Government Decentralization and the Size of the Nonprofit Sector: Revisiting the Government Failure Theory

Abstract

This article revisits government failure theory by examining the relationship between government decentralization and the size of the nonprofit sector (NPS). Government failure theory posits that nonprofits are most active in regions where the largest gap exists between the homogeneous supply of public service and heterogeneous citizen demands. Following this theory, government decentralization should decrease the size of the NPS, as it increases the efficiency and heterogeneity of government services. This article tests this hypothesis using a sample of U.S. counties. Decentralization is measured in two dimensions: vertical decentralization and horizontal fragmentation. After using instrumental regressions to eliminate the endogeneity bias, we find that counties with a more horizontally fragmented governmental system are associated with a larger NPS. Vertical centralization leads to a denser NPS but has no impact on the NPS revenue or assets. The impacts of resident heterogeneity are also mixed. As such, government failure theory is only partially supported, at best. Contrarily, interdependence theory is supported by this study.

Keywords

nonprofit size government–nonprofit relationship fiscal decentralization government failure

Introduction

The size of the nonprofit sector (NPS), as long noticed by scholars (e.g., Ben-Ner & Van Hoomissen, 1992; Boris & Steuerle, 1999), varies significantly across regions. Whereas Pennsylvania has about 15 nonprofit organizations (NPOs) per ten thousand capita, Nevada, although similar to Pennsylvania with respect to median household income, has less than five NPOs per ten thousand capita (Boris, 1999).¹ What explains the regional variation of the size of the NPS?

The NPS has become an increasingly important economic force in local economies (Gronbjerg & Paarlberg, 2001). According to the Urban Institute, nonprofits accounted for 9.2% of all wages and salaries paid in the United States in 2010 (Blackwood, Roeger, & Pettijohn, 2012). As such, understanding the factors that determine the size of the NPS has important policy implications for researchers, public and nonprofit managers, private philanthropy, and public policy makers. First, it will help them to understand important aspects of the roles that nonprofits play in our society and the impacts of the growth and capacity of the NPS on local economies. Such insights gained from the research may help to develop effective solutions to social problems and discourage those that are ineffective (Gronbjerg & Paarlberg, 2001). Second, it will help interested parties to appreciate the complex and dynamic government–nonprofit relationship. The development of public policies aimed toward the private or NPS requires a precise and holistic view of the government–nonprofit relationship; a distorted or incomplete view of the relationship will harm the design of effective public policies (Young, 2000).

This article focuses on government failure theory, one of the most influential theories used to explain the size of the NPS. Government failure theory contends that the NPS surfaces to supplement the homogeneous provision of government services (Weisbrod, 1977, 1991). Government services, due to their homogeneity, are incapable of satisfying the heterogeneous demands of consumers, who consequently have incentives to form NPOs and supply collective goods in a more heterogeneous way. Consequently, NPOs should be most active in regions where there exists the largest supply–demand gap between the homogeneous governmental services and the heterogeneous citizen preferences. Thus, a more heterogeneous and efficient provision of government services should decrease the size of the NPS.

Furthermore, the literature on fiscal federalism suggests that a more decentralized government system leads to a more efficient and diverse provision of public service. First, a fragmented government system,² with denser government units, provides a greater selection of local public services to its residents. Second, a decentralized government system physically close to its inhabitants is more likely to tailor its public service to satisfy the demands of its social members. Third, governmental decentralization induces inter-governmental competitions and consequently increases the efficiency of public services (Oates, 1999). Thus, if government failure theory can explain the regional difference in the size of the NPS, we can feasibly expect that the size of the NPS should vary inversely with the degree of fragmentations and fiscal decentralizations of governments.

This research empirically revisits government failure theory using county-level data in 2002, with a focus on examining the link between the fiscal decentralization of governments and the size of the NPS. Previous empirical studies of government failure theory emphasize the variation on the demand side of the public provision and capture it using population heterogeneities (Kim, 2015; Matsunaga & Yamauchi, 2004), but they either ignore the heterogeneity on the supply size of public service provision (Corbin, 1999; James, 1987) or try to capture it using merely the amount of governmental expenditure (Kim, 2015; Matsunaga & Yamauchi, 2004). We argue that the supply side of public provision also varies by regions in terms of its heterogeneities and efficiency. This research controls for the supply side of the governmental provision by including government fragmentation and fiscal decentralization into the analysis.

This study has a twofold contribution. First, it enhances our understanding of the government–nonprofit relationship. It is the first study that tests the government failure theory from the perspective of the diversity and efficiency of government services. We also control for more types of population heterogeneity than previous studies, including the seldom-examined educational heterogeneity. Second, it enriches the fiscal federalism literature by being the first research that explores the impacts of government decentralization on the NPS.

“Literature Review” section of this article reviews the related literature, including the government–nonprofit relationship literature and the government decentralization literature. “Theoretical Framework and Hypotheses” section presents the conceptual framework and hypotheses, followed by a discussion of the empirical model and data in “Empirical Method” section. “Estimation Results” section analyzes the regression results and the final section concludes with a summary of our findings, policy implications, and a brief discussion of limitations.

Literature Review

Government Failure, Interdependence, and Social Cohesive Theories

Weisbrod (1977, 1991) developed government failure theory by reconciling the persistence of NPOs with public economic theory. This theory regards the presence of NPOs as a response to the failure of governmental provision. In democratic societies, the level and the type of public goods are determined by the median voter’s preference through the process of public choice. As a result, social members whose preferences differ from that of the median voter are left unsatisfied. Thus, the heterogeneity of households’ preferences and the homogeneity of government goods generate a supply–demand gap, calling for a more diversified provision of collective services.

NPOs, with their capability to provide more diversified public goods, surface to fill this gap (Gronbjerg, 1993; Hansmann, 1987; Luksetich, 2008; Salamon, 1987; Smith & Gronbjerg, 2006; Young, 2000). NPOs are more likely to meet the heterogeneous demand of consumers for at least three reasons. First, NPOs are more fragmented and localized than governments. Hence, NPOs are closer to consumers and can tailor their services to consumers’ preferences, just as local governments are more responsive to residents than upper level governments. Second, NPOs are unconstrained by election pressure as faced by public policy makers. Thus, they may shape their services to meet the widely varied needs of different social groups rather than just cater to the median voter (Gronbjerg & Paarlberg, 2001; Matsunaga & Yamauchi, 2004; Young, 2000). Third, they are more consumer-oriented as they operate more like a business. All of these reasons contribute to the advantages of the NPS in providing heterogeneous collective goods. However, the “free rider” problem leads to the failure of nonprofits in providing pure public services. Thus, while the NPS effectively supplements governments in the provision of public or quasi-public services, it cannot substitute for the public sector (Young, 2000).

Previous tests of this theory generally examine whether it has explanatory power over the regional difference of the NPS (see Matsunaga & Yamauchi, 2004, and Lecy & Van Slyke, 2013, for a review). They posit that a region with a more heterogeneous population has a larger NPS, as social members in the region have more diverse preferences and thus are less likely to be satisfied with the public provision.

Unfortunately, these studies fail to sufficiently control for the supply side—the homogeneous degree of government provision. According to the tenets of government failure theory, both the supply side (the degree of the homogeneity of government services) and the demand side (the degree of the heterogeneity of citizens’ preferences) may affect the size of the NPS. Furthermore, the supply side’s homogeneity is associated with the degree of residents’ demand heterogeneity. Failing to sufficiently control for the heterogeneity of governmental provision may consequently lead to omitted variable bias and generate a distorted picture.

