Public Welfare Spending and Private Social Services in U.S. States

Abstract

Social services in the United States are supplied by both the public and private sectors. Previous political science research has focused on public transfers or Medicaid; I study “Other Public Welfare” programs that include contracts and grants to private social services providers, focusing on the relationship between the two sectors. My results imply that an increase in either Other Public Welfare spending or private individual and family services employees leads to an increase in the other sector. I find weaker evidence of similar relationships involving private residential care or day care services, and private social services employment is generally independent of public spending on transfers and Medicaid. My results have implications for the full effects of changes in public welfare spending, including the effects on the private sector, as well as the effects of organized interests on public welfare spending.

Keywords

public welfare spending social services nonprofit organizations interest groups

Although it is customary in capitalist countries to refer to separate public and private sectors, there are some spheres of activity where both sectors operate and may even engage in joint production. One such activity involves the provision of social (or human) services. Moreover, most of the actual delivery of social services (as opposed to their funding) occurs at the state and local level. In 2000, there were a total of just 9,028 federal employees engaged in providing “public welfare” services compared with 501,997 full-time equivalent state and local government employees (U.S. Census Bureau 2000b). Also in 2000, the private social services organizations analyzed in this article employed a total of more than 2.6 million persons during the pay period including March 12 (U.S. Census Bureau. Various-c).

This mixture of public and private service provision evolved over the past century. Social services were provided mainly by the private sector in the late nineteenth and early twentieth centuries, albeit with some subsidies from state and local governments (Grønbjerg 2001; Kramer 1987; Smith 2002). Beginning in the 1930s, federal government programs such as Aid to Families with Dependent Children (AFDC), Medicaid, and later Temporary Assistance for Needy Families (TANF) assumed major importance. By the 1980s, some opponents of the welfare state argued that private charitable organizations would expand to offset a reduction in public spending, but scholars such as Salamon (1995) emphasized that many services provided by private agencies were funded in part by government grants and contracts.¹ Still more recently, there have been claims that private social services agencies seek to influence public welfare policy and spending (J. M. Berry with Arons 2003; U.S. Congress 1995).

Despite the long-standing interconnection between the public and private social services sectors, little systematic empirical research examines the effect of changes in one sector on the other. Political science studies of public welfare policy mostly concentrate on transfer programs or Medicaid, and test for the effects of interparty or interstate competition (e.g., Bailey and Rom 2004; Barrilleaux, Holbrook, and Langer 2002; W. D. Berry and Baybeck 2005; Brown 1995; Fellowes and Rowe 2004; Volden 2002). Salamon’s (1995) work documenting government support for private nonprofit organizations is mostly descriptive, while multivariate studies of the variation in private nonprofit organizations across geographic units do not focus specifically on social services and either do not include measures of government spending (Corbin 1999; Gamm and Putnam 1999) or use very crude measures (Grønbjerg and Paarlberg 2001).²

I begin to fill this gap in the literature by using data on U.S. states during 1989–2000 to estimate dynamic models of state and local government public welfare spending and private social services employment. I ask,

What are the determinants of state and local government spending on public welfare programs that include grants and contracts with private sector providers? In particular, is public spending affected by the size of the private social services sector?

What are the determinants of the size of the private social services sector? In particular, what is the effect of spending on various types of public welfare programs?

In either case, how large are the effects?

Understanding the relationship between public and private sectors is especially important if we want to predict the full effects of a significant change in public welfare spending. The scale and composition of social services available to the public depends on both sectors, so the full consequences of a large cut in a particular kind of government spending, for example, depend on whether the private sector expands to compensate, contracts due to a drop in government grants and contracts, or remains unchanged. We would also like to know whether government spending on public welfare programs that provide funding for private organizations is influenced by these organized interests themselves.

