When government grants crowd out private donations. Anna Bremany October 13, 2008

Abstract This paper employs a previously unexplored panel dataset to test the e¤ect of two di¤erent types of government grants on private donations, controlling for changes in fundraising behavior. The data covers all registered charitable organizations in Sweden between 1989 and 2003. This dataset distinguishes between grants for domestic services and for international aid. The panel data allows us to test the e¤ect of di¤erent types of government grants, while controlling for unobserved organizational heterogeneity and time …xed e¤ects. Furthermore, I use a 2SLS speci…cation to control for possible endogeneity in government grants and fundraising expenditures. Total government grants have a small negative e¤ect on private donations at 7 percent. However, distinguishing between di¤erent types of government grants, I cannot reject full crowding out of government grants for international aid. For organizations targeting foreign recipients, the crowding out is, on average, 102 percent, and signi…cant. The paper provides evidence that government grants are heterogeneous in their e¤ect on private donations and it highlights the importance of …scal transparency as a factor in understanding the crowding out e¤ect.

Acknowledgements: I am grateful to the Swedish Foundation for Fundraising Control for providing the data. Moreover, I would like to thank Sophie Berwick for excellent research assistance. Comments from James Andreoni, Karin Edmark, Magnus Johannesson, Elena Paltseva, Jeremy Tobacman, Björn Tyrefors, Fredrik Wilander and from seminar participants at the Stockholm School of Economics, the 2006 ENTER Jamboree and the iHEA 5th World Congress are gratefully acknowledged. y Department of Economics, University of Arizona, 1130 E. Helen Street, McClelland Hall 401, Tucson, AZ 85721. E-mail: [email protected]

1

1

Introduction

Charitable organizations are typically …nanced by a mix of private contributions and government grants. This is the case in the United States where non-pro…t organizations provide more than 50% of the social services.1 Lately, European countries such as Sweden have seen an increase in the number of charitable organizations receiving government contracts to provide such services (Statskontoret 2004). At the same time, the number of charitable organizations has increased.2 One of the fundamental policy questions in public …nance is therefore how government grants to such organizations a¤ect private contributions. Do government grants crowd-out private donations, and is the relationship a¤ected by the type of government grant? The crowding-out hypothesis says that private givers, who are also taxpayers, will use their tax-…nanced donations as a substitute for their voluntary donations, thus reducing the net e¤ectiveness of grants.3 While theory predicts a one-to-one relationship between government grants and private donations, econometric and experimental studies have found evidence of partial or no crowding out.4 Andreoni and Payne (2003) argue that government grants reduce the organizations’ fundraising e¤orts, which may indirectly cause a decrease in private contributions. In a recent paper (Andreoni and Payne, 2008), the authors provide empirical evidence that reduced fundraising e¤orts explain 68 percent of crowding out, while classic crowding out represents 32 percent. This paper employs a previously unexplored panel data set to estimate the e¤ect of di¤erent types of government grants on private donations, controlling for changes in the organizations’fundraising behavior. The data set is a panel of all registered charities in Sweden between 1989 and 2003. The Swedish data is particularly suited for this 1

See, e.g., Salomon 1990. The data set employed in this paper shows that, in Sweden, the number of registered charities has grown by 146 percent between 1989 and 2003. 3 See, e.g., Warr 1982, 1983; Roberts 1984; Bernheim 1986; and Andreoni 1988. 4 See e.g. Khanna et al. 1995, Payne 1998, Khanna and Sandler 2000, and Okten and Weisbrod 2000. 2

2

purpose. First, it covers all registered charities in Sweden which have fundraising activities targeted towards the public. This eliminates the risk of selection e¤ects and allows us to estimate the e¤ect of government grants on the full population of charities as well as on di¤erent types of charities. Second, to facilitate the public’s trust in the organizations, the charities have to undergo yearly supervision and report detailed …nancial information. The accounting practices are the same for all these charities, which makes the data directly comparable between organizations and over time. Third, government grants are divided into two categories; sida grants allocated primarily to international aid projects, and other government grants, targeting domestic services. This distinction allows us to test whether private donations are a¤ected di¤erently by the two types of grants. There are 361 organizations in our data set which are divided into four categories; (1) health, (2) social services (other than health), (3) international aid and (4) other organizations. The …rst three groups comprise organizations that, for example, support the homeless, the disabled, and conducting foreign aid. The last group mainly consists of charities focusing on protecting the environment or advocating human rights. Although this is a rather diverse group, the common factor is that they are lobbying organizations, rather than social service providers. Typical examples are Amnesty and Greenpeace. To test whether government grants crowd out private donations, we control for changes in the organizations’ fundraising expenditures. We add other time-varying organizational variables as well as organizational …xed e¤ects and year …xed e¤ects. The organizational …xed e¤ects allow us to control for unobserved organizational heterogeneity, such as reputation, while the year …xed e¤ects control for macro economic shocks, political and economic e¤ects. A two-stage least squares speci…cation is employed to control for possible endogeneity in the key explanatory variables government grants and fundraising expenditures. Finally, we perform several robustness tests, such as tests of instrumental validity, over-identi…cation and autocorrelation. There are two main results. First, I …nd that government grants crowd out private 3

donations. The …rst-di¤erenced OLS regressions show a statistically signi…cant negative e¤ect of government grants on private donations for the full population at 2.1 percent. At the disaggregated level, we cannot reject zero crowding out in most speci…cations. The exceptions are for social services organizations and other organizations in the …rstdi¤erenced OLS regressions. For social services organizations, the estimated crowdout is, on average, 2.3%. For other organizations (mainly environmental and human rights), government grants signi…cantly crowd in private donations, on average, by 6.9%. As expected, the …rst-di¤erences OLS estimates are consistently upward biased as compared to the 2SLS results. The 2SLS regressions provide a crowding out estimate of 7.0 percent for the full population and 22.3 percent for health, 92.5 percent for international aid, 5.7 percent for social services and 9.6 percent for other organizations, respectively. The estimates for international aid and social services are statistically signi…cant. Second, di¤erent types of government grants a¤ect private donations di¤erently. The crowding out e¤ect is consistently larger for sida government grants as compared to other government grants. When we pool all organizations, targeting foreign recipients, the estimated crowding out is 102.4 percent for sida grants and statistically signi…cant. These results are interesting for three main reasons. First, it shows that di¤erent types of government grants may a¤ect private donations di¤erently. Second, it is the …rst non-experimental paper …nding full crowding out for a set of organizations. Third, it suggests that the degree of …scal transparency does matter for crowding out. This paper proceeds as follows: Section 2 discusses the results from previous studies. Section 3 describes the data set. Section 4 presents the empirical model for measuring crowding out and section 5 estimates this model. Section 6 provides a brief conclusion.