Previous empirical studies can be classified into three categories based on how they capture the impacts of governmental provision. The first group, perhaps due to data availability, makes no attempt to control for the supply side of the public provision. James’s (1987) research falls into this group and so does Corbin’s (1999) research, which in turn calls for a “true test” of government failure theory that should consider the supply side of the provision of public services.

While Matsunaga and Yamauchi (2004) regard their research as a response to Corbin’s work that calls for the “true test” of government failure theory, they simply consider the size of the provision. As a result, Matsunaga and Yamauchi attempt to correct for the flaw by including the aggregate government spending and government subsidy to the NPS into their models. Yet, the problem is that public services vary across regions in more than merely the size. As discussed in next sections, even with the same level of public expenditure, the unsatisfied level of the population demand may vary with the degree of government fragmentation and government fiscal decentralization. Thus, this group of studies, including Ben-Ner and Van Hoomissen (1992), Kim (2015), Lecy and Van Slyke (2013), Matsunaga and Yamauchi (2004), Paarlberg and Gen (2009), and Salamon, Sokolowski, and Anheier (2000), fail to conduct a true test of government failure theory.

The third group mainly refers to Abzug and Turnheim (1998), who notice that difference in government provision is more than the government expenditure amount. However, their proxy for the difference of governments across regions—Moody’s municipal bond credit rating—fails to capture the heterogeneity of the governmental provision. Municipal bond ratings measure the default risk of municipal debts rather than the diversity or the efficiency of government services. Moreover, their study fails to control for the total size of the government provision.

These studies have mixed findings on the association between population heterogeneity and the size of the NPS. Some find a positive relationship between the demand heterogeneity and the size of the NPS (Corbin, 1999; James, 1987; Matsunaga & Yamauchi, 2004), as the government failure theory predicts; others fail to find a significant relationship (Abzug & Turnheim, 1998; Salamon et al., 2000) or find a negative relationship (Gronbjerg & Paarlberg, 2001); still others find that the relationship depends on the functional sectors examined (Ben-Ner & Van Hoomissen, 1992; Gronbjerg & Paarlberg, 2001; Kim, 2015). Many of them found that racial heterogeneity and religious heterogeneity cannot significantly predict the size of the NPS.

Ben-Ner and Van Hoomissen (1992) propose social cohesion theory as an attempt to reconcile these inconsistent findings, arguing that a society with more cohesive members is more likely to form NPOs. Social cohesion theory explains the cross-regional difference in NPS sizes from the perspective of social members’ willingness to contribute and form NPOs, rather than their demand for NPOs. It predicts that population heterogeneity leads to a smaller NPS. It is imperative to note that social cohesion theory and government failure theory are not necessarily in conflict with each other, as they may both have explanatory power on the size of the NPS (Corbin, 1999). Thus, an examination of the relationship between the population heterogeneity and the size of the NPS may be inadequate to justify or reject government failure theory. Even if we fail to find a significant positive relationship between population heterogeneity and the size of the NPS, it is still premature to conclude that government failure theory is rejected. From this perspective, any findings based solely on population heterogeneity should not be considered direct evidence for supporting or rejecting the government failure theory. Instead, examining the explanatory power of government efficiency and heterogeneity (as captured by governmental fiscal decentralization and fragmentation) over the size of the NPS may be a more direct test of the government failure theory.

Interdependence theory is yet another theory that focuses on the government–nonprofit relationship to explain the size of the NPS. By portraying nonprofits as supplements to government, government failure theory is unable to explain the fact that much of the nonprofit revenue comes from governmental funding. To rectify this impasse, Salamon (1987, 1995) develops interdependence theory, which Young (2000) refers to as complementary theory. It argues that it is a win-win strategy for governments to delegate the delivery and production of some public services to the NPS (Lecy & Van Slyke, 2013; Salamon, 1987; Young, 2000). Government uses its coercive power to collect taxes to avoid the free-riding of public goods (Olson, 1965), while contracting out some service provision to the NPS allows government to capitalize on the expertise, flexibility, diversity, and efficiency of nonprofits. As such, the public sector and the NPS have a complementary or interdependent relationship. The interdependence theory and government failure theory are not necessary mutually exclusive. Instead, they explain the different perspectives of the complex government–nonprofit relationships (Young, 2000).

Government Decentralization

On the supply side, the diversity of government provisions cannot be fully captured by the amount of government spending. This article contends that the degree of government decentralization also affects the diversity and efficiency of government services.

The following hypothetical example demonstrates how government decentralization may affect the size of NPS. Assume that there are two counties, A and B. In both county areas, the amounts of government spending are similar, but County A has a more fragmented provision of public services. Assume County A has three general governments and five public schools, while County B has only one general government and one school. We can envision that residents in County A have more options for public education and other public services than their counterparts in County B. Ceteris paribus, people in County A are more likely to be satisfied with their public education and other public services. If government failure theory is correct, then residents in County B should have more incentives to form NPOs for the reasons discussed below.

First, even though the population in County A and that in County B may have the same level of demand heterogeneity, the group collectively consuming public services in County A is actually smaller and thus less heterogeneous than in County B. In other words, the whole population in County A is divided into smaller groups to consume collectively provided goods, which thus are more likely to be tailored to residents’ preferences.

Second, given the population mobility in the United States, consumers can “shop around” for public services that best match their preferences, resulting in a smaller government failure (Tiebout, 1956). This migration further leads to the aggregation of people with similar preferences to consume the same public services. Consequently, the more fragmented a government system, the smaller the government failure would be. In other words, assuming a mobile population, government fragmentation allows a rearrangement of the population according to their demand for public goods, thereby increasing the social efficiency. Furthermore, a more fragmented public provision means a “closer” policy decision and implementation to consumers and hence is more likely to satisfy households’ demands (Zax, 1989). These two rationales focus on allocative efficiency, arguing that public services provided by governments at a lower level can be better matched to the preferences of the population.

Third, even without the interregional relocation of the population, fragmentation may still improve government efficiency through government competition (Oates, 1999). This argument focuses on productive efficiency rather than allocative efficiency referred to above. The rationale is that governments in a more fragmented public system would be more likely to compete for potential tax base, leading to a more efficient provision of public services. In addition, decentralization can also improve productive efficiency by increasing electoral control over incumbents and lowering the possibility of corruption (Fisman & Gatti, 2002; Seabright, 1996). Contrarily, it has also been suggested that decentralization may undermine productive efficiency if economies of scale is important for the production or delivery of specific public goods (Sow & Razafimahefa, 2015).³ Whereas decentralization, theoretically, may result in such an efficiency loss, it is generally perceived that “for most local services provided locally, the provision in a given city is independent of the provision in other cities. The welfare losses attributable to economies of scale that would result from decentralization are probably minimal” (Prud’Homme, 1995, p. 209). In essence, how fiscal decentralization affects the productive efficiency is an empirical question. Previous studies have empirically examined the effect of decentralization on corruption (Fisman & Gatti, 2002), immunization coverage rate (Khaleghian, 2004), educational attainment (Barankay & Lockwood, 2007), economic growth (Akai & Sakata, 2002; Iimi, 2005; Lin & Liu, 2000; Xie, Zou, & Davoodi, 1999; Zhang & Zou, 1998), and the size of public provision (Bates & Santerre, 2006; Oates, 1985; Zax, 1989). Most of these studies have found evidence that decentralization increases productive efficiency.