My results vary depending on the type of social services analyzed, but the clearest finding is that government spending on nonmedical vendors and “Other Public Welfare” programs and employment in organizations providing individual and family services (IFS) are complements. Dynamic statistical models imply causal effects in both directions, although error correction models imply that the effect of public spending on private employees is short-lived. Results for other types of private social services are similar but stop short of statistical significance, with the exception of child day care services during 1990–96. Estimates of the magnitudes of the effects suggest that an increase in public funding may partially crowd out private organization revenues from other sources. I also find some evidence that private IFS act as a substitute for public spending on transfers, but the coefficients do not reach statistical significance at conventional levels and the estimated magnitudes of the effects are small.

Although my main focus is on the interaction between public and private sectors, I also obtain some interesting results regarding their exogenous determinants. Personal preferences (ideology) are determinants of public sector spending and private social services, but in opposite directions: State and local government spending on Other Public Welfare programs tends to be greater in states with more liberal political elites, but private social services employment tends to be greater in states with more conservative citizens. Consistent with Peterson’s (1981) theory regarding redistributive spending, spending on Other Public Welfare programs does not depend on measures of need. Contrary to the implications of some other research (Alesina, Baqir, and Easterly 1999; Corbin 1999), I find little evidence that social cohesion as measured by racial heterogeneity matters for either sector.

The following two sections summarize previous research and tentative hypotheses for the determinants of public welfare spending on programs that involve third-party providers and the private social services sector, respectively. Next, I discuss data sources and measurement issues. The empirical analysis first presents dynamic models estimated for public and private sectors using data from all states except Alaska for two separate time periods: 1989 or 1990–96, and 1999–2000. I then discuss the estimated magnitudes of the effects for selected variables and exploratory analysis of the effects of religious adherents in each state. The conclusion summarizes my findings and suggests questions for further research.

Explaining Government Support for Third-Party Providers

State and local government public welfare services are delivered through several different channels, including transfer programs, direct provision of services, means-tested health insurance, and payments to third-party suppliers. As noted above, previous political science research focuses on transfers (generally AFDC or TANF) and health insurance (Medicaid). This article focuses on programs that are funded by taxes but implemented by private providers supported in part by grants and contracts. What might explain the variation in state and local government spending on these types of public welfare programs?

Public choice theory implies that the potential recipients of government grants and contracts will advocate for increased spending (Mueller 2003). This could be motivated either by a desire to increase their own revenues and personal benefits, or a desire to increase services to their clients (Walker 1991). It may be argued that this logic does not apply here, as nonprofit organizations account for most of the revenues and employees in the private social services sector (Grønbjerg 2001; Smith 2002). Not only are staff at nonprofit organizations limited in their ability to pocket financial rents but also organizations that qualify as “philanthropic” nonprofits are subject to specific restrictions on their ability to lobby and are prohibited from engaging in partisan campaign activities (J. M. Berry with Arons 2003).

Nonetheless, there is ample reason to hypothesize that an increase in the size of the private social services sector leads to an increase in spending on grants and contracts. There are many ways to influence public spending that do not fall within the legal definition of lobbying, and public administrators may enlist the help of private organizations to gain support for their programs (J. M. Berry with Arons 2003; Smith and Lipsky 1993). Private social services providers may also form “peak” state associations that engage in more traditional lobbying and employees, volunteers or board members are free to engage in advocacy as individuals. In 1995, the U.S. House of Representatives passed the so-called “Istook Amendment,” which would have dramatically tightened the restrictions on advocacy by nonprofit organizations receiving government grants. Proponents argued that these organizations were diverting resources away from services to advocate successfully for more grants (Reid 1999; U.S. Congress 1995). J. M. Berry and Arons conclude based on interviews and a mail survey of nonprofit human services organizations that “in state and local politics, the lobbying of politically active nonprofits is obvious to anyone following politics in that venue” (2003, 67). They note that nonprofit organizations advocating for redistributive programs often face little opposition as “No one organizes to work against the frail elderly” (J. M. Berry with Arons 2003, 94).