4

2

Previous literature

The classical model of charitable giving predicts that there is a one-to-one relationship between government grants and private donations. The key assumption in the model is that individuals solely bene…t from their private consumption and the realization of a public good. The intuition is that government grants are indirect private contributions collected through taxes and, as private givers only care about the overall level of the provision of the public good, private donations and government grants should be perfect substitutes.5 There are, however, several reasons why complete crowding out might not be observed in the data. First, if the assumption of neutrality between tax-collected contributions and direct private contribution is relaxed by assuming that individuals may derive private enjoyment from the act of giving so-called “warm-glow giving”, the complete crowd-out may fail (Andreoni 1989, 1990, Ribar and Wilhelm 2002). Ribar and Wilhelm (2002) show that such warm-glow preferences will make the crowding out hypothesis sensitive to the number of donors: a small number of donors (as in an experimental setting) will increase the marginal utility of each individual’s own gift, thus causing complete crowding out. In a setting with many donors, the marginal utility of an individual’s own contribution will be small and the warm-glow e¤ect will dominate biasing crowding-out towards zero. Second, complete crowding out is also based on the assumption of …scal transparency. Eckel et al (2005) argue that donors/taxpayers might not understand the sources and opportunity costs of funding for charities they support, which will cause complete crowding out to fail. In their study, they implement an experiment with two di¤erent settings, which are identical except for the framing. Consistent with the crowding-out hypothesis, they report almost complete crowding out when subjects are told that they are “taxed” with contributions to charitable organizations, but partial crowding out otherwise. 5

See e.g. Warr 1982, 1983, Roberts 1984, Bernheim 1986 and Andreoni 1988.

5

Third, government grants can be seen as a signal that a charity is reliable. If the government is perceived as having superior information, this endorsement e¤ect might cause government grants to crowd in private donations (Rose-Ackerman 1982). The crowding-in hypothesis has not been directly tested using the government as a third party donor, but it has been experimentally tested using private third-party donors (List and Lucking-Reily 2002, Vesterlund 2003, Potters, Sefton and Vesterlund 2005). These studies provide strong evidence of an endorsement e¤ect. A number of experimental and empirical studies have been conducted to test the crowding-out hypothesis. The experimental setting has been that of a public goods experiment. Andreoni (1993) was the …rst to perform such an experiment …nding evidence of partial crowding out. This result was con…rmed in a later experiment by Bolton and Katok (1998). Early empirical literature relied on income tax returns or expenditure surveys as their source of private donations (see Payne 1998 for an excellent survey of these studies). There are several problems in using income tax returns; inability to match the source of government spending with the private donations, and inability to control for organizational-speci…c heterogeneity and behavior. Therefore, more recent empirical studies have used data at the organizational level, which allows them to match private donations to government grants given to the same organization. Furthermore, panel data has made it possible to control for unobserved organizational heterogeneity and time …xed e¤ects. Khanna et al (1995), the …rst to employ a panel data set, used a 7-year panel of 159 charities in the U.K., …nding no evidence of crowding out. In a more recent contribution using the same U.K. data, Khanna and Sandler (2000) improve upon their previous study by applying 2SLS to control for possible endogeneity of government grants. Once more, they …nd evidence of crowding in rather than crowding out for the full sample. Payne (1998) exploits a dataset on 430 American non-pro…t organizations between 1982 and 1992, and also uses a 2SLS speci…cation to control for possible endogeneity of government grants. The estimated crowding out was signi…cantly di¤erent from zero, and a one dollar increase in gov6

ernment grants, on average, crowded-out private donations by about 50 cents. Okten and Weisbrod (2000), with data on American non-pro…t organizations, …nd that government grants do generally not crowd out private donations. On the contrary, in most industries, there are signi…cant positive e¤ects. No crowding out is equally the result in Ribar and Wilhelm (2002) who use a panel for 1986-93 of 125 international relief and development organizations in the United States. On the other hand, Gruber and Hungerman (2005) …nd evidence of crowding out in a study on faith-based charity during the great depression. The estimated crowd-out was small as a share of total New Deal spending (3%), but large as a share of church spending (30%). Note that among the above mentioned empirical studies, only Khanna et al (1995), Khanna and Sandler (2000) and Okten and Weisbrod (2000) include fundraising expenditures as an explanatory variable, but none of them consider endogeneity in both government grants and fundraising expenditures. Andreoni and Payne (2003) argue that the incomplete crowding out observed in most empirical and experimental literature is due to a change in the behavior of the organization, rather than to a change in the behavior of private givers. Using data on American charities, they …nd that government grants signi…cantly reduce the organizations’ fundraising activities. Andreoni and Payne (2008) estimate that 68 percent of crowding out is due to reduced fundraising and only 32 percent from classic crowding out. In total they …nd that crowding out is about 56 percent. Straub (2004) takes into account the possible e¤ect of government grants on fundraising activities when estimating crowding out for a set of public radio stations in the U.S. The advantage of the public radio data is that it matches household- and …rm-level information. The hypothesis that crowd-out is dollar-for-dollar is rejected while the hypothesis that crowd-out is zero cannot be rejected. The study uses cross-sectional data, making it di¢ cult to control for unobserved …rm-level heterogeneity, and it is restricted to one sector, thereby reducing the possibilities of drawing any conclusions about other types of charitable organizations. The Swedish data has the advantage of being a panel covering all types of charitable 7

organizations. All organizations are publicly scrutinized each year, which requires them to report government grants, private donations as well as fundraising expenditures. The empirical speci…cation takes into account changes in fundraising behavior. The aim is to estimate the total e¤ect of di¤erent government grants on charitable contribution, and not separate out the e¤ect of classic crowding out from crowding out from reduced fundraising e¤orts. The following section presents the data set.

3

The SFI Dataset

The data on charitable organizations comes from the Swedish Foundation for Fundraising Control (SFI)6 . The data covers all registered charities in Sweden that have fundraising activities targeted towards private givers. I have collected data for the period 1989 to 2003, which gives us a panel of 361 organizations over 15 years. Since the 1940s, Sweden has a system where bank accounts starting with the number 90 are exclusively used for public fundraising by charitable organizations. To receive a so-called 90-account, the organizations must be approved by SFI and thereafter, they have to follow the SFI rules regarding fundraising and undergo yearly supervision. If an organization does not ful…ll the statues, norms and guidelines of SFI, the 90account is immediately withdrawn. In Sweden, the 90-account is seen as a guarantee of the seriousness of charitable organizations. The bene…t of this system is thus that it increases the public’s trust in the work of charitable organizations and reduces the risk of fraud. It also facilitates comparisons between organizations and hence, renders it a more reliable dataset for the researcher. In 2004, these organizations received a total of SEK 4.1 billion from private donors, which is equivalent to approximately SEK 450 ('USD 60) per person. The equivalent number for the United States is considerably higher. There are several reasons for the 6

See http://www.insamlingskontroll.a.se/. The Swedish name is ”Stiftelsen för Insamlingskontroll” (SFI). This non-pro…t organization is …nanced through administrative fee taken from the charitable organizations entitled to use a 90-account. The board of SFI includes representatives from the government, private companies and the unions.