The analysis above rests on the presumption that government policies are all designed and implemented at the same level. Yet, in the current federalism system, all three levels of government have their own authority and responsibilities in providing public services. How different levels of governments divide their resources and responsibilities in supplying public services may affect the efficiency of public services and the satisfaction of the public. Consistent with the dimension of horizontal government fragmentation, the degree of vertical decentralization is also positively related to government efficiency and the heterogeneity of public services (Oates, 1999).

Theoretical Framework

Based on the above literature review, we have a generalized version of government failure theory for explaining the regional difference in the NPS size. First, unlike previous literature claiming that public services are provided by governments in a uniform fashion across regions, we acknowledge that some regions may have less “failure” than others in their governmental provision. Thus, differences in the government “failure” may lead to different sizes of the NPS across regions. One option for generating a more satisfying supply of public services is to decentralize government services, assuming that regional spillover is no threat.

Second, when local governments fail to satisfy households’ demand for public services, local residents can vote against the policy, “vote with their feet” by migrating to the region where the tax-public service matches their preferences best, or form NPOs to provide public services. This line of reasoning is demonstrated in Figure 1. This framework differs from the traditional fiscal federalism theory in that it suggests that in addition to emigration and voting against the current policy, unsatisfied residents can also adopt a third strategy: aligning to form NPOs for supplying public services. For most residents, this strategy is probably more feasible than relocating to other regions, as relocation involves higher costs and is constrained by other limitations (e.g., job location). Furthermore, residents may be unsatisfied with only some services, but not with the service package as a whole. In this case, the NPS can satisfy their unmet demands without pushing them to switch the entire public service package.

Figure 1.

An illustration of the generalized government failure theory.

Based on the above analysis, we posit that, holding all else constant, the NPS is smaller in regions where public services are decentralized and more diverse. In addition, governments’ direct expenditures may also determine the size of the NPS. Government failure theory predicts that a higher direct government expenditure is associated with a smaller size of the NPS because more sufficient government provision will lead to less government failure. However, the complementary theory proposed by Salamon et al. (2000) holds a different prediction. This theory regards governments and NPOs as complementary with each other. Thus, under the lens of this theory, a larger public sector is associated with a more sizable NPS.

Empirical Methodology

Model

Based on the above theoretical framework, we have the following empirical model:

\begin{array}{l} S i z e o f N P S = β_{0} + β_{1} H e t e r o g e n e i t y + β_{2} G o v e r n m e n t s e r v i c e \\ + β_{3} C o n t r o l v a r i a b l e s + β_{4} s t a t e d u m m y v a r i a b l e s + ε, \end{array}

where the dependent variable—the size of the NPS—is measured with the number of NPOs per ten thousand capita, the NPS revenues or assets per capita in a logarithmic form. A detailed description of the data is provided in Table 1.

Table 1.

Variable Description.

Variables	Description	Expected sign
Dependent variables
Number of tax-filing NPOs per 10,000 capita	Total number of 501(c)3 public charities filing 990 or 990 EZ with the IRS × 10,000/population; based on Digitized Data
Number of registered NPOs per 10,000 capita	Total number of active 501(c)3 public charities that have registered for tax-exempt status with the IRS × 10,000/population; based on BMF
NPS revenue per capita	Aggregate revenue of 501(c)3 public charities filing with the IRS/population
NPS assets per capita	Aggregate assets of 501(c)3 public charities filing with the IRS/population
Government service-related variables
Governments per 10,000 capita	Number of governments in the county area, including both general governments and special purpose governments, per 10,000 capita	−
County share of local total revenue	County government revenue/total governmental revenue in the county	+
Government expenditure per capita	Aggregated government expenditures in the county per capita	−
Government grants per capita	Revenue from government grants/population	+
Government grants/total NPS revenue	Revenue from government grants/total revenue of NPS in the county	+/−
Resident heterogeneity
Educational heterogeneity	Population heterogeneity in education, see section “Measures and Data” for details
Racial heterogeneity	Population heterogeneity in race, see section “Measures and Data” for details	+/−
Religious heterogeneity	Population heterogeneity in religion, see section “Measures and Data” for details	+/−
Gini index	Gini Index of the county in 2000	+/−
Other control variables
Median age	Median age of the county population	+
Crime index	Total number of crimes based on Uniform Crime Reporting Index, including arson.	−
Unemployment rate	Unemployment rate	−
Urbanization ratio	Urban population/total population, in 2000	−
Household median income	Median value of household income	+
Median educational attainment	Median years of educational attainment	+
Population	Population in the county
Instrumental variables
Number of streams	Number of streams in the county as defined by USGS
Number of ranges	Number of ranges in the county as defined by USGS

Note. NPO = nonprofit organizations; IRS = Internal Revenue Service; NPS = nonprofit sector; BMF = Business Master Files; USGS = U.S. Geological Survey.

Heterogeneity is a vector of population heterogeneity, including the heterogeneity of race, religion, education, and income. The next subsection “Measures and Data” will discuss in detail how these variables are measured.

Government service is a vector of variables capturing the quantity and diversity of public provision, including the degree of government fragmentation, the degree of government vertical decentralization, government expenditures, and government grants to the NPS.

Control variables include the median household income, unemployment rate, urbanization rate, median highest educational attainment, median age, crime index, and population.

Some state-specific variables, such as colonial history, political regimes, state expenditures and policy, and so on, may affect the size of the NPS (Matsunaga & Yamauchi, 2004). Hence, state-fixed effect models are used to absorb these unobservable effects over the size of the NPS.

In addition, the endogeneity of government fragmentation variables may also challenge the estimation. Government fragmentation (as measured by government density or the number of governments per ten thousand capita) may be affected by some population characteristics. While population heterogeneity is, to some extent, controlled for, in our model, it is unrealistic to assume that the endogeneity bias can be completely eliminated by ordinary least square (OLS) regression. Regions with a tradition of preferring more public services may have both a larger public sector and a greater NPS. To generate consistent estimates, this research utilizes instrumental regressions. This study derives instrumental variables, the numbers of streams and ranges, from geographical features. The number of streams has been found to correlate with the government density (Alesina, Baqir, & Hoxby, 2004; Hoxby, 2000). It is reasonable to expect that these geographical features are uncorrelated with unobservable effects. Further empirical tests, as explained in the section “Estimation Results,” confirm the validity of these instrumental variables.

Measures and Data

We focus the examination on the county level in the United States, as county-level data offer some advantages for the purpose of this research. First, unlike countries or states, counties are characterized by the cross-border mobility of households, an important assumption for some underlying arguments of the present research. Second, while counties are not too large to mask the heterogeneity within the same area, they are large enough to contain the externalities of most NPOs, ensuring that the NPS in one observation would not directly and substantially influence the habitats in others. Finally, county-level data provide a larger sample, potentially yielding more reliable estimation.

While the number of general-purpose governments increased only slightly from 1952 to 2002, the number of special district governments almost tripled during the same period (U.S. Census Bureau, 2002). Scholars have attributed the forming of jurisdictions to such factors as population growth and movement along the search for lower taxes and for a particular bundle of taxes and services (Burns, 1994). Alesina et al. (2004) argues that in addition to the aforementioned factors, the size of jurisdictions is also a result of the trade-off between population heterogeneity and the economies of scale. Following this argument, one may be concerned that some unobserved factors (e.g., economies of scale) may affect both the density of governments and the size of the NPS. If so, fragmentation measured by the government density will be endogenously determined in Equation 1. To eliminate the endogeneity, we use instrumental regressions to estimate Equation 1.

