012 D.C. and 2011 Massachusetts regulatory changes. According to Venables LLP (2012), one of the most important implications of the former is that executive committees and other committees exercising board authority must consist only of directors that are appointed by a majority of the directors in office, yielding a rise in control over the board selection process. This is depicted as a good governance indicator, for instance by Harris et al. (2015). Besides, the act provides procedures for approaching conflicts of interest, implying formal written policies for conflicts. Lastly, the act permits for electronic membership meetings and voting, which slightly improves the voting process in NPOs. The introduction of harsher financial penalties for non-compliants in Massachusetts, which vary from 50 to 10,000 U.S. dollar, induces better compliance of nonprofits. The Massachusetts nonprofit law requires among other things that NPOs have their financial statements audited and have an election process for board members, causing an increase in the compliance rate to yield a raise in having these governance characteristics. C. Hypotheses Following directly from the framework of past research and regulations spelled out above, four expectations are posed as hypotheses in order to give a structured answer to the research question guiding this thesis. First, from literature in both the nonprofit and for-profit field, it is expected that compensation has a negative effect on the amount of donations raised in NPOs. Second, since governance provisions are designed to deal with, or prevent excessive compensation, I expect the negative effect of compensation to disappear, or become positive when NPOs possess such provisions. The literature also mentioned endogeneity concerns due to simultaneous causality of governance and donations. To deal with these problems, state level regulatory changes are employed in a difference-in-differences set-up. Regarding this set-up, I expect governance levels to rise significantly after the two depicted changes in D.C. and Massachusetts, causing an increase in donations in this way. This is the third hypothesis. Finally, from the fact that documented literature is focused on CEO compensation, rather than Officer or Total
11
compensation, I expect all regression results to be most pronounced for the models with CEO compensation as explanatory variable.
Methodologically, this research builds on papers of Balsam and Harris (2014), Harris, Petrovits and Yetman (2015), Bebchuk, Cohen and Ferrell (2009), Chaochharia and Grinstein (2009), and Card and Krueger (1994). This is extensively explained in section IV, methodology. III. Data To compose the panel data sample needed to conduct this research, data is gathered from Form 990 filings provided by the Internal Revenue Service for the years 2008 up to and including 2013. Non-profit organizations are required to file this form in order to preserve their tax-exempt status. The year range is selected merely due to data availability: the IRS has not published data for the years after 2013, while at the same time, filings of the Form 990 before 2008 are less informative, and do not contain the variables I need to conduct the research. The initial NPO data set consists of 121,609 unique filings. 86,257 of these filings were delivered by charitable organizations, defined by a 501(c)(3) status, excluding private foundations. Consecutively, 4,188 observations are dropped because they are provided by firms that are permitted to file the less informative Form 990-EZ or Form 990-N, i.e. because total assets and gross receipts are smaller than 500,000 and 200,000 dollars, respectively. Ultimately, the sample is reduced by selecting the entities of interest, being educational institutions. This brings forward a final sample of 11,508 firm year observations resulting from tax-filings of 2,376 unique tax-exempt companies over the years 2008-2013. Most of the regressions performed use lagged values of some of the variables, so that the number of observations in these regressions is 9,023. The data set is characterized as a so called panel data set, yielding a large number of entities and a small number of years. The number of entities in every year from 2008 until 2013 is 1,903, 1,976, 1,839, 1,870, 1,918 and 2,002, in increasing order of years. Descriptive statistics can be found in table I, reported below.
Descriptive statistics containing the mean, median and type of variables retrieved from the IRS Form 990 data set are given in table I. Even though the variables containing numbers are regressed in their logarithmic form, here they are presented in their original, absolute values. In all regressions, these variables are winsorized at the
12
1% level to circumvent the effects of extreme observations. Moreover, four of the six factors capturing good corporate governance according to Harris et al. (2015) are included, videlicet formal written policies, independent audits and audit committees, review and approval of executive compensation, and invoice in board election process. In addition, as mentioned by Chhaochharia and Grinstein (2009), the percentage of independent voting members to the total number of members in the governing body, and the resulting aggregate measure of corporate governance are reported as well. Table I Descriptive Statistics Table I presents financial and governance statistics for U.S. non-profit educational institutions between 2008 and 2013. The total sample consists of 2,376 NPOs that filed 11,508 Form 990 filings with the IRS. In Panel A presents the financial statistics. Total Donations include grants and contributions to the nonprofit received from all sources. CEO, Officer, and Total Reported Compensation follow from their Form 990 columns. All values are in millions of U.S. dollars ($). In Panel B, Governance Indicators are presented. An independent voting member is defined as a director that has not been employed by the organization in the last 3 years and that has no relatives working at the NPO. Under coding, the variable is denoted as continuous, dummy or discrete with integers. The dummy variables are equal to 1 if the nonprofit features the governance indicator during a given year. The aggregate measure is composed of the four dummy variables and a dummy variable that equals 1 if the % of independent voting members >50%.