8

relatively low donations per person in Sweden. Most importantly, charitable contributions are not tax deductible in Sweden. In combination with a high overall tax burden, social services have primarily been the responsibility of the state. Still, a large share of the population do contribute to charitable organizations each year. The data also reveals several di¤erences between the two countries in the pattern of giving. In Sweden, almost 40 percent of charitable contributions target foreign aid, while the equivalent number in the United States is two percent (Giving USA 2006). Moreover, neither education nor religious organizations are de…ned as separate categories in the Swedish statistics. Universities that receive considerable charitable contributions in the United States are primarily …nanced by taxation in Sweden. Private contributions to universities are virtually non-existent. Approximately 19 percent of all organizations in Sweden have a religious a¢ liation. From table 1, we see that the religious organizations are mainly found in the categories "social services" and "other" organizations. Religious organizations are also overrepresented among those targeting foreign recipients, re‡ecting the fact that many religious organizations carry out missionary work. The data in earlier studies faced several measurement problems (see, e.g., Andreoni and Payne, 2003, for an overview). There is a number of ways in which this dataset improves upon the previous literature.

1. It covers all registered charities in Sweden. Previous studies have used subgroups of charities within a given country. There is no selection bias in the Swedish dataset due to the type of charities investigated. 2. The data is directly comparable between organizations and over time. There are no diverging accounting practices between the organizations; SFI has detailed guidelines on how to report on the economic activities of the organization. All economic reports are controlled by a certi…ed auditor according to SFI’s speci…c directives. 9

3. The data distinguished between two types of government grants; grants targeting domestic services, and grants targeting international aid. The di¤erences between the two types of grants are described in more detail below. 4. There are more than 20 …nancial variables in the dataset. The richness of the dataset allows us to control for organization-speci…c variables that may in‡uence the behavior of the charity itself as well as private givers and government decision makers. Such variables include …nancial assets, membership fees, debts, etc. 5. Program service revenues are included in government grants. One of the major di¢ culties in the earlier studies using U.S. data is that program service revenue is not included in government grants. Program service revenue is payment for speci…c services performed by a charity under a government contract. It is thus taxpayers’ money and should cause crowding out to the same extent as other government grants. 6. Fundraising expenditures are only used towards the collection of private donations and there is a clear separation between fundraising costs and administrative costs. In the U.S. data, it is unclear whether fundraising may also include the costs associated with applying for government funding or even the costs of reporting and complying with the grants. In this dataset, this problem does not arise since it is clearly speci…ed in the instructions to the organizations that fundraising expenditures should show the “costs that the organization has incurred to collect the donations accounted for from the public”.7

Although the dataset comes to terms with several of the potential sources of measurement errors in earlier studies, it is not without problems. The major problem of this dataset is the fact that government expenditures can include donations from other charitable organizations. This is done for accounting purposes, however. The 7

SFI’s accounting guidelines 2002.

10

headquarters of large organizations with regional o¢ ces generally receive some of the money collected regionally. To avoid that these private contributions are counted twice (…rst at the regional level and then at the national level), thus overstating total private contributions, these donations are added to government grants instead of private contributions. For the purpose of our study, this accounting practice will overstate the total amount of government grants. Therefore, we exclude those organizations representing headquarters of large organizations with regional o¢ ces (for example, the Red cross, the Salvation Army). Moreover, this dataset shares the disadvantage of the U.S. datasets that there are many zeros reported in the measurements of interest. Some data cleaning is therefore necessary. Organizations are excluded according to these (sequentially followed) rules8 :

1. All organizations with two or fewer years of observations in the sample (77 organizations, 107 observations). 2. All organizations that are national headquarters of regional organizations (6 organizations, 82 observations). 3. All organizations that report zero government grants for all years for which the organization is in the sample (103 organizations, 835 observations).

This leaves 2503 observations and 252 organizations, where 78 organizations (845 observations) are health-related, 51 (515) social services, 70 (652) international aid, and 53 (491) other organizations. Health-related are those organizations delivering health services or supporting health-related research, such as The Swedish Cancer Society9 . The social services organizations are those delivering social services such as food and shelter. In this group, we …nd organizations such as the Red Cross and Stockholm’s 8

These are the only criteria by which organizations are deleted. No outliers or otherwise strange observations have been observed or deleted. 9 Cancerfonden.

11

City Mission (where the latter primarily targets the homeless). The international aid organizations work in many …elds such as education, health and humanitarian aid. For example, SOS Children’s Villages …nance housing, education, food and health care for orphans in developing countries. The last category "Other" organizations mainly consists of environmental and human rights organizations including, among others, Greenpeace, the World Wide Fund for Nature (WWF), Amnesty and the Swedish Peace and Arbitration Society10 .

3.1

The two types of government grants

The accounting guidelines of the Swedish Organization for Fundraising Control stipulate that the organizations distinguish between (1) government grants administrated by the Swedish development agency, Sida, and (2) other government grants. Sida administrated government grants are given to international aid projects, while other government grants primarily target domestic services. Unlike international aid, there is not one agency, but several government agencies that administrates the domestic grants. International aid has a long tradition in Sweden. The o¢ cial government policy has been that one percent of gross domestic income should go to foreign aid, a high percentage in international comparison. However, due to budgetary restrictions, the actual contributions to foreign aid has varied between 0.7 to 1.0 percent of GDI in the past 20 years. Two features are important here. First, the government target is well-known by the general public. When the foreign aid budget is cut or expanded, it will be noted by the media. In contrast, other government grants focusing on domestic services are not well known nor well publicized. The allocation of money to domestic social services is less clear and less visible to the public. Table 2 provides the number of organizations (number of observations within parentheses) where the recipients of aid are domestic or foreign or both. The majority of 10

Svenska Freds- och Skiljedomsföreningen.

12

organizations are domestically oriented. However, 39 percent of the charities target their funds to foreign or both foreign and domestic recipients. The organizations targeting foreign recipients dominate among international aid organizations and other organizations. Health and social services organizations are primarily focusing on domestic recipients or both domestic and foreign recipients. Table 3 reports the relative importance of the two types of government grants, total private donations and total fundraising expenditures. The sums reported are adjusted for in‡ation, using the Consumer Price Index provided by Statistics Sweden11 . 1990 is the base year. International aid organizations receive 87 percent of total government grants as Sida grants, while the equivalent number for health organizations is 3 percent, social services organizations 8.6 percent and other organizations 81 percent. It is also worth noting that international aid organizations and other organizations receives the most government grants per organization and year, with SEK 11.2 million and SEK 9.6 million respectively. In total, almost 55 percent of total government grants are Sidaallocated government grants. Hence, even though a minority of the organizations target foreign recipients, they receive a majority of government grants. Finally, graph 1 shows the yearly trend in total expenditures for our key variables government grants, private donations and fundraising costs. Since fundraising expenditure is considerably lower than private donations and government grants, it is measured by the scale on the right-hand side of the graph, and the other two are measured on the left-hand side. While private donations and fundraising expenditures have increased over time, government grants have remained at a more constant level and even decreased in some years. From this graph alone, we cannot see any clear indication that private donations have declined when government grants have increased or vice versa. The following section gives the empirical speci…cation used to test the crowding-out hypothesis. 11

The Consumer Price Index is available at Statistics Sweden’s website www.scb.se.