We use data from various sources, including the U.S. Census Bureau, the GuideStar National Nonprofit Research Database (the “Digitized Data”) and the Business Master Files (BMF) of the National Center for Charitable Statistics (NCCS), U.S. Department of Justice, and so on. The Digitized Data are collected from Internal Revenue Service (IRS) Form 990, Form 990-EZ and other related forms, which are required to file for NPOs with an annual gross income of no less than US$25,000. Thus, the Digitized Data set includes the vast majority of the public charities exempted from the federal income tax under section 501(C)(3) of the Internal Revenue Code. 501(C)(3) organizations with less than US$25,000 in annual gross receipts are excluded, as they are not required to file Forms 990 or 990-EZ with the IRS.⁴ The Digitized Data set is an updated subset of 6 years (1998-2003) from the NCCS full data set (core files). It has been checked for accuracy⁵ and includes some variables that are not available in other NCCS datasets (e.g., governmental grants). We use the data of 2002, as it is the latest year available from the Digitized Data set that can match the year of government finance census conducted every 5 years. As a robust check, we also use the number of 501(C)(3) public charities from the NCCS BMF, which contains descriptive information for all active organizations that have registered for the tax-exempt status with the IRS. The difference between the Digitized Data and the BMF is that while the former contains more detailed financial information, the latter includes only public charities required to file tax forms with the IRS. Thus, the BMF data include NPOs of all size, while the Digitized Data include only those that are large enough to be required to file for tax forms.

Most variables are for year 2002, during which a government finance census was conducted by the Census Bureau. When the data for 2002 are unavailable, data for 2000 are used (e.g., population heterogeneity variables). The final sample, after dropping those with missing values, has 3,032 observations. Table 2 shows the descriptive statistics of all variables.

Table 2.

Descriptive Statistics (N = 3,032).

	M	SD	Minimum	Maximum
Dependent variables
Number of tax-filing NPOs per 10,000 capita, Digitized Data	14.68	8.56	0.78	115.38
Number of registered NPOs per 10,000 capita, BMF data	28.44	24.35	2.64	761.72
NPS revenue per capita	1,396.73	2,193.89	0.76	46,644.17
NPS asset per capita	2,232.83	4,819.45	0	72,259.99
Government service-related variables
Governments per 10,000 capita	6.29	12.00	0.01	168.33
County share of total local government revenue	0.29	0.21	0	1
Government expenditure per capita	2,757	1,484.36	501.87	44.79
Government grants per capita	152.24	350.33	0	7,323.75
Government grants/total NPS revenue	0.15	0.19	0	1
Resident heterogeneity
Educational heterogeneity	0.76	0.05	0.39	0.84
Racial heterogeneity	0.25	0.18	0.01	0.75
Religious heterogeneity	0.46	0.15	0.07	0.84
Gini index	0.43	0.04	0.33	0.59
Other control variables
Median age	37.41	3.81	20.70	54.30
Crime index	0.02	0.02	0	0.11
Unemployment rate	5.72	1.86	2.10	19.70
Urbanization ratio	0.59	0.30	0.00	1.00
Median household income	36,314.49	9,500.25	16610.00	93,927.00
Median educational attainment (in years)	12.39	0.74	8.32	15.71
Population	94,218.48	294,998.20	567.00	9,776,424.00

Note. NPOs = nonprofit organizations; BMF = Business Master Files; NPS = nonprofit sector.

The size of the NPS

Previous studies measure the size of the NPS using the number or the density of NPOs, or the employment in NPS. Yet, NPS revenues and assets per capita could more precisely capture the size of the NPS. Thus, we utilize the number of NPOs per ten thousand capita, NPS revenues per capita and NPS assets per capita to measure the size of the NPS. We use the logarithmic values for these dependent variables.

Figure 2 shows the distribution of NPOs, in terms of both the number and the expenditure, in our data set. The Northeast has the densest NPOs, with most counties in the region having a density of more than 10 tax-filing NPOs per 10,000 people. It is also home to the NPS with the highest expenditure per capita. The Rocky Mountain area and the South have a relatively small size of the NPS as measured by either the NPO density or expenditure per capita.

Figure 2.

Distribution of nonprofit organizations by counties in 2002.

As shown in Table 1, counties in our sample have an average of approximately 15 tax-filing NPOs per ten thousand capita. The average revenue per capita of the NPS is US$1,397, and the average asset per capita of the NPS is US$2,233.

Government service

While there is no perfect measure of government fiscal decentralization, Oates (1985) and others (Arikan, 2004; Zax, 1989, to name a few) use the density of government units to measure government fragmentation and the fiscal share among vertical governments to measure fiscal centralization/decentralization.

We measure government fragmentation as the number of government units in a county per 10,000 people, including both general-purpose and single-purpose local governments. There are about six governments per ten thousand capita on average in a U.S. county. Following Zax (1989), we measure centralization as the ratio between the county government’s revenue and the total revenue of all governments in the county. Approximately 29% of government revenues in a county are by the county government.

We also include governmental expenditure per capita and government grants as control variables. In models using the NPO density as the dependent variable, the variable of government grants is measured using its logarithmic value of the per capita government grants to the NPS. In models using the NPO revenue or assets as the dependent variable, it is not appropriate to measure government grants using the per capita value, because government grants are part of the NPO revenue. Thus, in those models, the variable of government grants is measured using government grants as a ratio of total NPS revenue. The findings on the effects of government funding on the NPS are mixed. While some researchers have found a crowding out effect (Andreoni, 1993; Bergstrom, Blume, & Varian, 1986; Duncan, 1999; Warr, 1982), others have found that government funding can crowd in private contributions (Rose-Ackerman, 1982; Schiff, 1985; Seaman, 1981).

Demand heterogeneity

Population heterogeneity is measured in four dimensions: race, religion, education, and income. Most previous studies focus on racial and religious heterogeneities. Educational and income heterogeneities have not received sufficient attention. Since an individual’s demand on public service is affected by his or her income, income heterogeneity of the community is more likely to influence the heterogeneity of a community’s demand. Indeed, sociologists and other social science scholars have identified income heterogeneity as an important factor that affects the social preference, social conflict, and policy decision (Alesina et al., 2004; Wilson, 1996). Excluding the dimension of income heterogeneity overlooks an important perspective of the population heterogeneity. Similar reasons exist for educational heterogeneities.

This study uses the Gini coefficient as the measure for income heterogeneity.⁶ The average Gini coefficient of counties in our sample is 0.43. We follow Alesina et al.’s (2004) measure of racial heterogeneity as the probability that two randomly drawn persons belong to different races, which are divided into the five categories as used in the Census of Population. More specifically, racial heterogeneity is calculated as follows:

R a c e = 1 - \sum_{i} {(\frac{r a c e_{i}}{t o t a l_p o p u l a t i o n})}^{2} .

Religious heterogeneity is measured in a similar way. As education can be measured as a numerical variable, we calculate educational heterogeneity as the coefficient of variation of education using the following formula:

E d u = \frac{s (e d u)}{\bar{e d u}},

where the numerator and denominator are the standard deviation and the mean of educational attainment (in years) of the population in the county, respectively. All four heterogeneity measures have a value ranging from 0 to 1, with a higher value representing more heterogeneity.

Control variables

Table 1 lists the definitions of the control variables and their expected signs. These expected signs largely follow Matsunaga and Yamauchi’s (2004) research, which offers a detailed review of the determinants of the NPS size. Lecy and Van Slyke (2013) also summarize the empirical findings of major determinants in previous studies.