Panel A: Financial Statistics
Description Coding Mean Median
Total Donations Continuous 25.80 4.11 CEO Compensation Continuous 0.65 0.37 Officer Compensation Continuous 1.20 0.68 Total Reported Compensation Continuous 1.63 0.98 Total Assets (Size) Continuous 378.00 77.10 Fundraising Expenses Continuous 2.01 0.74 Government Grants Continuous 13.40 0.23 Panel B: Governance Indicators
Description Coding Mean Median
Number of Voting Members in Governing Body Continuous 24.41 24 Percentage of Independent Voting Members Continuous 0.916 0.971 Audited Financial Statements = 1 for Yes = 0 for No 0.929 1 Written Policy or Procedure = 1 for Yes = 0 for No 0.037 0 Election of Board Members = 1 for Yes = 0 for No 0.139 0 Compensation Process CEO = 1 for Yes = 0 for No 0.899 1 Aggregate Measure of Governance Discrete 0-5 2.468 2
13
IV. Methodology
The data obtained as described in the data section is carefully analyzed following three subsequent research methodologies. By doing so, the composed hypotheses can be tested to finally arrive at the conclusion of this thesis. To start with, the direct relationship between executive compensation and the level of donations is examined by an analysis of the panel data with OLS and with industry- and year fixed effects regressions. The analysis builds on three different variables for compensation, i.e. CEO Compensation, All Officer Compensation, and Total Reported Compensation. The “All Officers” variable includes compensation to board members and all other directors. I include all three measures, because I expect a possibility for different outcomes. CEO compensation is more salient to the public and donors, while compensation to all officers and all employees represent larger amounts of cash, which should be of more concern to donors.
Thereafter, these regressions are expanded by the inclusion of five measures that capture good corporate governance: four of them as documented in a prior research of Harris, Petrovits and Yetman (2015), the other as described by Chhaochharia and Grinstein (2009). In this second regression I introduce a novel aspect by carrying out regressions both with the five separate measures of good corporate governance as indicated in the data section, as well as with one measure aggregating these indicators. This aggregate measure is created using a kind of entrenchment score index, analogous to the methodology Bebchuk, Cohen and Farell exploit in their 2009 paper. I add both regressions to the results because I expect both outputs to be informative. From the aggregate measure it can be derived whether the level of total contributions and donations is influenced by the addition of more governance factors. With the factors included separately, on the other hand, I can derive which of the factors matters most to donors making these contributions. This presents a broader horizon of the research than focusing on one of the two methods.
Then, at last, to address some possible endogeneity issues, a difference-in-differences regression is performed using an exogenous change in non-profit regulation at the state level. Here, the methodology pursued is akin to Card & Krueger (1994).
14
A. Total Donations and Three Compensation Measures
The easiest and plainest way of establishing the impact the level of compensation has on the amount of donations a non-profit organization raises, is by exerting a cross-sectional Ordinary Leased Squares regression featuring the amount of donations as dependent, and the level of compensation as the explanatory variable. Since the data contains multiple firm-years, the models are estimated using both random firm effects and industry- and year fixed effects. In these regressions, I use three different measures for the compensation the NPO grants: compensation to the CEO, compensation to all officers, and total compensation granted. All three these measures are included in the output due to expected differences in salience and dollar amounts.
As Balsam and Harris (2014) pointed out, donors amend the heights of their donations based on the compensation level in the foregoing year. Therefore, lagged values of CEO compensation are used in this regression. The main control variables are similar to the ones Harris et al. (2015) use in their research being denoted by fundraising expenditures, government grants, and firm size as represented by the total assets variable. As is customary in regressions of this kind (Woolridge, 2002), total donations, and the lagged values (t-1) of compensation levels, fundraising expenses, government grants and total assets are all scaled by taking the natural logarithm of these variables. Likewise, the common habit of financial studies to correct for outliers by winsorizing at the 1% level is adopted by this study as well.
Next, the data is analyzed treating the data as panel data. Following the definition as presented by Stock & Watson (2012), a panel data model combines a longitudinal, cross-sectional analysis with time-series data, aiming at the observation of the behavior of entities across time. Industry- and year fixed effects are added to the random effects model to control for year-by-year trends and industry-specific unobserved heterogeneity in the factors that drive total donations. Industry codes are derived from NTEE core codes as presented by the NPOs in their 990 filings. These codes are equivalent to the SIC codes for for-profit firms. In this research, nonprofits are assumed to be in the same industry if the first 3 digits of their code are equal, reflecting the major subsector, specific activity areas and specific types of organizations.
All in all, the tests are performed using variants of the panel data regression equation as stated below in equation (1). Following the literature review, we expect coefficient 𝛽" to be negative, and 𝛽# and 𝛽$ to be positive (A.O. Weisbrod and
15
Dominguez, 1986). For government grants, the results from documented papers are mixed. Among others Emanuele and Simmons (2004) and Yetman and Yetman (2009) found evidence of donors refraining from contributing to higher government grants firms. On the other hand, others such as Okten and Weisbrod (2000), and Petrovits, Shakespear and Shih (2011) report a positive relation between donations and government grants. 𝟏 𝑙𝑛𝑇𝑂𝑇𝐴𝐿 𝐷𝑂𝑁𝐴𝑇𝐼𝑂𝑁𝑆1,3 = 𝛽5+ 𝛽"𝑙𝑛𝐶𝑂𝑀𝑃𝐸𝑁𝑆𝐴𝑇𝐼𝑂𝑁 𝐼𝑁𝐷𝐼𝐶𝐴𝑇𝑂𝑅1,3<"+ 𝛽#𝑙𝑛𝑇𝑂𝑇𝐴𝐿 𝐴𝑆𝑆𝐸𝑇𝑆1,3<" + 𝛽$𝑙𝑛𝐹𝑈𝑁𝐷𝑅𝐴𝐼𝑆𝐼𝑁𝐺 𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑆1,3<"+ 𝛽A𝐺𝑂𝑉𝐸𝑅𝑁𝑀𝐸𝑁𝑇 𝐺𝑅𝐴𝑁𝑇𝑆1,3<" + ∑𝛾1𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦1+ ∑𝛿3𝑌𝑒𝑎𝑟3+ 𝜀1,3 B. Total Donations, Compensation and Five Governance Measures
Since this research is in the first place aiming at making a contribution to the knowledge in the field of agency problems and corporate governance, it is interesting to see whether the expected negative effect of CEO compensation disappears when it is interacted with variant measures of good corporate governance. For the construction of these measures, I use the binary factors as described by Harris et al. (2015), of which 4 are available in my data set, being independent audits and audit committees, a written policy or procedure for disagreement, review and approval of executive compensation, and the election of board members. In addition, a fifth measure is added indicating the independency of the majority of the voting board members, with the threshold at 50% independent board members. The importance of this factor is stressed by a paper of Chhaochharia and Grinstein (2009). Guthrie, Sokolowsky and Wan (2012) cast doubt on their work, keeping in mind the managerial power hypothesis, so the significance of this factor may be lower than expected.