13

4

Empirical Speci…cation

The data matches private donations and government grants organization by organization. All data is reported on a yearly basis, and includes other organizational speci…c characteristics that vary over time which might in‡uence private donations, such as fundraising expenditures and membership fees. Several U.S. studies (see e.g. Payne 1998, Payne and Andreoni 2003) focus on statelevel charities and therefore include time-varying economic and political indicators at the state level. However, this approach is not suitable for the Swedish context for the following reasons. First, Sweden is a small homogenous country and most charities are nation-wide, targeting the entire population in their fundraising e¤orts12 . Second, regional income di¤erences are relatively small in Sweden due to extensive redistribution at the national level. We will lose many organizations if we only include those speci…c to a certain region, and little will be picked up due to small variations in regional variables. The following empirical model tests the crowding-out hypothesis:

Pit =

+ Govit + Fit + Zit +

i

+

t

+ "it ;

(1)

where Pit denotes private donations received by charitable organization i at time t. Govit is the government grants received by the charitable organization and Fit is fundraising expenditures by the charitable organization, directed at private donations. Zit represents the vector of other revenue and/or expenditure measures at the organizational level.

i

and

t

represent the unobserved heterogeneity at the organization-

and time-level, respectively. The crowd-out parameter is measured by , i.e. the coe¢ cient on government grants. We estimate this speci…cation with ordinary least-squares (OLS) and two-stage least-squares (2SLS). Since the data is a panel dataset, I include organization and year …xed e¤ects. The organization …xed e¤ects are incorporated to 12

The populations of Sweden is approximately 9.2 million.

14

control for time-invariant heterogeneity in the charities, such as their reputation, age and/or method of operation that may a¤ect the collection of government grants. The year-…xed e¤ects capture the macro-level time varying shocks that a¤ect all charities similarly, such as nation-wide economic shocks and political measures. In addition to the above measures, we included various measures to help control for time-varying changes at the organizational level. For all organizations, we include fundraising expenditures, membership fees, and the charitable output . It has been shown that organizations might devote fewer resources to fundraising after receiving government grants (Andreoni and Payne 2003). Therefore, we are particularly interested in fundraising expenditures. An organization’s fundraising expenditures include salary costs and marketing expenditures directly targeted at increasing private donations. We can test whether the organizations are net revenue maximizers, i.e. whether a one-unit increase in fundraising expenditures increases private donations by one unit. The coe¢ cient on fundraising expenditures should then be equal to one. If it is larger (smaller) than one, less (more) than the optimal amount is spent on fundraising. Moreover, we include membership fees in the regressions. Fundraising, particularly in smaller organizations, is often done on a voluntary basis. Membership fees can be seen as a measure of the amount of voluntary activity within an organization. Including this measure helps control for voluntary activities not included in fundraising expenditures. On the one hand, it can thus be expected to increase private donations, while on the other hand, membership fees might crowd out other private giving by volunteers who consider their free time as their main contribution. The expected net e¤ect of membership fees on private donations is therefore uncertain. Finally, we include the variable "e¢ ciency in charitable output" which is de…ned as

P = 1=[1

(f + a)];

(2)

where the denominator denotes the proportion of the charity’s total expenditure spent 15

on charitable output (Weisbrod and Dominguez 1986). In equation (2), f and a denote the proportions of total expenditure used for fundraising and administration, respectively. Thus, the variable e¢ ciency in charitable output shows how much a donor has to give in order for the charitable cause to receive SEK 1. The less e¢ cient the organization is, the less donors should be willing to contribute. In the tables of results, I will refer to this variable as "cost" to emphasize that we expect the variable to have a negative sign. Unlike the U.S. and the U.K., charitable contributions are not tax deductible in Sweden (neither for private individuals nor for companies). Therefore, we can disregard the marginal tax rate in the "cost" variable. We take the …rst-di¤erences of equation (1) to eliminate unobserved heterogeneity at the organizational level. The …rst-di¤erence estimator rather than the withinestimator is chosen, since …rst-di¤erencing is more appropriate if we want to use lagged variables as instruments in the 2SLS speci…cations (Wooldridge, 2002). The empirical speci…cation is thus

Pit =

where

Pit = Pit

Govit +

Fit +

Pit 1 ; Govit = Govit

Govit

Zit +

1

"it ;

(3)

and so on.

We …rst estimate the total e¤ect of government grants on private donations for the full population of charities and by category. First, using the …rst-di¤erenced OLS estimator and second using the 2SLS estimator. The third step is to measure the e¤ect of the two types of government grants separately. Here, we use the 2SLS speci…cation and focus on the (1) full sample of organizations, (2) international aid organizations, and (3) all organizations targeting foreign recipients13 . 13

The other categories have few grants for international aid. That makes it di¢ cult to estimate the e¤ect of the two types of government grants seperately. Regressions are available from the author upon request.

16

5

Estimation

In estimating equation (3), there are two potential problems; serial correlation and endogeneity. Serial correlation in the error process is a common problem in panel data and, if present, the standard errors can be severely biased (see Kézdi, 2003). We perform the Wooldridge test for autocorrelation in panel data (Drukker, 2003; Wooldridge, 2003, p. 283). If there is no serial correlation in the original dataset, taking …rst di¤erences causes serial correlation in the error terms equal to -0.5. Therefore, we can take the …rst-di¤erences and test whether the correlation in the error terms is equal to -0.5. The null is that there is no serial correlation. No serial correlation can be rejected (p = 0.0001). A reasonable assumption is that the error terms are serially correlated within each organization, but not in-between organizations. Therefore, we use robust standard errors controlling for serial correlation within each organization, but not in-between di¤erent organizations. In addition, the calculated standard errors are robust to arbitrary heteroskedasticity. Furthermore, endogeneity might arise for two reasons. First, private donations and government grants might be jointly determined. For example, after a catastrophe such as a hurricane, the services of a social services organization (e.g. the Red Cross) are in high demand. Contributions are thus likely to be sought from both private givers and government donors. In other words, unmeasured in‡uences, which only in‡uence a subset of organizations, may be increasing both government grants and private donations, and the measure of crowd-out would then be positively biased. Second, we might encounter a problem of reversed causality between private donations and fundraising expenditures. Fundraising expenditures are targeted to increase private donations, but low private contributions in one year can cause fundraising expenditures to increase within the same year. The measure of fundraising expenditures would then be negatively biased. To correct for this possible endogeneity of government grants and fundraising expenditures, I choose to use instrumental variables. I use four set of instruments to 17

control for possible endogeneity in government grants and fundraising expenditures; repayments, administrative costs, …nancial assets and material assets. The set of instruments relevant for government grants are repayments of government grants lagged one year, and administrative costs. Repayments of government grants are made if an organization has not used the grants in year t-1. It is then legally obliged to repay the unused money to the government (i.e. positive repayments), and it is likely to receive a smaller amount in the following year. Repayments are therefore included as an instrument for government grants in the subsequent year. Administrative costs are associated with applying for and complying with government grants and that variable is therefore included as a third instrument. These instruments are thus relevant. Are they also exogenous? Repayments of government grants are not publicly available information. The organizations have no obligation to report repayments and it is not made available via the Swedish Foundation for Fundraising Control. This implies that lagged repayments of government grants are uncorrelated with the error term and thus exogenous. Administrative costs are associated with government grants. There is a clear distinction between fundraising costs that are associated with private donations and administrative costs that are associated with applying for and complying with government grants. Administrative costs can thus be expected to be uncorrelated with the error term14 . The set of instruments relevant for fundraising expenditures are …nancial assets and material assets. Financial assets and material assets are a measure of the wealth of the organization and are therefore included as an instrument for fundraising expenditures. The material and …nancial assets included as instruments are inventories, stocks, buildings and total material and …nancial assets. The higher the wealth, the more can be spent on fundraising. Material and …nancial assets can be expected to be exogenous. First, charitable donations are not tax deductible in Sweden and are therefore not as14