Intuitively, we may expect that a region with a higher median household income has a more active NPS because wealthier residents are more likely to contribute to charitable organizations. Researchers have found some empirical evidence (Gronbjerg & Paarlberg, 2001; Hall, 1987; Lecy & Van Slyke, 2013; Saxton & Benson, 2005). However, others have found a negative relationship between household income and NPS size (Ben-Ner & Van Hoomissen, 1992; Matsunaga & Yamauchi, 2004). They attribute this negative relationship to the asymmetric information between consumers and service providers. When consumers are unable to evaluate the quality of services and products from for-profits, they prefer nonprofit services because NPOs have less incentive to cheat consumers (Young, 2000). Because rich consumers are more likely to afford the transaction costs associated with differentiating the trustworthiness of for-profit organizations, wealthier communities may rely more heavily on the for-profit sector for services than on the NPS.

Most previous studies have found that a community with a higher educational attainment is associated with a more active NPS (Wolpert, 1993). Findings about the effect of the crime rate are inconclusive. Matsunaga and Yamauchi (2004) hypothesize that crime rate decreases the size of NPS, as crimes lessen people’s social activities and thus discourages social cohesion and consequently charitable activities. Their empirical study supports this hypothesis. Lecy and Van Slyke (2013) have failed to find a significant relationship between crime rate and the size of the NPS. Urbanization level is considered a determinant with a negative impact, as urbanization decreases the level of social cohesion (Matsunaga & Yamauchi, 2004).

Estimation Results

This section discusses the regression results. Tables 3 and 4 present the results for models measuring the size of the NPS using NPO density and NPO financial information, respectively. All models are estimated using both an OLS regression and a two-stage least square (2SLS) instrumental regression. All models control for state-fixed effects and account for the heteroscedasticity of error terms by estimating and reporting the robust standard errors of coefficients.

Table 3.

Regression Results for NPO Density (N = 3,032).

Variables	Log (NPO no. per 10,000 capita)
	NPO no. from Digitized Data		NPO no. from BMF
	Model 1	Model 2	Model 3	Model 4
	OLS	2SLS	OLS	2SLS
Log (government density)	0.06** (0.020)	0.95*** (0.210)	0.07*** (0.014)	0.76*** (0.159)
County share of local total revenue	−0.13 (0.080)	0.53** (0.191)	−0.10 (0.055)	0.42** (0.144)
Log (government grants per capita)	0.05*** (0.003)	0.05*** (0.003)	0.02*** (0.002)	0.02*** (0.002)
Log (government expenditure per capita)	0.10*** (0.030)	−0.16* (0.071)	0.12*** (0.022)	−0.08 (0.054)
Educational heterogeneity	−4.74*** (0.436)	−5.64*** (0.570)	−2.48*** (0.288)	−3.19*** (0.399)
Racial heterogeneity	−0.17* (0.082)	−0.14 (0.102)	0.21*** (0.057)	0.23** (0.076)
Religious heterogeneity	−0.06 (0.086)	0.39** (0.150)	0.05 (0.067)	0.40*** (0.120)
Gini index	1.47*** (0.353)	0.95* (0.446)	1.45*** (0.232)	1.04*** (0.312)
Median age	0.02*** (0.003)	0.01* (0.005)	0.01*** (0.003)	0.01 (0.004)
Crime index	4.72*** (0.705)	8.26*** (1.206)	2.31*** (0.499)	5.09*** (0.864)
Unemployment rate	−0.01 (0.006)	−0.02* (0.009)	−0.01** (0.005)	−0.02** (0.007)
Urbanization ratio	−0.13* (0.053)	−0.53*** (0.115)	−0.08* (0.034)	−0.40*** (0.085)
Log (household median income)	0.87*** (0.171)	0.59* (0.228)	1.09*** (0.126)	0.87*** (0.170)
Median educational attainment	0.57*** (0.027)	0.65*** (0.040)	0.35*** (0.019)	0.42*** (0.030)
Log (population)	−0.97*** (0.170)	−0.25 (0.280)	−1.20*** (0.123)	−0.63** (0.208)
Constant	−0.80 (0.466)	2.40** (0.878)	2.21*** (0.314)	4.60*** (0.676)
State fixed effect	√	√	√	√
Kleibergen–Paap rk LM statistic		35.92		35.92
Cragg–Donald Wald F statistic^a		12.21		12.21
Hansen J statistic^b		1.89		1.89
Endogeneity test^c		34.05		34.05
Adjusted R²	.64	.41	.68	.42

Note. Robust standard errors are in parentheses; coefficients for state dummy variables are not shown; excluded Instrumental variables: number of streams and number of ranges. NPO = nonprofit organizations; BMF = Business Master Files; OLS = ordinary least square; 2SLS = two-stage least square; LM = Lagrange Multiplier

Stock-Yogo weak ID test critical values (10% maximal IV relative bias): 9.08.

Critical value: χ²(df = 2, α = .05) = 5.99.

χ²(df = 1, α = .05) = 3.84.

†

p < .1. *p < .05. **p < .01. ***p < .001.

Table 4.

Regression Results for NPS Revenue and Assets (N = 3,032).

Variables	Log (NPS revenue per capita)		Log (NPS asset per capita)
	Model 5	Model 6	Model 7	Model 8
	OLS	2SLS	OLS	2SLS
Log (government density)	0.15* (0.059)	1.75** (0.586)	0.18** (0.059)	1.64** (0.557)
County share of local total revenue	−1.48*** (0.239)	−0.30 (0.526)	−1.33*** (0.256)	−0.25 (0.511)
Government grants/NPS revenue	−0.40** (0.149)	−0.38* (0.165)	−0.80*** (0.150)	−0.78*** (0.162)
Log (government expenditure per capita)	−0.16 (0.095)	−0.62** (0.196)	−0.12 (0.100)	−0.55** (0.192)
Educational heterogeneity	−7.19*** (1.266)	−8.82*** (1.525)	−9.29*** (1.335)	−10.79*** (1.539)
Racial heterogeneity	−0.79*** (0.235)	−0.73** (0.271)	−0.82*** (0.248)	−0.78** (0.271)
Religious heterogeneity	−0.71** (0.260)	0.11 (0.411)	−0.46^† (0.269)	0.29 (0.408)
Gini index	6.95*** (1.013)	5.99*** (1.141)	6.67*** (1.076)	5.78*** (1.179)
Median age	−0.01 (0.011)	−0.03^† (0.013)	−0.02^† (0.011)	−0.03* (0.013)
Crime index	9.59*** (2.071)	15.95*** (3.268)	9.93*** (2.299)	15.73*** (3.297)
Unemployment rate	0.01 (0.019)	−0.00 (0.021)	0.01 (0.019)	−0.01 (0.021)
Urbanization ratio	−0.90*** (0.152)	−1.62*** (0.310)	−1.03*** (0.162)	−1.69*** (0.305)
Log (household median income)	2.06*** (0.453)	1.54** (0.558)	2.31*** (0.477)	1.85** (0.571)
Median educational attainment	0.72*** (0.076)	0.88*** (0.103)	1.00*** (0.081)	1.14*** (0.102)
Log (population)	−1.76*** (0.448)	−0.44 (0.713)	−2.03*** (0.472)	−0.83 (0.710)
Constant	1.58 (1.273)	8.89*** (2.211)	1.04 (1.377)	7.83*** (2.197)
State fixed effect	√	√	√	√
Kleibergen–Paap rk LM statistic		32.89		32.89
Cragg–Donald Wald F statistic^a		17.37		17.37
Hansen J statistic^b		0.38		1.23
Endogeneity test^c		8.77		8.32
Adjusted R²	.44	.31	.48	.39

Note. Robust standard errors are in parentheses; coefficients for state dummy variables are not shown; excluded Instrumental variables: number of streams and the number of ranges. NPS = nonprofit sector; OLS = ordinary least square; 2SLS = two-stage least square; LM=Lagrange Multiplier.