First, the compensation measures are interacted with an aggregate measure of these governance indicators. This aggregate measure is created following the methodology of Bebchuk et al. (2009), who created an entrenchment index with six measures of corporate governance selected from the twenty-four provisions included in the Gompers, Ishii and Metrick index. These factors, however, are based on for-profit firm characteristics such as stock returns. Therefore, I create the measure similarly using the
16
factors from Harris et al. (2015), as mentioned above, resulting in a measure that takes on a value between 0 and 5, increasing in governance provisions the organization meets during the given year. The different levels of the aggregate measure are included as dummy variables, to account for nonlinearity of the included factors. That is, the difference between having 0 or 1 provision might diverge from the difference between having 4 or 5 provisions. From the coefficient on the interaction variable, it can be learned whether non-profits with better corporate governance enjoy more donations following a rise in compensation. The regression model used to retrieve the coefficients for the aggregate measure is reflected in equation (2). 𝟐 𝑙𝑛𝑇𝑂𝑇𝐴𝐿 𝐷𝑂𝑁𝐴𝑇𝐼𝑂𝑁𝑆1,3 = 𝛽5+ 𝛽"𝑙𝑛𝐶𝑂𝑀𝑃𝐸𝑁𝑆𝐴𝑇𝐼𝑂𝑁1,3<"+ 𝛽#𝐴𝐺𝐺𝑅𝐸𝐺𝐴𝑇𝐸 𝐺𝑂𝑉𝐸𝑅𝑁𝐴𝑁𝐶𝐸1,3<" + 𝛽$𝑙𝑛𝐶𝑂𝑀𝑃𝐸𝑁𝑆𝐴𝑇𝐼𝑂𝑁1,3<"∗ 𝐴𝐺𝐺𝑅𝐸𝐺𝐴𝑇𝐸 𝐺𝑂𝑉𝐸𝑅𝑁𝐴𝑁𝐶𝐸1,3<" + 𝛽A𝑙𝑛𝑇𝑂𝑇𝐴𝐿 𝐴𝑆𝑆𝐸𝑇𝑆1,3<"+ 𝛽R𝑙𝑛𝐹𝑈𝑁𝐷𝑅𝐴𝐼𝑆𝐼𝑁𝐺 𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑆1,3<" + 𝛽S𝐺𝑂𝑉𝐸𝑅𝑁𝑀𝐸𝑁𝑇 𝐺𝑅𝐴𝑁𝑇𝑆1,3<"+ ∑𝛾1𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦1+ ∑𝛿3𝑌𝑒𝑎𝑟3+ 𝜀1,3
After reviewing these results, it might be interesting to decompose the aggregate index into its original five factors to see which of the governance indicators matters the most to donors. To test this, the separate measures are now interacted with the three compensation measures to distinguish an intuition on specific importance. From Harris et al. (2015) we know that theoretically, the measures itself have a positive impact on the amount of donations raised. The coefficient on the aggregate measure is therefore obviously expected to have the same direction. The sign of the coefficients on the aggregated and separate interaction variables is negative if donors are still convinced that their money is wasted on excessive compensation of management, officers or employees in general. A positive coefficient indicates that donors now believe that the higher compensation level is justified, for example because they associate a rise in compensation with more skilled employees. They might make this association because of the presence of good governance indications.