The price of giving which can be expected to in‡uence private donations is fundraising and administrative costs measured as the share of total expenditures. However, the correlation between administrative costs and the price of giving is close to zero (.005). The reason is that the relative shares of fundraising and administrative costs vary greatly between di¤erent charities and over time.

18

sociated with …nancial planning. Second, information on material and …nancial assets is not available via the Swedish Foundation for Fundraising Control. The general overall economic environment may a¤ect private donations (the stock market and house prices) as well as the …nancial and material wealth of charities. However, the general economic environment is controlled for by year …xed e¤ect and the instruments capture organizational speci…c …nancial and material assets. Thus, it should not in‡uence private donations. Material and …nancial assets are therefore uncorrelated with the error term and valid as instruments for fund-raising expenditures. In this section, I …rst report results without controlling for endogeneity, and then turn to results using a two-stage least-squares regression methodology. In both the OLS and the 2SLS regressions, I include organization and year …xed e¤ects. Robust standard errors are used in all speci…cations.

5.1

First-di¤erenced OLS Regressions

Table 5 reports the results from the …rst-di¤erenced OLS regressions for the full sample of organizations. All variables are in thousand SEK15 (in‡ation-adjusted with 1990 as the baseline year). Two parameters are of main interest, the crowding out parameter and the fundraising parameter : If government grants crowd out fundraising rather than private donations, we expect to …nd zero crowding out ( = 0) after controlling for changes in fundraising expenditures. Furthermore, I want to test whether organizations are net revenue maximizers, i.e. whether

= 1:

For the full sample, I can reject no crowding out, i.e.

= 0: There is a small, but

signi…cant e¤ect of government grants on private donations. On average, a SEK 1000 increase in government grants decreases private donations by SEK 21, after controlling for both year and organization …xed e¤ects. I cannot reject that organizations are net revenue maximizers. Fundraising has a positive and signi…cant e¤ect on private donations; the coe¢ cient is close to 1 which indicates that charities, on average, are 15

1 USD ' 7 SEK.

19

net revenue maximizers, spending up to the point where the marginal e¤ect of SEK 1 extra spent on fundraising gives SEK 1 in return. Table 6 presents the results for the …rst-di¤erenced OLS regressions for di¤erent types of organizations. For health and international aid organizations, zero crowding out cannot be rejected. For social services and other, no crowding out can be rejected. The measures of crowding out are negative and signi…cant for social services organizations (2.3%) and positive and signi…cant for the category other organizations (6.9%), indicating that government grants crowd in private donations. Furthermore, the e¤ect of fundraising is positive, highly signi…cant and close to one, indicating net revenue maximization for health, international aid and other organizations. For social services organizations, I can reject net revenue maximization. The coe¢ cient on fundraising expenditures is less than one, indicating that social services organizations overspend on fundraising activities. Finally, we note that the variable "e¢ ciency in charitable output" has a large negative impact on private donations, except for the category other organizations. Membership fees enter with a positive sign for health and social services organizations, while it has a small negative impact for international aid and other organizations. These estimates of government grants and fundraising expenditures could potentially be biased due to endogeneity. If endogeneity is present, we would expect the measure of crowd-out would then be positively biased in the OLS regression, while the measure of fundraising expenditures would be negatively biased. Therefore, we turn to a 2SLS speci…cation.

5.2

Two-Stage Least-Squares Regressions

Tables 7 and 8 report the results from the 2SLS …rst-di¤erenced regressions, using total government grants and total fundraising. It is clear from this table that including the instruments provide evidence that the OLS regressions consistently caused crowding out to be biased upwards and fundraising expenditures to be biased downwards. For 20

the full sample, and for each category, the estimated crowd-out is larger than in the …rst-di¤erenced OLS regressions. For the full sample, the estimated crowding out has increased from 2.1 percent to 7.0 percent, but the e¤ect is not statistically signi…cant in the 2SLS estimation. In table 8, we report the 2SLS …rst-di¤erenced estimates for the four types of organizations. The health, international aid, social services show crowding out at 22.3, 92.5, 5.7 and 9.6 percent, respectively. The estimated crowding out signi…cant for international aid and social services organizations. Moreover, for the full sample, the point estimate of fundraising is now larger at 1.45 and highly signi…cant, indicating that the organizations, on average, fall short of net revenue maximizing. For health, social services and other organizations, the 2SLS shows that fundraising expenditures were biased downward, and I can now reject net revenue maximization for health and other organizations. The exception is international aid organizations, but the estimate is far from signi…cant. The most remarkable result is the high level of crowding out for international aid organizations and the change from crowding in to a signi…cant crowding out e¤ect for the category other organizations. This leads us to believe that those organizations are potentially di¤erent from the others. I therefore estimate the e¤ect of the two types of government grants separately, focusing on international aid and other organizations. The two types of government grants are (1) Sida government grants (targeting foreign aid projects), and (2) other government grants (mainly focused on domestic services). The international aid organizations are predominantly …nanced by Sida grants, but also some organizations in the other organization category receives substantial grants from Sida, while health and social services organization receives very small amounts of Sida grants. I therefore run the 2SLS speci…cation for (1) international aid organizations (69 organizations), (2) other organizations (53 organizations), and (3) all organizations who solely target foreign recipients (88 organizations - all international aid organization and a subset of other organizations). The instruments are the same as before. Table 916 presents the results of the 2SLS regressions estimated for sida grants 16

Table 10 reports the results for health, social services and all organizations, from the 2SLS regres-