Stock-Yogo weak ID test critical values (10% maximal IV relative bias): 9.08.

Critical value: χ²(df = 2, α = .05) = 5.99.

χ²(df = 1, α = .05) = 3.84.

†

p < .1. *p < .05. **p < .01. ***p < .001.

As shown in Tables 3 and 4, endogeneity test statistics are much higher than the critical value in all models, revealing that the government density is endogenously determined.⁷ Tests also suggest that the excluded instrumental variables work as appropriate instruments for the government density in all three models. First, Lagrange multiplie (LM) statistics reject the underidentification of the model, thus supporting the use of our instrumental variables in explaining the government density. Second, Cragg–Donald Wald F statistics reject the weak explanatory power of the instrument. Third, tests fail to reject the irrelevance of instrumental variables and unobserved effects, as suggested by the Sargan–Hansen statistics. As government density is endogenously determined, OLS regressions generate biased and inconsistent estimates. We will focus on interpreting the results of 2SLS estimations (Models 2, 4, 6, and 8) below.

Density of the NPS

Models 1 through 4 measure the size of the NPS using the logarithmic value of NPO density. While Models 1 and 2 use the number of tax-filing NPOs from the Digitized Data to calculate the NPO density, Models 3 and 4 use the data of registered NPOs from BMF.

The estimated coefficients of all variables aside from racial heterogeneity have consistent signs in the two 2SLS models of NPO density, although they may have different statistical significance. Thus, the two NPO density models basically tell the same story. Government fragmentation has a positive sign and is highly significant. The coefficients are 0.95 and 0.76, respectively, suggesting that ceteris paribus, a 1% increase in the government fragmentation (measured as government density) increases the density of tax-filing NPOs and registered NPOs by 0.95% and 0.76%, respectively. To better understand the magnitude of the impacts, we can convert the elasticity effect to the marginal effect. Based on Model 2, one more government per ten thousand capita will increase the NPO density by 2.22 tax-filing NPOs (0.95 × 14.68 / 6.29) per ten thousand capita, in which 14.68 and 6.29 are the mean values of NPO density and government density, respectively, and 0.95 is the elasticity coefficient. The government fragmentation dimension of fiscal decentralization has a positive coefficient, thus failing to support the government failure theory, which predicts a negative coefficient for government fragmentation. Instead, this result seems to be consistent with the interdependence theory, which suggests that governments and NPOs rely on each other for public service provision. In both 2SLS models, the county government share of total local government revenue—the measure for vertical centralization—has a positive sign and is significant at 99% significance level. Hence, the vertical centralization/decentralization result is consistent with the prediction of government failure theory.

The findings on the impacts of resident heterogeneity are not conclusive. Whereas the coefficients of religious heterogeneity and income heterogeneity (Gini index) are signed positive and statistically significant, educational heterogeneity has a significant and negative coefficient. Racial heterogeneity is not significant in Model 2, but is signed positive and statistically significant in Model 4. It suggests that racial heterogeneity may have an impact on small size NPOs but not on medium or large NPOs. Thus, income heterogeneity and religious heterogeneity are consistent with the predictions of government failure theory. In contrast, educational heterogeneities decrease the density of the NPO, which may be a result that the social cohesion effect suppresses the demand for nonprofits brought by government failure.

In Models 2 and 4, the coefficient of government expenditure per capita has the negative sign as predicted by the government failure theory. The coefficient is statistically significant in Model 2 but not in Model 4. Holding government grants to NPS per capita and other factors constant, a 1% increase of government expenditure per capita decreases NPO density by 0.16%, according to Model 2. This finding is consistent with the government failure theory prediction that higher government spending reduces the need for public service and consequently lowers the size of the NPS.

Government grants per capita have a highly significant and positive coefficient. Ceteris paribus, a 1% growth in government grants per capital leads to an increase of tax-filing NPOs and registered NPOs by about 0.05% and 0.02%, respectively. Based on Model 2, one extra dollar per capita of government grants in a county can increase the NPO density by about 0.048 (=0.05 × US$14.68/US$152.24) NPOs per ten thousand capita, in which US$14.68 and US$152.24 are the mean values of NPO density and government grants, respectively, and 0.05 is the elasticity coefficient. It is also equivalent to say that an extra US$208 (US$1/0.048) per capita government grants is needed to increase the number of one NPO per ten thousand capita. Converting the unit of the effect into one used in Lecy and Van Slyke (2013), we can predict that a government grants needed to add a NPO is US$2.08 million.⁸ This result is smaller than the US$4.7 million found by Lecy and Van Slyke, but their result is based on a sample of solely human service NPOs.

NPS Revenue and Assets

Coefficients in Model 6 have exactly the same signs and similar statistical significance as those in Model 8. Thus, it makes no difference to measure the NPS size using their revenue or assets. Most results for the NPS revenue and assets are also consistent with those from the NPO density models.

Government horizontal fragmentation has a positive and statistically significant coefficient. This result is contrast to the prediction of government failure theory. Although the coefficient of vertical centralization has a negative sign as predicted by the government failure theory, it is insignificant in Models 6 and 8. Thus, the vertical centralization increases the NPO density but not the revenue or assets of the NPS.

In the models of NPS revenue and assets, income heterogeneity and religious heterogeneity have a positive sign as predicted by government failure theory, but religious heterogeneity is not significant. Both educational heterogeneity and racial heterogeneity are statistically significant and have a negative sign. In sum, population heterogeneity again shows inconclusive results.

The coefficient of government expenditure is negative and statistically significant in Models 6 and 8. Thus, government expenditure per capita reduces the revenues and assets of the NPS. This finding supports government failure theory. Holding the ratio of government grants to NPS revenue and other factors constant, a 1% increase of government expenditure per capita decreases per capita NPS revenues and assets by 0.62% and 0.55%, respectively. In other words, one dollar per capita government spending reduces NPS revenue per capita and assets per capita by approximately US$0.31 and US$0.45, respectively.⁹

When NPS assets or revenue serves as the dependent variable, it is problematic to directly include government grants as a control variable, as government grants then becomes collinear with the dependent variable. Therefore, Models 6 and 8 use government grants as a ratio of the NPS revenue instead of the value of government grants per capita. Government grants ratio has a negative and significant coefficient in both Models 6 and 8, suggesting that ceteris paribus a NPS relying more heavily on government direct support is associated with a smaller size. There are two possible explanations for this result. First, it may be possible that smaller NPOs have less diverse revenue sources and rely more heavily on governmental support. Second, it may be a consequence of the “crowding out” effect in which direct government support crowds out other revenues of the NPS, leading to a smaller NPS, as documented by Andreoni (1993), Bergstrom et al. (1986), and Duncan (1999).¹⁰ Further examination is needed before a conclusion can be reached about the effects of government funding on other NPS revenue sources.