For these regression models, again, the natural logarithm of total donations is used to construct the dependent variable and the natural logarithms of the lagged (t-1) values of the compensation measures interacted with the governance indicators reflect the explanatory variables. Control variables are again lagged total assets, fundraising
17
expenses and government grants, all in their natural logarithmic form. The regressions are run in OLS and fixed effects form. For the separate indicators, the corresponding results are extracted by estimating variants of the regression equation as represented below in equation (3). 𝟑 𝑙𝑛𝑇𝑂𝑇𝐴𝐿 𝐷𝑂𝑁𝐴𝑇𝐼𝑂𝑁𝑆1,3 = 𝛽5+ 𝛽"𝑙𝑛𝐶𝑂𝑀𝑃𝐸𝑁𝑆𝐴𝑇𝐼𝑂𝑁1,3<"+ 𝛽#(𝐺𝑂𝑉𝐸𝑅𝑁𝐴𝑁𝐶𝐸 𝐼𝑁𝐷𝐼𝐶𝐴𝑇𝑂𝑅𝑆1,3<") + 𝛽$𝑙𝑛𝐶𝑂𝑀𝑃𝐸𝑁𝑆𝐴𝑇𝐼𝑂𝑁1,3<"∗ (𝐺𝑂𝑉𝐸𝑅𝑁𝐴𝑁𝐶𝐸 𝐼𝑁𝐷𝐼𝐶𝐴𝑇𝑂𝑅𝑆1,3<") + 𝛽A𝑙𝑛𝑇𝑂𝑇𝐴𝐿 𝐴𝑆𝑆𝐸𝑇𝑆1,3<"+ 𝛽R𝑙𝑛𝐹𝑈𝑁𝐷𝑅𝐴𝐼𝑆𝐼𝑁𝐺 𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑆1,3<" + 𝛽S𝐺𝑂𝑉𝐸𝑅𝑁𝑀𝐸𝑁𝑇 𝐺𝑅𝐴𝑁𝑇𝑆1,3<"+ ∑𝛾1𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦1+ ∑𝛿3𝑌𝑒𝑎𝑟3+ 𝜀1,3 C. Difference-In-Differences Analysis Endogeneity is a common concern in governance studies. Although I tried to cover the endogeneity concern of omitted factors driving the results by the inclusion of control variables and industry- and year fixed effects, here, it might well be that both the level of executive compensation or corporate governance factors as well as the amount of donations raised are suffering from reversed, or simultaneous causality. This means that both variables simultaneously determine each other, which violates the Gauss-Markov OLS assumptions. Considering that non-profit regulation is increasingly based on state level rather than on federal level legislation (Hitoshi Mayer, 2016), a difference-in-differences design exploiting regulatory changes regarding governance of non-profit organizations on the state level may be feasible to tackle this endogeneity issue. In this way the impact of governance factors is examined solely through an exogenous change in regulation. I follow a model as it is used by Card and Kruger in their 1994 paper. To be able to do so, the regulatory change can only affect the amount of donations through its effect on non-profit governance. To disentangle the effect of the change, two differences are taken. To start with, the difference between the state that is affected by the change compared to a control group resembling the trend, i.e. a comparable state that is unaffected by the change. The second difference is then taken for the affected and the control state before and after the regulatory change. Methodologically, this brings forward the creation of two dummy variables, which are interacted to represent the total difference-in-differences estimator. To reduce the possibility of omitted variable bias, the same control variables are added as in the regressions earlier discussed, i.e.
18
size, fundraising expenses and government grants, in natural logarithmic form. Since the regulation is state-based, no state or year fixed effects are used here, as the effects of the treatment and time dummies would be controlled away. The difference-in-differences model is accordingly designed as reflected below in equation (4). 𝟒 𝑙𝑛𝑇𝑂𝑇𝐴𝐿 𝐷𝑂𝑁𝐴𝑇𝐼𝑂𝑁𝑆1,3X" = 𝛽5+ 𝛽"𝑇𝑅𝐸𝐴𝑇𝑀𝐸𝑁𝑇1 𝑥 𝑇𝐼𝑀𝐸3+ 𝛽#𝑇𝑅𝐸𝐴𝑇𝑀𝐸𝑁𝑇 𝐷𝑈𝑀𝑀𝑌1 + 𝛽$𝑇𝐼𝑀𝐸 𝐷𝑈𝑀𝑀𝑌3+ 𝛽A𝑙𝑛𝑇𝑂𝑇𝐴𝐿 𝐴𝑆𝑆𝐸𝑇𝑆1,3<" + 𝛽R𝑙𝑛𝐹𝑈𝑁𝐷𝑅𝐴𝐼𝑆𝐼𝑁𝐺 𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑆1,3<"+ 𝛽S𝐺𝑂𝑉𝐸𝑅𝑁𝑀𝐸𝑁𝑇 𝐺𝑅𝐴𝑁𝑇𝑆1,3<"+ 𝜀1,3 Here, the coefficient of interest, 𝛽", is the difference-in-differences estimator, defined by the interaction of the treatment and the time dummy. For the regulatory change, I use two state level changes in laws governing non-profit organizations in states with a relatively large number of educational institutions. The changes are retrieved from the website of Venable LLP (June, 2017), a large legal service provider to nonprofits in the U.S. An intuition on these changes is presented in section II.B, under Regulation.
First, in the District of Columbia (D.C.) the ‘District of Columbia Nonprofit Corporation Act went into effect on January 1st, 2012. This act yields various changes in governance, and as most significant aspects, it improves default rules for director standards of conduct and liability, indemnification, member voting rights and ballot voting procedures. It furthermore implies expanded record keeping requirements, codified standards of conduct and the permission for the creation of a so called designated body, which resembles a new governing body that may exercise powers comparable to the board of directors. As indicated in section II.B, these amendments in general should raise governance levels through a higher amount of nonprofits classified by an ‘election of board members process’, and an increase in use of the ‘formal written policy’ provision. Both these factors are included in the factors capturing good governance in NPOs that are reviewed in this research.
Second, in Massachusetts there has been a nonprofit law titled “an Act Further Regulating Public Charities”, imposing new fees and penalties on charities which fail to comply with the nonprofit filing requirements. The act has been effective as of January 1st, 2011. Governance levels of the nonprofits eligible to the new law are expected to increase significantly due to increased compliance, caused by the raise in penalty fees.
19
Complying firms should, among other things, have their financial statements independently audited and have an election process for board members. These governance provisions are included in the aggregate governance measure, which is therefore expected to increase following the regulatory change.