21

and other grants separately. Consistently for the tree di¤erent speci…cations are that sida grants have a negative impact on private donations, while other government grants have a positive impact. The strongest results is found in the sample of 88 organizations targeting solely foreign aid recipients. The estimated crowding out of sida grants is 102 percent and statistically signi…cant, while the e¤ect of other government grants is positive, but not statistically signi…cant. Why would sida grants have cause complete crowding out of private donations, while other government grants do not? Government grants can cause crowding out via two channels; (1) a direct negative e¤ect on private donations, (2) an indirect e¤ect via reduced fundraising e¤orts. There is no obvious reason why sida government grants and other government grants should a¤ect the organizations’ fundraising behavior di¤erently. Private donors, on the other hand, are likely to be in‡uenced by the information available on government grants. As described in chapter 3, sida government grants are well-publicized and coordinated by one government agency, while other government grants are administrated by several di¤erent agencies and no coherent information is easily available about the amounts or sources of those grants. Those year that the government increase the budget for international aid, i.e. the share of taxes designated for foreign aid increases, the public might respond by decreasing private donations to international aid organizations, and conversely increase contributions in years when the budget is cut. Since other government grants are less known to the general public and the allocation of these funds is less clear, the crowding out e¤ect of other government grants is less pronounced. Therefore, it is reasonable that sida grants have a larger impact on private donations than other types of government grants. The results are in line with Eckel et al. (2005) who show that …scal transparency matter for crowding out in the laboratory. sions estimated for sida grants and other government grants separately. These results are less precise as a very small share of total government grants are sida grants for these types of organizations. Still, for all organizations and for social services organizations, sida government grants have a larger negative impact as compared to other government grants. Health organizations reveal no e¤ect of sida grants on private donations, but a mere three percent of total government grants are sida grants in this category.

22

Ideally, we would like to separate out the e¤ect of government grants on fundraising e¤orts of the organization and the direct e¤ect of reduced private donations. However, the sample size is limiting such estimations. Considering that we are already using the full population of charitable organizations within the country, it is not possible to collect a larger dataset. The main contribution of this paper is therefore to show that di¤erent types of government grants may have diverging e¤ects on private donations and that government grants can have a larger negative e¤ect than previously found in the empirical literature.

5.3

Robustness

There are two potential problems with instrumenting; …rst, the instrument might be weak, which can cause biased estimates. Second, government grants and fundraising expenditures might not be endogenous variables in the model. Therefore, I perform several robustness tests. To assess the instruments, I carry out the Hansen-Sargan test for over-identifying restrictions. The null is that instruments are valid instruments, i.e. uncorrelated with the error term, and that the excluded instruments are correctly excluded from the estimated equation. A rejection of the null casts doubt on the validity of the instruments. I cannot reject the null that the instruments are valid in any speci…cation. The second test of instrumental validity is an F-test on the instruments in the …rst-stage regression. In tables 7 and 8 when we test the e¤ect of total government grants, the instruments are strong in all regressions for total government grants and for fundraising expenditures, except for total government grants for international aid organizations. In tables 9 and 10 when we estimate the e¤ect of sida and other government grants separately, the instruments also perform well for sida grants, other government grants and fundraising expenditures. The instruments are somewhat weak for fundraising expenditure for international aid and for other government grants for all organizations targeting foreign recipients. 23

Finally, I want to test if government grants and fundraising expenditures are endogenous. If they are not, the …rst-di¤erenced OLS regressions are more appropriate, and we should base our conclusion on those estimates. The assumption necessary to carry out a Hausman test of endogeneity is that the instruments are exogenous. The null hypothesis in the Hausman test is that government grants and fundraising expenditures are exogenous. The results in tables 7 and 8 suggest that I cannot reject exogeneity of total government grants and fundraising expenditures for the full sample (p = 0.44) nor for any of the sub-categories, except from other organizations (p = .00). In table 9, when we estimate the e¤ect of sida grants and other government grants separately, we can reject exogeneity in all three speci…cations. International aid organizations and other organizations are also the two categories where we …nd the largest di¤erence between the OLS and the 2SLS estimates. The robustness analysis indicates that it is valuable to estimate the e¤ect of di¤erent types of government grants separately. In those estimations, exogeneity is rejected for international aid organizations and other organizations. These are the categories where there is a substantial share of both sida grants and other government grants, and the di¤erences between the OLS estimates and 2SLS estimates are large. This seems intuitive, considering the origin of endogeneity in this sample. For government grants to be endogenous, there should be an unobserved event that simultaneously a¤ects private donations and fundraising expenditures for a subset of organizations within one category, and is therefore not controlled for by organizational …xed e¤ects or time …xed e¤ects. Typical examples would be a hurricane, a famine or a natural disaster. These types of events are the most likely to in‡uence the international aid organizations or the environmental organizations (included in the category "other"). It is hard to …nd similar events that would cause an endogeneity problem for health organizations, or for domestic social services organizations. Second, in the full sample, we can expect unobserved events to cause an endogeneity problem, but if these events are rare, the e¤ect is likely to be small. I do …nd that the crowding out estimate is upward biased in the OLS compared to the 2SLS regression, 24

but the bias is small. This reinforces the interpretation that endogeneity is a problem for some types of organizations, and that the e¤ect is not su¢ ciently large to have a substantial impact in the full sample.

6

Conclusions

Government grants to charitable organizations can crowd-out private donations for two reasons. First, the classic crowding-out hypothesis says that donors let their involuntary tax contributions substitute for their voluntary contributions. Second, the strategic response of the charitable organization on receiving a government grant will be to reduce its fundraising e¤ort; thus, indirectly reducing private donations. This paper estimates the e¤ect of two di¤erent types of government grants on private donations, controlling for changes in fundraising behavior, using a new panel dataset covering all registered charities in Sweden over 15 years. Government grants are divided into two categories; sida government grants targeting international aid projects, and other government grants, primarily targeting domestic services. Sida government grants are administrated by the Swedish development agency and the overall budget is public knowledge, while other government grants are administrated through several government agencies and information on the amount and allocations of these grants are not readily available. There are three main …ndings in this paper. First, I …nd that government grants crowd out private donations. The crowding out e¤ect is consistently upward biased in the …rst-di¤erenced OLS estimates as compared to the 2SLS results. The 2SLS regressions provide a crowding out estimate of 7.0 percent for the full population and 22.3 percent for health, 92.5 percent for international aid, 5.7 percent for social services and 9.6 percent for other organizations, respectively. The estimates for international aid and social services are statistically signi…cant. Second, di¤erent types of government grants a¤ect private donations di¤erently. The crowding out e¤ect is larger for sida government grants as compared to other

25

government grants for the full sample of organizations as well as for international aid, social services and other organizations. Third, I cannot reject full crowding out for sida government grants for organizations targeting foreign recipients. The estimated crowding out of sida grants is 102.4 percent and statistically signi…cant. What do our results suggest for future research? First, both the restrictions following government grants and the public’s knowledge about the size and sources of government grants might di¤er for the various types of charities. In line with previous laboratory experiments, the results indicate that …scal transparency may impact the e¤ect of government grants on private donations. The observed crowding out is likely to be caused by reduced private donations, but also reduced fundraising e¤ort from the organizations, which indirectly will a¤ect private donations. The data in this paper do not allow us to further investigate the organizations responses to di¤erent types of government grants. Further research with more data could greatly contribute to our understanding of the organizations’behavioral response to di¤erent sources of government support. What do our results imply for policy? In contrast to previous empirical results, the hypothesis that government grants crowd out private donations one-to-one cannot be rejected for some types of government grants. This paper provides evidence that government grants are heterogeneous in their e¤ect on private donations, and it highlights the importance of …scal transparency as a factor in understanding the crowding out e¤ect.