Most control variables have the expected sign in 2SLS models. Both median household income and educational attainment have a positive and highly significant impact on the size of the NPS. It is consistent with the expectation that richer and more educated residents are more likely to donate and participate in nonprofit service. It is worthwhile to emphasize the magnitude of the effect of median household income. On average, a 1% increase of median household income raises the density of tax-filing NPOs by 0.59%, the density of registered NPOs by 0.87%, NPS revenue per capita by 1.54%, and NPS assets by 1.85%. Equivalently, we can interpret the effects based on the marginal effects of household income. Using the marginal effect on NPS revenue as an example, we have 1.54 × US$1,396.73 / US$36,314.5 = US$0.06, in which US$1,396.73 and US$36,314.5 are the mean values of the NPS revenue and median household income, respectively. Our results thus suggest that every one dollar increase in median family income leads to approximately a 6-cent increase in NPO revenue. A higher family income can increase the NPS revenue by boosting private donation, program revenues and government grants and contracts. The predicted 6-cent increase of NPO revenue for every 1-dollar increase in median family income mainly captures the effect of family income on NPS revenue via private donation and program revenue, because government grants and government spending have been controlled for in the regression. Urbanization level and unemployment rate decrease the size of the NPS, as explicated in the section of “Measures and Data.” Median age has a significant positive impact on the density of NPOs but a negative impact on the assets and revenues of the NPS.

Conclusion

This study revisits the government failure theory by investigating the impacts of fiscal decentralization on the size of the NPS. Previous tests of the theory focus on the demand side of public services—the heterogeneity of population demands. We argue that the failure of governmental provision depends on both the demand side and the supply side of public service provision. Furthermore, the supply of public service differs in not only how much governments spend but also the diversity and the efficiency of their provisions. Because it has been documented that more decentralized governments can improve government efficiency and increase the heterogeneity of public services, government decentralization should decrease the need for nonprofit services if government failure theory can explain the size of the NPS.

Our analyses use a county-level data set to test this line of reasoning and utilize 2SLS instrumental regressions to eliminate the endogeneity bias. The findings, at best, partially support the government failure theory. Government expenditure, conditional on government grants to the NPS, is negatively associated with the size of the NPS. This finding is consistent with the prediction of the government failure theory. The findings on the two dimensions of government decentralization on the governmental failure theory are mixed. Contrary to the prediction, we find that government fragmentation increases the size of the NPS as measured by its density, revenue, or assets. The vertical centralization increases the NPO density as expected by the government failure theory but has no impacts on the NPS revenue or the NPS assets.

On the demand side, as found in most previous research, the relationship between population heterogeneity and the size of the NPS is inconclusive. Where evidence from income heterogeneity supports the government failure theory, educational heterogeneity provides no evidence for government failure theory. Religious heterogeneity supports the government failure theory when the size of the NPS is captured by the NPO density, but this heterogeneity measure rejects the theory when the NPS size is measured by its revenue and assets. This inconsistency, however, is not surprising. More religious diversity encourages residents to form more religious NPOs to satisfy their demands (Corbin, 1999). Conversely, the heterogeneity may also reduce the social cohesion, leading to less private contributions. While the former argument is consistent with the government failure theory, the latter can be explained by the social cohesion theory.

While this study focuses on testing government failure theory, it also has implications for complementary theory (i.e., interdependency theory), which emphasizes the collaborative relationship between the public sector and the NPS. As previously discussed, complementary theory argues that governments may delegate the production and the delivery of some public services to the NPS (Lecy & Van Slyke, 2013; Salamon, 1987). Government grants per capita, the most critical variable for interdependence theory, has an expected positive and highly significant coefficient. Under this theory, we may also expect a positive relationship between the number of NPOs and the number of governments, which is congruent with the empirical findings of this research.

As discussed in the introduction, understanding the government–nonprofit relationship is important for our policy design. The results of this study suggest that while it is still early to reject or accept government failure theory, interdependence theory is consistent with our findings. These two theories are not necessarily incompatible. Our results simply suggest that the government–nonprofit relationship is more responsive to the side of resource availability than to the demand for public service.

While the government fragmentation dimension is not signed as predicted by government failure theory, it is highly significant and the magnitude of the effect is substantial. For every ten thousand capita, one more government will increase 2.22 tax-filing NPOs; or 1% increase of government density leads to an increase of NPO density by 0.95%. On the contrary, governments need to provide extra grants of US$208, an increase of more than 135% of the current mean level at US$152 per capita, to increase one more NPO per ten thousand capita. Thus, government density seems to be a more sensitive factor for increasing the number of NPOs than does government grants. Because this effect comes after the variable of government grants is held constant, the channel for the impact of government density may be program revenues or contracts. While the mechanism of this relationship needs to be further explored, it may be a consequence that interdependence relationship outperforms the impacts caused by government failure. Regardless of the rationale of this complementary relationship, it needs to be taken into account when policies are designed.

The mixed findings on government failure theory suggest that our focal issue is more complicated than expected. Further examination is needed to better understand factors driving the existence of the NPS and its relationship with the public sector. For instance, our analysis is based on the aggregated data of all functional sectors in the NPS; yet, the impacts of government decentralization on the size of the NPS may depend on functional sectors. The positive relationship between fiscal decentralization and the NPS size may not be applied to functional sectors associated with cross-regional spillover. In fact, studies have found that while fiscal competition improves the efficiency of most public sectors, it may also impair the efficiency of public sectors characterized with externalities, such as social welfare and health services (see Oates, 1999, for a review). Different functional sectors of the NPS may also be related to governments in different fashions (Kim, 2015; Lecy & Van Slyke, 2013). As such, more insights may be gained through examining functional sectors separately instead of the entire NPS as a whole.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biography

Gao Liu is an assistant professor in the School of Public Administration at Florida Atlantic University. His research focuses on public finance, fiscal federalism, and issues related to issuance and market of municipal bonds. His work has been published in National Tax Journal, Public Budgeting and Finance, Public Finance Review and Municipal Finance Journal.

References

Abzug

Turnheim

J. K.

(1998). Bandwagon or band-aid? A model of nonprofit incorporation by state. Nonprofit and Voluntary Sector Quarterly, 27, 300-322.

Akai

Sakata

(2002). Fiscal decentralization contributes to economic growth: Evidence from state-level cross-section data for the United States. Journal of Urban Economics, 52, 93-108.

Alesina

Baqir

Hoxby

(2004). Political jurisdictions in heterogeneous communities. The Journal of Political Economy, 112, 348-396.

Andreoni

(1993). An experimental test of the public-goods crowding-out hypothesis. American Economic Review, 83, 1317-1327.

Arikan

(2004). Fiscal decentralization: A remedy for corruption? International Tax and Public Finance, 11, 175-195.

Barankay

Lockwood

(2007). Decentralization and the productive efficiency of government: Evidence from Swiss cantons star, open. Journal of Public Economics, 91, 1197-1218.

Bates

L. J.

Santerre

R. E.

(2006). Leviathan in the crosshairs. Public Choice, 127, 133-145.

Ben-Ner

Van Hoomissen

(1992). An empirical investigation of the joint determination of the size of the for-profit: Nonprofit and government sectors. Annals of Public and Cooperative Economics, 63, 391-415.

Bergstrom

T. C.

Blume

L. E.

Varian

H. R.

(1986). On the private provision of public goods. Journal of Public Economics, 29, 25-49.

10.

Blackwood

Roeger

Pettijohn

(2012). The nonprofit almanac 2012. Washington, DC: Urban Institute.

11.