As a control group, I selected the state of New York, since this state is featured by a large number of educational institutions and is comparable in geographical location and political preferences in the elections of 2008, 2012 and 2016 (democratic). For both states, the parallel trends assumption prior to the treatment holds, i.e. coefficients of the pre-treatment years interacted with the treatment indicator are not significantly different from zero. This indicates that D.C. and Massachusetts follow similar trends as the control state New York. All in all, I expect both these regulatory changes to significantly impact the amount of donations received by the institutions. The second one with greater statistical significance due to a larger amount of observations. V. Results Below, an overview of the econometric results is given following the methodology presented in the last section, followed by an interpretation of the numbers. To test for robustness of the results that I expect to find as reported by Balsam and Harris (2014), a Hausman test as described in Stock & Watson (2012) is performed to test whether the panel data matches a regression with Random Effects (RE) or Fixed Effects (FE). The chi-squared estimator of the Hausman test is significant at the 1% level, indicating that the Fixed Effects model is at least consistent and therefore preferred over the Random Effects model. This result brings forward the that there are time-invariant, organization-specific factors driving the amount of donations. Performing the regressions with firm fixed effects, however, brings forward insignificant results, probably due to the low number of observations per nonprofit, i.e. 6 at most. Therefore, I decided to exert the regression controlling for unobserved heterogeneity using fixed effects for industry and year variables. A test for the inclusion of State fixed effects yielded insignificant results due to multicollinearity issues.
20 A. Total Donations and Three Compensation Measures Table II contains the econometric results of the regressions as discussed in the first part of the methodology section, derived from equation (1). It includes both standard OLS regression models (columns (1), (3) and (5)) and panel data models (columns (2), (4) and (6)) with NTEE industry- and year fixed effects for the years 2008-2013. All variables are lagged one period (t-1), causing the number of observations to shrink from 11,508 to 9,023, provided by 2,162 non-profit organizations.
When examining the results presented in table II, it can be seen that in the OLS regressions, the coefficients on the three explanatory variables are all negative, with CEO compensation and Total Compensation significant at the 5% and 10% levels, respectively. This is consistent with my expectation derived from the paper of Balsam & Harris (2014), and indicates a negative correlation between donor grants in the year following a raise in compensation, at least displayed by CEO and total compensation. What is striking, however, is that in the following panel data regressions with industry- and year fixed effects, the sign of all three the coefficients changes. The coefficient on CEO compensation is insignificant, but the ones on officers’ compensation and total compensation granted are positive and significant at the 5% and 10% level, respectively. Economically, this yields that a 1% rise in all officers’ compensation increases the amount of total donations by 0.012%, whereas the same rise in total reported compensation raises total donations by 0.018%. The R-squared values of the panel data models are slightly higher than their OLS equivalents, indicating that the former models are more sophisticated in explaining the variance of the observations.
The changing direction of the coefficients is remarkable. As of factors that are omitted in the OLS regressions, but that are captured by the time-invariant industry fixed effects, firm and manager reputation are likely to play a role. As Lichtenstein et al. mentioned in their 2004 paper, a good, reliable reputation may cause a raise in compensation to cast a signal of improved performance and ability to donors, inducing them to raise their contributions rather than cut them. Conyon (2014) states that this effect of reputation is captured by running a firm fixed effects regression, and in a similar way it might run through reputation of an industry as well. Another explanation can be the presence of unobserved good corporate governance indicators. This possibility is addressed in the next section. The control variables are all positive and significant at the 1% level, as expected from the related literature. The year fixed effects
21
dummy for 2013 is significantly positive at the 5% level, implying that in this year, unobserved time trend factors positively influence the amount of donations. An F-test for joint significance of the variables is significant at the 1% level for all specifications.
Table II
Total Donations and Compensation
The table below presents estimates of OLS (columns (1), (3), and (5)) and panel data regressions (columns (2), (4), and (6), where the dependent variable is Total Donations to educational institutions and the dependent variables are specified above the columns. The sample consists of 9,023 observations from 2,162 nonprofits. The explanatory variables are CEO, all officer, and total compensation. Control variables are fundraising expenses, NPO size, and government grants. All variables are lagged (t-1), in natural logarithmic form, and winsorized at the 1% level. In the panel data models, industry – and year fixed effects are added. The numbers in parentheses reflect robust standard errors, clustered at the NPO-level. *, **, and *** indicate significance of the coefficients at the 10%, 5%, and 1% levels, respectively.
Dependent Variable: (1) (2) (3) (4) (5) (6)
Total Donations CEO CEO Officers Officers Total Total
CEO Compensation -0.026** 0.002 (0.012) (0.005) Officer Compensation -0.021 0.012** (0.013) (0.006) Total Compensation -0.036* 0.018* (0.019) (0.010) Fundraising Expenses 0.149*** 0.067*** 0.145*** 0.066*** 0.146*** 0.067*** (0.019) (0.010) (0.019) (0.010) (0.019) (0.010) Size 0.568*** 0.625*** 0.566*** 0.618*** 0.575*** 0.614*** (0.038) (0.039) (0.039) (0.039) (0.040) (0.038) Government Grants 0.097*** 0.055*** 0.096*** 0.055*** 0.096*** 0.055*** (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Dummy(Year = 2010) 0.038 0.037 0.039 (0.028) (0.028) (0.027) Dummy(Year = 2011) -0.002 -0.002 -0.002 (0.027) (0.027) (0.027) Dummy(Year = 2012) 0.036 0.035 0.037 (0.027) (0.027) (0.027) Dummy(Year = 2013) 0.064** 0.062** 0.064** (0.030) (0.030) (0.030) Observations 9,023 9,023 9,023 9,023 9,023 9,023 R-squared 0.509 0.568 0.508 0.568 0.508 0.567 Industry & Year FE - + - + - + Number of NPOs 2,162 2,162 2,162 B. Total Donations, Compensation and Aggregate Governance Measure In table III the estimated results are reflected emerging from the regression models analyzed as described in part B of the methodology section, based on equation (2). Again, columns (1), (3) and (5) provide simple OLS regression results, whereas columns
22
(2), (4) and (6) contain panel data models with year- and industry fixed effects, all for the years 2008-2013. Both explanatory as well as control variables are lagged (t-1). The three compensation measures are interacted with five dummy variables indicating whether a nonprofit has a certain amount {0,5} of governance provisions during a given year, regardless of the type of provision. The provisions that are included in this aggregate measure are described in section B of the methodology. Combined with the exclusion of entities omitting information on one or more governance factors, this brings forward 8,992 observations from 2,158 non-profit educational institutions.