References [1] Andreoni, J. 1989. “Giving with Impure Altruism: Applications to Charity and Ricardian Equivalence”. Journal of Political Economy. 97(6): 1447-1458.

26

[2] Andreoni, J. 1990. "Impure Altruism and Donations to Public Goods: A Theory of Warm-Glow Giving". The Economic Journal. 100(401): 464-477. [3] Andreoni, J. 1993. “An Experimental Test of the Public-Goods Crowding Out Hypothesis”. American Economic Review 83(5): 1317-1327. [4] Andreoni, J., and A.A. Payne. 2003. “Do Government Grants to Private Charities Crowd Out Giving or Fund-raising?”American Economic Review 93(3): 792-811. [5] Andreoni J. (2004). “Philanthropy” in Handbook of Giving, Reciprocity and Altruism. Edited by L.-A. Gerard-Varet, Serge-Chritsophe Kolm and Jean Mercier Ythier. Elsevier/North-Holland. [6] Bernheim, Douglas B. (1986). “On the Voluntary Provision of Public Goods”. American Economic Review 76(4): 789-793. [7] Bolton, Gary E. and Elena Katok. (1998). “An experimental test of the crowding out hypothesis: The nature of bene…cent behaviour”. Journal of Economic Behavior and Organization 37: 315-331. [8] Drukker, D.M. (2003). “Testing for serial correlation in linear panel-data models” Stata Journal (3)2: 168-177. [9] Eckel, C.C., Grossman, P.J. and M.R. Johnston (2005). “An Experimental Test of the Crowding Out Hypothesis”. Journal of Public Economics, 89: 1543-1560. [10] Gruber, Jonathan and Daniel M. Hungerman. 2005. "Faith-Based Charity and Crowd Out During the Great Depression". NBER Working Paper 11332. [11] Kézdi, Gábor. 2003. “Robust Standard Error Estimation in Fixed-E¤ects Panel Models”. Mimeo. Budapest University of Economics, IE/HAS, and CEU. [12] Khanna, J., J. Posnett and T. Sandler. 1995. “Charity donations in the UK: New evidence based on panel data”. Journal of Public Economics. 56: 257-272. 27

[13] Khanna, J., J. Posnett and T. Sandler. 2000. “Partner sin giving: The crowding-in e¤ects of UK government grants”. European Economic Review. 44: 1543-1556. [14] List, John A. and Lucking-Reilly. 2002. JPE [15] Okten, Cagla and Burton A. Weisbrod. 2000. “Determinants of donations in private non-pro…t markets”. Journal of Public Economics. 75: 225-272 [16] Payne, A. A. (1998). “Does the government crowd-out private donations? New evidence from a sample of nonpro…t …rms”. Journal of Public Economics. 69: 323345. [17] Potters, Jan et al (2005). "After You - Endogenous Sequencing in Voluntary Contribution Games," Journal of Public Economics, August 2005, 1399-1419. [18] Ribar, David C. and Mark O. Wilhelm. (2002) “Altruistic and Joy-of-Giving Motivations in Charitable Behavior”. Journal of Political Economy. 110(2): 425-457. [19] Rose-Ackerman, S. 1982. "Charitable Giving and Excessive Fund-raising". Quarterly Journal of Economics, 97:193-212. [20] Salomon, L.M. 1990. The non-pro…t sector and the government: the American experience in theory and practice. In: Anheier, H., Wolfgang, S. (Eds.), The Third Sector: Comparative Studies of Nonpro…t Organizations. Walter de Gruyter, New York, pp. 219-240. [21] Statskontoret. 2004. Bidrag till idéella organisationer. Kartläggning, analys och rekommendationer. (Report from the Swedish Agency for Public Management to the Swedish government on charitable organizations) [22] Straub, John D. (2004). Fundraising and Crowd-out of Charitable Contributions: New Evidence from Contributions to Public Radio”. Mimeo Texas A&M University.

28

[23] Vesterlund, Lise. 2003. "The Informational Value of Sequential Fundraising." Journal of Public Economics, March 2003. [24] Warr, Peter G. (1982). “Pareto optimal Redistribution and Private Charity”. Journal of Public Economics. 19: 131-138. [25] Warr, Peter G. (1983). “The Private Provision of a Public Good is Independent of the Distribution of Income”. Economic Letters. 13: 207-211. [26] Woolridge, J.M. (2002), Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press

29

Table 1: Cause and Religion Religious Class No Yes Health 77 1 (840) (5) International aid 54 16 (465) (187) Social Services 36 15 (347) (168) Other 38 15 (315) (166) Total 205 47 (1977) (526)

Total 78 (845) 70 (652) 51 (515) 53 (491) 252 (2503)

Table 2: Cause and Targeted Recipients Recipient Class Domestic Foreign Both Health 78 0 0 (845) (0) (0) International aid 0 70 0 (0) (652) (0) Social Services 45 0 6 (463) (0) (52) Other 30 17 6 (218) (200) (73) Total 153 87 12 (1526) (852) (125)

30

Total 78 (845) 70 (652) 51 (515) 53 (491) 252 (2503)

Table 3: Summary Statistics

Health Total private donations Sida government grants Other government grants Total government grants Total fundraising expenditures International aid Total private donations Sida government grants Other government grants Total government grants Total fundraising expenditures Social Services Total private donations Sida government grants Other government grants Total government grants Total fundraising expenditures Other Total private donations Sida government grants Other government grants Total government grants Total fundraising expenditures

Number of …rms 78

70

51

53

31

Number of observations 845

Mean (SEK 1,000)

Standard deviation

3465 124 3905 3960 972

7230 2985 11084 11149 3631

10881 9751 1435 11186 1594

25332 28361 4013 39138 5524

6733 680 7210 7918 851

14178 3685 33649 33831 2969

8376 7748 1827 9551 1068

15091 23413 3876 24624 3346

652

515

491

Table 5: Private donations and government grants, OLS All organizations Dependent variable OLS(1) OLS(2) Private donations Government grants

-.021** (.010) .970*** (.075) -3.21 (3.125) -.150 (.213) 0.2214

-2.13 (4.737) -.256 (.204) 0.0162

Number of observations

Organization and year 2224

Organization and year 2224

Number of organizations

252

252

Fundraising expenditures Costs Membership fees R2 Fixed e¤ects

-.015* (.009)

Notes: Robust standard errors in parentheses. *** denotes signi…cance at p<0.01, ** at p<0.05 and * at p<0.10.