Boris

E. T.

(1999). Nonprofit organizations in a democracy: Varied roles and responsibilities. In Boris

E. T.

Steuerle

E. C.

(Eds.), Nonprofits and government: Collaboration and conflict (pp. 1-33). Washington, DC: Urban Institute Press.

12.

Boris

E. T.

Steuerle

E. C.

(1999). Nonprofits and government: Collaboration and conflict. Washington, DC: Urban Institute Press.

13.

Burns

(1994). The formation of American local governments: Private values in public institutions. Oxford, UK: Oxford University Press.

14.

Corbin

J. J.

(1999). A study of factors influencing the growth of nonprofits in social services. Nonprofit and Voluntary Sector Quarterly, 28, 296-314.

15.

Duncan

(1999). Modeling charitable contributions of time and money. Journal of Public Economics, 72, 213-242.

16.

Fisman

Gatti

(2002). Decentralization and corruption: Evidence across countries. Journal of Public Economics, 83, 325-345.

17.

Gronbjerg

K. A.

(1993). Understanding nonprofit funding. San Francisco, CA: Jossey-Bass.

18.

Gronbjerg

K. A.

Paarlberg

(2001). Community variations in the size and scope and the nonprofit sector: Theory and preliminary findings. Nonprofit and Voluntary Sector Quarterly, 30, 684-706.

19.

Hall

P. D.

(1987). A historical overview of the private nonprofit sector. In Powell

W. W.

Steinberg

Clemens

E. S.

(Eds.), The nonprofit sector: A research handbook (pp. 32-65). New Haven, CT: Yale University Press.

20.

Hansmann

(1987). Economic theories of nonprofit organization. In Powell

W. W.

Steinberg

Clemens

E. S.

(Eds.), The nonprofit sector: A research handbook (pp. 27-42). New Haven, CT: Yale University Press.

21.

Hayashi

(2000). Econometrics. Princeton, NJ: Princeton University Press.

22.

Hoxby

C. M.

(2000). Does competition among public schools benefit students and taxpayers? American Economic Review, 90, 1209-1238.

23.

Iimi

(2005). Decentralization and economic growth revisited: An empirical note. Journal of Urban Economics, 57, 449-461.

24.

James

(1987). The nonprofit sector in comparative perspective. In Powell

W. W.

Steinberg

Clemens

E. S.

(Eds.), The non-profit sector: A research handbook (pp. 43-54). New Haven, CT: Yale University Press.

25.

Khaleghian

(2004). Decentralization and public services: The case of immunization. Social Science & Medicine, 59, 163-183.

26.

Kim

(2015). Socioeconomic diversity, political engagement, and the density of nonprofit organizations in U.S. counties. The American Review of Public Administration, 45, 402-416.

27.

Lampkin

Boris

(2002). Nonprofit organization data: What we have and what we need. American Behavioral Scientist, 45, 1675-1715.

28.

Lecy

J. D.

Van Slyke

D. M.

(2013). Nonprofit sector growth and density: Testing theories of government support. Journal of Public Administration Research and Theory, 23, 189-214.

29.

Lin

J. Y.

Liu

(2000). Fiscal decentralization and economic growth in China. Economic Development and Cultural Change, 49, 1-21.

30.

Luksetich

(2008). Government funding and nonprofit organizations. Nonprofit and Voluntary Sector Quarterly, 37, 434-442.

31.

Matsunaga

Yamauchi

(2004). Is government failure theory still relevant? A panel analysis using U.S. state level data. Annals of Public & Cooperative Economics, 75, 227-263.

32.

Oates

W. E.

(1985). Searching for leviathan: An empirical study. The American Economic Review, 75, 748-757.

33.

Oates

W. E.

(1999). An essay on fiscal federalism. Journal of Economic Literature, 37, 1120-1149.

34.

Olson

(1965). Logic of collective action public goods and the theory of groups. Cambridge, MA: Harvard University Press.

35.

Paarlberg

L. E.

Gen

(2009). Exploring the determinants of nonprofit coproduction of public service delivery. The American Review of Public Administration, 39, 391-408.

36.

Prud’Homme

(1995). The dangers of decentralization. The World Bank Research Observer, 10, 201-220.

37.

Rose-Ackerman

(1982). Charitable giving and “excessive” fundraising. The Quarterly Journal of Economics, 97, 193-212.

38.

Salamon

L. M.

(1987). Of market failure, voluntary failure, and third-party government: Toward a theory of government-nonprofit relations in the modern welfare state. Nonprofit and Voluntary Sector Quarterly, 16, 29-49.

39.

Salamon

L. M.

(1995). Partners in public service: Government-nonprofit relations in the modern welfare state. Balitmore, MD: John Hopkin University Press.

40.

Salamon

L. M.

Anheier

H. K.

(1998). Social origins of civil society: Explaining the nonprofit sector cross-nationally. Voluntas: International Journal of Voluntary and Nonprofit Organizations, 9, 213-248.

41.

Salamon

L. M.

Sokolowski

S. W.

Anheier

H. K.

(2000, December). Social origins of civil society: An overview (Working papers of the Johns Hopkins Comparative Nonprofit sector Project). Baltimore, MD: Johns Hopkins University.

42.

Saxton

G. D.

Benson

M. A.

(2005). Social capital and the growth of the nonprofit sector. Social Science Quarterly, 86, 16-35.

43.

Schiff

(1985). Does government spending crowd out charitable contributions. National Tax Journal, 38, 535-546.

44.

Seabright

(1996). Accountability and decentralisation in government: An incomplete contracts model. European Economic Review, 40, 61-89.

45.

Seaman

B. A.

(1981). Economic theory and the positive economics of arts financing. The American Economic Review, 71, 335-340.

46.

Smith

S. R.

Gronbjerg

K. A.

(2006). Scope and theory of government-nonprofit relations. The Nonprofit Sector: A Research Handbook, 2, 221-242.

47.

Sow

Razafimahefa

M. I. F.

(2015). Fiscal decentralization and the efficiency of public service delivery (No. 15-59). Washington, DC: International Monetary Fund.

48.

Tiebout

C. M.

(1956). A pure theory of local expenditures. Journal of Political Economics, 64, 416-424.

49.

U.S. Census Bureau. (2002). Census of governments: Government units in 2002. Retrieved from http://www2.census.gov/govs/cog/2002COGprelim_report.pdf

50.

Warr

P. G.

(1982). Pareto optimal redistribution and private charity. Journal of Public Economics, 19, 131-138.

51.

Weisbrod

B. A.

(1977). The voluntary nonprofit sector: An economic analysis. Lexington, MA: Lexington Books.

52.

Weisbrod

B. A.

(1991). The nonprofit economy. Cambridge, MA: Harvard University Press.

53.

Wilson

W. J.

(1996). When work disappears: The world of the new urban poor. New York, NY: Knopf.

54.

Wolpert

(1993). Patterns of generosity. New York, NY: The Twentieth Century Fund.

55.

Xie

Zou

Davoodi

(1999). Fiscal decentralization and economic growth in the United States. Journal of Urban Economics, 45, 228-239.

56.

Young

(2000). Government failure theory. In Ott

J. S.

(Ed.), The nature of the nonprofit sector (pp. 190-197). Boulder, CO: Westview Press.

57.

Zax

J. S.

(1989). Is there a leviathan in your neighborhood? The American Economic Review, 79, 560-567.

58.

Zhang

Zou

(1998). Fiscal decentralization, public spending, and economic growth in China. Journal of Public Economics, 67, 221-240.