As shown by the fixed effects regressions in table II, the sign of the coefficients of the compensation measures changed when industry-specific controls and a time trend indicator were included as fixed effects. Whereas the conjecture that firm- or industry reputation is driving the positive results is hard to test empirically, an analogous investigation can be performed employing available data on non-profit corporate governance factors, to test if they impact the compensation – donations relationship. Table III shows the coefficients of the evaluation of the aggregate measure as described in the methodology.
Inspecting the baseline effects of the aggregate measure presents some clear intuitions. First of all, it is notable that, with inclusion of the aggregate measure, all compensation measures itself are insignificant. The measures itself, now, do not significantly impact donations. Second, a look at the dummy variables clarifies the effects that different amounts of governance provisions have on the intercept of the total donations function. It is evident that the effects are most pronounced when examining CEO compensation as explanatory variable. As for economic significance: when a nonprofit features 2, 3, 4, or 5 provisions during a given year, the amount of donations rises with 0.64%, 0.67%, 0.76%, or 2.25%, respectively. These corresponding coefficients are all significant at the 5% level, and increasing in the amount of provisions that are included. The fact that the governance indicators only seem to matter when reviewing CEO compensation can be attributed to salience of CEO compensation. Donors are usually well informed as it comes to CEO compensation, whereas they are more ignorant when it comes to officer and total compensation. This is supporting evidence for the fourth hypothesis, predicting salience of CEO compensation.
23 Table III Total Donations, Compensation and Aggregate Governance Table III reflects estimates of OLS (columns (1), (3), and (5)) and panel data regressions (columns (2), (4), and (6), where the dependent variable is Total Donations to educational institutions and the independent variables are specified above the columns. The sample consists of 8,992 observations from 2,158 nonprofits. The explanatory variables CEO, all officer, and total compensation, are interacted with 5 dummies representing the levels of the aggregate governance measure. One of these dummies equals 1 if a nonprofit has {1,5} governance provisions in a given year, regardless of the type of provisions. Control variables are fundraising expenses, NPO size, and government grants. All variables are lagged (t-1), winsorized at the 1% level, and in natural logarithmic form. In the panel data models, industry – and year fixed effects are added. The numbers in parentheses reflect robust standard errors, clustered at the NPO-level. *, **, and *** indicate significance of the coefficients at the 10%, 5%, and 1% levels, respectively.
Dependent Variable: (1) (2) (3) (4) (5) (6)
Total Donations CEO CEO Officers Officers Total Total
CEO Compensation -0.207 -0.022 (0.128) (0.032) Officer Compensation -0.196 -0.040 (0.121) (0.032) Total Compensation -0.180 -0.035 (0.129) (0.060) Dummy Aggregate = 1 0.039 0.013 -0.400 -0.360 0.018 -0.462 (0.852) (0.368) (0.671) (0.471) (1.234) (0.743) Dummy Aggregate = 2 0.010 0.642** -0.348 0.317 -0.576 0.181 (0.846) (0.311) (0.676) (0.383) (1.239) (0.798) Dummy Aggregate = 3 0.524 0.672** 0.237 0.358 0.553 0.313 (0.836) (0.314) (0.651) (0.387) (1.232) (0.787) Dummy Aggregate = 4 -0.200 0.757** -1.500 0.486 -2.948 -0.229 (1.137) (0.349) (1.432) (0.453) (2.808) (1.005) Dummy Aggregate = 5 -3.925 2.248** -7.414*** 1.585 -8.628*** 0.521 (2.577) (1.137) (2.807) (2.070) (2.756) (2.105) Aggregate = 1 x Comp. Measure 0.186 0.067* 0.187* 0.095** 0.118 0.089 (0.129) (0.035) (0.096) (0.039) (0.132) (0.055) Aggregate = 2 x Comp. Measure 0.195 0.022 0.187** 0.051 0.173 0.054 (0.128) (0.032) (0.094) (0.033) (0.130) (0.061) Aggregate = 3 x Comp. Measure 0.166 0.020 0.155* 0.047 0.102 0.044 (0.128) (0.032) (0.093) (0.033) (0.130) (0.060) Aggregate = 4 x Comp. Measure 0.217 0.008 0.279** 0.033 0.348 0.079 (0.140) (0.033) (0.130) (0.036) (0.219) (0.074) Aggregate = 5 x Comp. Measure 0.504** 0.097 0.703*** 0.038 0.747*** 0.032 (0.230) (0.087) (0.220) (0.144) (0.217) (0.145) Fundraising Expenses 0.140*** 0.067*** 0.137*** 0.066*** 0.138*** 0.066*** (0.019) (0.010) (0.019) (0.010) (0.019) (0.010) Size 0.577*** 0.617*** 0.570*** 0.612*** 0.581*** 0.607*** (0.038) (0.038) (0.039) (0.038) (0.040) (0.038) Government Grants 0.095*** 0.056*** 0.095*** 0.055*** 0.094*** 0.055*** (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Observations 8,992 8,992 8,992 8,992 8,992 8,992 R-squared 0.516 0.571 0.516 0.571 0.516 0.570 Industry & Year FE - + - + - + Number of NPOs 2,158 2,158 2,158
24
By switching to a review of the interaction variables, acquaintance is offered regarding the impact the governance provisions have on the slope of the donations – compensation interplay. To start with, all the interaction coefficients have a positive direction, indicating that the intuition presented in the literature is confirmed at first sight, i.e. good corporate governance can address the detrimental effects of excessive compensation on donations. Regarding significance, however, the results are less convincing. For CEO and officer compensation in the panel data specification, the inclusion of 1 governance indicator is positively significant at the 10% level and 5% level, respectively. These can be interpreted economically as such that for a nonprofit organization meeting one governance factor (aggregate = 1), a 1% rise in CEO compensation to cause donations in the next year to go up by 0.07% when taking into account significant coefficients only. For officer compensation this rise mounts to 0.10%.