32

Table 6: Private donations and government grants, OLS Type of organization

Health

Social Services

Other

OLS(1)

International aid OLS(2)

Dependent variable Private donations

OLS(3)

OLS(4)

-.098 (.069) .964*** (.073) -64.04** (31.39) 3.16* (1.80) 0.2561

-.004 (.43) 1.122*** (.042) -24.13 (19.13) -.85 (.87) 0.3071

-.023*** (.005) .42 (.29) -8.154 (5.24) 2.314*** (.683) 0.0966

.069** (.030) .818*** (.134) 3.397 (3.604) -.319*** (.034) 0.3188

Number of observations

Organization and year 764

Organization and year 569

Organization and year 457

Organization and year 434

Number of organizations

78

70

51

53

Government grants Fundraising expenditures Costs Membership fees R2 Fixed e¤ects

N o te s: R o b u st sta n d a rd e rro rs in p a re nth e se s. * * * d e n o te s sig n i…c a n c e a t p <0 .0 1 , * * a t p <0 .0 5 a n d * a t p <0 .1 0 .

33

Table 7: Private donations and government grants, 2SLS All organizations Dependent variable Private donations Government grants

2SLS(1) -.070 (.063) 1.448*** (.456) -4.045 (4.304) -.100 (.228)

Fundraising expenditures Costs Membership fees

R2 on second stage Results from …rst stage Instrument set

0.1632 Financial assets Administrative costs Repayments Material assets

F-test on instruments for government grants (p-value) F-test on instruments for fundraising expenditures (p-value) Over-identi…cation test J-statistic (p-value)

20.02 (.00)

7.327 (.50)

Hausman test F-statistic (p-value)

.88 (.41)

5.18 (.00)

Fixed e¤ects

Organization and year 2224 252

Number of observations Number of organizations

N o te s: R o b u st sta n d a rd e rro rs in p a re nth e se s. * * * d e n o te s sig n i…c a n c e a t p <0 .0 1 , * * a t p <0 .0 5 a n d * a t p <0 .1 0 .

34

Table 8: Private donations and government grants, 2SLS Type of organizations

Health

Dependent variable Private donations Government grants

2SLS(1)

Fundraising expenditures Costs Membership fees

R2 on second stage Results from …rst stage Instrument set

F-test on instruments for government grants (p-value) F-test on instruments for fundraising expenditures (p-value) Over-identi…cation test J-statistic (p-value) Hausman test F-statistic (p-value) Fixed e¤ects Number of observations Number of organizations

International aid 2SLS(2)

Social services

Other

2SLS(3)

2SLS(4)

-.925*

-.096 (.356) 2.243** (1.129) -2.035 (5.551) -.212*** (.072)

-.223 (.147) 1.282*** (.259) -79.40** (40.49) 3.603* (2.140)

(.557) -.109 (1.252) -75.34 (101.87) -1.244 (3.102)

-.057* (.033) .982 (1.052) -5.021 (6.792) 2.283*** (.672)

0.2206

.7758

.0402

.3076

Financial assets Material assets Repayments, Administrative costs 40.44 (.00)

Financial assets Material assets Repayments, Administrative costs 1.04 (.42)

Financial assets Material assets Repayments, Administrative costs 2312.90 (.00)

Financial assets Material assets Repayments, Administrative costs 19.27 (.00)

1532.22 (.00)

13.0 (.00)

11.96 (.00)

17.34 (.00)

7.800 (.45)

9.316 (.32)

3.563 (.47)

8.097 (.42)

0.70 (.50) Organization and year 764 78

1.17 (.32) Organization and year 569 70

1.70 (.19) Organization and year 457 51

14.22 (.00) Organization and year 434 53

N o te s: R o b u st sta n d a rd e rro rs in p a re nth e se s. * * * d e n o te s sig n i…c a n c e a t p <0 .0 1 , * * a t p <0 .0 5 a n d * a t p <0 .1 0 .

35

Table 9: Di¤erentiating between di¤erent types of government grants, 2SLS Type of organizations Dependent variable Private donations Sida Government grants Other Government grants Fundraising expenditures Costs Membership fees

R2 on second stage Results from …rst stage Instrument set

International aid 2SLS(1)

Other organizations 2SLS(2)

All organizations with foreign 2SLS(3)

-.624 (.680) 2.087** (1.05) 1.584** (.761) -133.949 (195.196) -4.030 (3.993)

-.155 (.171) .143 (.1.19) 2.386*** (.681) -2.794 (5.460) -.173 (.114)

-1.024* (.558) .987 (2.435) .987 (1.261) -91.62. (101.30) 1.347 (4.611)

0.2279

.4798

0.7717

Financial assets Administrative costs Repayments Material assets

Financial assets Administrative costs Repayments Material assets

Financial Assets Administrative cost Repayments Material Assets

21.33 (.00)

27.52 (.00)

4.70 (.00)

41.57 (.00)

13.91 (.00)

2.16 (.02)

F-test on instruments for Sida government grants (p-value) F-test on instruments for other government grants (p-value) F-test on instruments for fundraising expenditures (p-value) Over-identi…cation test J-statistic (p-value)

2.35 (.02)

17.34 (.00)

4.24 (.00)

4.659 (.79)

7.149 (.41)

6.929 (.54)

Hausman test F-statistic (p-value)

2.64 (.056)

11.22 (.00)

4.14 (0.009)

Organization and year 507 69

Organization and year 434 53

Organization and year 751 88

Fixed e¤ects Number of observations Number of organizations

N o te s: R o b u st sta n d a rd e rro rs in p a re nth e se s. * * * d e n o te s sig n i…c a n c e a t p <0 .0 1 , * * a t p <0 .0 5 a n d * a t p <0 .1 0 .

36

Table 10: Di¤erentiating between di¤erent types of government grants, 2SLS Type of organizations Dependent variable Private donations Sida Government grants Other Government grants Fundraising expenditures Costs Membership fees

R2 on second stage Results from …rst stage Instrument set

All organizations 2SLS(1)

Health 2SLS(2)

Social Services 2SLS(3)

-.229 (.338) -.069 (.057) 1.391*** (.417) -4.645 (4.920) -.118 (.226)

.004 (.110) -.168 (.148) 1.065*** (.307) -67.74* (36.19) 3.496 (2.291)

-2.131 (4.406) -.046** (024.) .820 (.855) -7.25 (6.36) 2.431*** (.673)

0.1358

.1745

0.2829

Financial assets Administrative costs Repayments Material assets

Financial assets Administrative costs Repayments Material assets

Financial Assets Administrative costs Repayments Material Assets

.58 (.83)

15.35 (.00)

8.11 (.00)

11.27 (.00)

560.45 (.00)

2053.32 (.00)

F-test on instruments for Sida government grants (p-value) F-test on instruments for other government grants (p-value) F-test on instruments for fundraising expenditures (p-value) Over-identi…cation test J-statistic (p-value)

5.18 (.00)

1532.22 (.00)

11.96 (.00)

7.090 (.42)

8.136 (.32)

12.502 (.09)

Hausman test F-statistic (p-value)

.99 (.40)

1.82 (.15)

1.62 (.20)

Organization and year 2224 252

Organization and year 764 78

Organization and year 457 51

Fixed e¤ects Number of observations Number of organizations

N o te s: R o b u st sta n d a rd e rro rs in p a re nth e se s. * * * d e n o te s sig n i…c a n c e a t p <0 .0 1 , * * a t p <0 .0 5 a n d * a t p <0 .1 0 .

37