In the OLS models, two things are noteworthy. First, the positive impact the aggregate measure has interacted with the compensation measure, is only significant when an NPO features all five governance indicators in a given year. This could indicate statistical significance support due to clustering of the provisions, i.e. if a nonprofit has one provision, it is very likely to have all, or at least most of the provisions. In that case, however, the effect should also appear in the panel data models. On top of that, correlation between the five governance provisions is low, as reflected in table VII. The second oddity is that all governance level dummies are positively significant in impacting the slope of the officer compensation effect on donations. On the other hand, economically, the coefficients are only slightly different from the ones on CEO compensation interaction variables. Statistical significance is derived, thus, from more precise estimations.
Overall, the table provides supporting evidence for the hypothesis that the presence of good corporate governance factors positively impacts the correlation between compensation and the amount of donations in NPOs. When examining CEO compensation, the base level is also positively correlated with inclusion of the governance indicators itself, increasing in the amount of governance provisions met. Furthermore, since the compensation measures itself are all insignificant now, the significantly positive effect the unobserved industry- and year fixed effects capture in table II, are partly driven by the governance indicators included in table III, at least for
25 CEO and officer compensation at the aggregate = 1 level. The F-statistic is significant at the 1% level for all models specified. The most important limitation of the regression specifications presented in table III, is that the use of an aggregate measure of governance only disentangles what impact the amount of provisions has on the compensation – donations correlation. To examine which of the included governance indicators matters the most to donors, in the next section an analysis is performed using the provisions separately.
C. Decomposition of the Aggregate Measure
In table IV, estimates are presented following the regression models as presented in part B of the methodology. They all have their origin in regression equation (3), and once more, columns (1), (3) and (5) contain OLS regressions, while columns (2), (4) and (6) are filled with estimates following the fixed effects model with industry- and year fixed effects, all for the years 2008-2013. Governance factors as well as compensation measures and control variables are regressed using their lagged values (t-1). The compensation measures are interacted with five dummy variables indicating whether an NPO possesses the corresponding type of provision during a given year. Total observations cumulate to 8,992, filed by 2,158 nonprofit educational institutions.
As brought forward by the coefficients reported in table III, the corporate governance factors taken together in a single, aggregate measure seem to have a positive effect (if any) on the slope of the correlation between the compensation measures and the amount of donations. When CEO compensation was used as explanatory variable in the panel data model, the intercept of this relation was significantly altered as well. This intercept was negatively influenced, however, in all three OLS regressions for NPOs that feature all five governance measures. The results are, so to say, somewhat ambiguous, which requires a closer look on the included governance factors. To extract which of the governance indicators drives the positive impact on the slope, analogous regressions are performed with the compensation measures interacted with the decomposed corporate governance indicators, rather than the aggregate measure. I will first comment on the effects the individual measures have on the intercept, followed by an analysis of the impacts regarding the slope of the compensation – donations relationship.
26
First of all, it is notable that for CEO and all officer compensation, the compensation measures itself are significantly positive again at the 5% and 10% level, respectively. The OLS coefficients are all negative, but insignificant. The direct and indirect, i.e. through compensation, impact of the governance provisions are discussed simultaneously this time.
At first sight, the coefficients expressing these effects altogether present a mirroring, somewhat difficult to interpret story. As for the panel data regressions, the provision of independent audit committees positively influences the amount of donations in all three compensation specifications. For the model with CEO compensation as explanatory variable, additionally, independent boards and a formal written policy for disputes are significantly positive as well. The striking aspect follows from the corresponding interaction variable coefficients of these significant estimates: the matching interactions that are significant are all carrying the sign opposite to their individual governance factors. In the CEO sample, for instance, the provision of an independent audit committee is positive and significant at the 5% level, while at the same time, the interaction variable of Independent Audits x CEO compensation is negative and significant at the 5% level.
Strange as this contradiction may seem, these results might as well be plausibly containing some information on the way donors amend the size of their donations. For the significant provisions, donors amend the amount of donations according to their presence. Considering the model where CEO compensation counts as the explanatory measure, the correlation for firms lacking any of the governance provisions is positive at the 5% level. Then for NPOs that possess one or multiple governance factors, the amount of donations rises regardless of the compensation level with 0.57%, 0.47%, and 0.51% for Independent Boards, Independent Audits and Formal Written Policy, respectively. This rise might indicate an increase in trust that the nonprofit uses the donations. Inherent to the positive impact of the provisions itself, there is a negative direction on the corresponding interaction variables, moving the overall effect of CEO compensation to around zero (-0.014) for firms who have all significant provisions, indicating a lower sensitivity to CEO compensation. The same holds for all officer and total compensation, retrieving a total significant effect of 0.007 and 0.014, respectively.
27 Table IV Total Donations, Compensation and Separate Governance Table IV presents estimates of OLS (columns (1), (3), and (5)) and panel data models (columns (2), (4), and (6), where the dependent variable is Total Donations to educational institutions and the independent variables are specified above the columns. The sample consists of 8,992 observations from 2,158 nonprofits. The explanatory variables CEO, all officer, and total compensation, are interacted with 5 dummies representing five different provisions of good governance. The dummy variables are equal to one if the NPO possesses the provision during a given year. Control variables are fundraising expenses, NPO size, and government grants. All variables are lagged (t-1), winsorized at the 1% level, and in natural logarithmic form, except the dummies. In the panel data models, industry – and year fixed effects are added. The numbers in parentheses reflect robust standard errors, clustered at the NPO-level. *, **, and *** indicate significance of the coefficients at the 10%, 5%, and 1% levels, respectively.
Dependent Variable: (1) (2) (3) (4) (5) (6)
Total Donations CEO CEO Officers Officers Total Total
CEO Compensation -0.044 0.060** (0.067) (0.024) Officer Compensation -0.077 0.047* (0.054) (0.027) Total Compensation -0.086 -0.015 (0.061) (0.043) Independent Board 0.679 0.571* 0.292 0.404 -0.273 -0.568 (0.537) (0.302) (0.490) (0.286) (0.624) (0.438) Independent Audits 0.110 0.468** 0.022 0.374* 0.671* 1.010*** (0.307) (0.201) (0.313) (0.220) (0.391) (0.329) Formal Written Policy 0.125 0.508* -0.917* 0.460 -3.670** 0.201 (0.615) (0.266) (0.550) (0.282) (1.643) (0.467) Election of Board Members -0.453 0.040 -0.933* -0.035 -1.014* 0.005 (0.384) (0.170) (0.525) (0.199) (0.593) (0.421) CEO Compensation Process 0.445 0.112 0.592* 0.286 0.784** -0.027 (0.326) (0.183) (0.344) (0.210) (0.397) (0.304) Ind. Board x Comp. Measure 0.020 -0.033 0.057 -0.005 0.090* 0.080** (0.055) (0.024) (0.047) (0.026) (0.052) (0.040) Ind. Audits x Comp. Measure 0.012 -0.029* 0.018 -0.018 -0.040 -0.066*** (0.026) (0.015) (0.027) (0.015) (0.031) (0.023) Written Policy x Comp. Measure -0.019 -0.045** 0.055 -0.040* 0.239** -0.021 (0.047) (0.022) (0.039) (0.023) (0.110) (0.036) Board Election x Comp. Measure 0.034 -0.005 0.069* 0.002 0.071* -0.001 (0.030) (0.012) (0.038) (0.014) (0.042) (0.030) Comp. Process x Comp. Measure -0.029 -0.002 -0.039 -0.020 -0.048 0.010 (0.031) (0.014) (0.030) (0.015) (0.033) (0.024) Fundraising Expenses 0.136*** 0.066*** 0.132*** 0.064*** 0.133*** 0.064*** (0.019) (0.010) (0.018) (0.010) (0.018) (0.009) Size 0.582*** 0.613*** 0.583*** 0.610*** 0.597*** 0.604*** (0.038) (0.038) (0.038) (0.038) (0.039) (0.037) Government Grants 0.094*** 0.055*** 0.093*** 0.055*** 0.092*** 0.055*** (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Observations 8,992 8,992 8,992 8,992 8,992 8,992 R-squared 0.515 0.576 0.516 0.573 0.517 0.574 Industry & Year FE - + - + - + Number of id 2,158 2,158 2,158
28
D. Difference-in-Differences Analysis
In order to overcome possible endogeneity issues caused by reversed, simultaneous causality of the dependent and the explanatory variables, a difference-in-differences factor is distinguished from a change in state level non-profit regulation. The 2012 “District of Columbia Nonprofit Corporation Act”, and the 2011 “Act Further Regulating Public Charities” of Massachusetts are employed in this analysis. Summary statistics for the affected states (D.C. and Massachusetts) and the control state (New York) can be found in the appendix in table IX, including time trends of frequencies and average values of the variables analyzed in the difference-in-differences model. The results and corresponding coefficients of the difference-in-differences (DiD) estimators for the two regulatory changes as elaborated in the fourth part of the methodology, are presented in table V. To carve out the effects of the regulatory changes, the DiD-estimator is regressed as independent variable on the amount of total donations received, the three compensation measures, and the aggregate measure of governance. The first will be the main model of interest, whereas the other four specifications present some information on the channel through which the effects run.
The statistical results are presented in table V. In Panel A, the treatment dummy equals 1 for NPOs registered in the District of Columbia during a given year, and 0 otherwise. The time dummy equals 1 for the years 2012 and 2013. 160 D.C.- and 1,145 New York observations accumulate to a total of 1,305 nonprofit observations. In Panel B, analogously, treatment is 1 for Massachusetts, and the time dummy even so for the years 2011, 2012 and 2013. The total sample here contains 1,877 observations: 735 from Massachusetts and 1,145 from New York. For both panels, dependent variables are specified in the header of the table, and control variables are lagged (t-1), natural logarithms of fundraising expenses, size and government grants.
The main coefficient of interest in both panels is the one describing the correlation between the DiD estimator – i.e. the Treatment x Time interaction variable – and the amount of donations. Both estimated coefficients are positive, pointing at higher donations to nonprofit educational institutions. Only for Massachusetts, however, it is significant at the 10% level. Since both are about equal as regards to economic significance, the difference in statistical significance can mainly be attributed to a lower amount of observations in the D.C. sample, leading to low power of the t-test.