Previous studies of redistributive spending lack measures of organized interests, but they do suggest a number of control variables and their expected effects. Peterson (1981) argues that redistributive spending by subnational governments should depend primarily on available resources rather than measures of need. He theorizes that governments concerned with the out-migration of taxpayers will concentrate on “developmental” services that enhance the economic position of the community and “allocational” services that are economically neutral and provide relatively uniform benefits. Although Peterson formulated his theory in terms of local governments, he tested it with data on combined state and local government spending. Subsequent research has challenged Peterson’s central claim even at the local level (e.g., Percival, Johnson, and Nieman 2009). If Peterson is correct, then public welfare spending should increase with per capita income and federal transfers, but be unaffected by the share of residents in vulnerable age groups. If, however, demand matters, then the effect of income may be zero or negative, while spending should increase with the shares of the population who are children or seniors.

More recent research on local government spending on redistributive programs has focused on social cohesion, or the lack thereof. Alesina, Baqir, and Easterly (1999) show that more ethnically fragmented cities and counties in the United States devote smaller shares of public spending to redistribution and productive public goods, and larger shares to crime prevention and patronage. They argue that fragmentation results in a lack of social cohesion, which in turn leads to a lack of support for redistributive programs. If they are correct, and if racial and ethnic homogeneity is a proxy for social cohesion, then spending on public welfare programs should be greater in states with higher percentages of non-Hispanic whites.

Ideology is another potential explanatory variable. Peterson’s (1981) argument implies that the personal preferences of policy makers should not matter at the local level, but previous studies of state transfer programs and Medicaid typically include controls for ideology (Bailey and Rom 2004; Barrilleaux, Holbrook, and Langer 2002; Fellowes and Rowe 2004; Volden 2002). In the case of programs implemented through grants and contracts with private agencies, the mechanism for delivery makes things more complicated. Whereas liberals should prefer more spending on public welfare programs in general, conservatives might prefer that services be delivered through private, third-party providers rather than directly by the government. I therefore control for ideology, but I am agnostic about the expected effect on spending on programs that are implemented by private providers.

Two other variables frequently included in studies of transfer programs and Medicaid are interparty competition and comparable policy outputs in neighboring states (e.g., Bailey and Rom 2004; Barrilleaux, Holbrook, and Langer 2002; W. D. Berry and Baybeck 2005; Brown 1995; Fellowes and Rowe 2004; Volden 2002). The theoretical argument for the former is that noncompetitive states should tend to have less generous public welfare programs because incumbents do not need to compete for the votes of the “have nots” to remain in office. The argument for the latter is that politicians do not want to encourage immigration by people with low incomes in search of generous benefits. Neither argument applies as well to public spending on programs implemented through grants and contracts for two reasons. First, these programs often provide services such as foster care or transportation for the disabled, where eligibility is limited to people with specific conditions. Second, the connection between public spending and the level of benefits received by individuals is indirect, as government funds are not the only source of revenue for private social services providers. I therefore omit these variables from my main analysis, but report on alternative specifications in the section summarizing my results.

Explaining Variation in Private Social Services

Resource dependence theory implies that the size of the private social services sector should depend on available resources, whereas market failure theory implies it should be greater where there is greater demand for services not provided by for-profit businesses (Corbin 1999). Resources include both public spending on grants and contracts and per capita income. Low per capita income could also be a measure of demand for services, as is the share of the population in vulnerable age groups.

Although previous research has argued that spending on programs involving grants and contracts should affect the scale of private social services (Salamon 1995; Smith and Lipsky 1993), empirical tests of the effect are lacking. This effect could be at least partially offset for two reasons. First, managers of nonprofit social services organizations who are not revenue maximizers may respond to increased public funding by decreasing their efforts to raise money from other sources (Lowry 1997). Second, private donors who are aware that a nonprofit organization is receiving public funds may perceive less need for their donations (Brooks 2000). Other types of public welfare spending may have different effects. Spending on transfer programs and direct services might act as a substitute for services provided through third-party organizations. Medicaid should primarily affect private health care providers, but there may be an effect on social services as well depending on what is covered by Medicaid and which private organizations are included.

Other theoretical discussions of factors that might affect the size of the private social services sector highlight the importance of social cohesion or heterogeneity, although this can cut both ways. On one hand, racial and ethnic diversity might indicate demand for specialized services that are not likely to be provided directly by government. On the other hand, a lack of social cohesion might lead to a failure to voluntarily supply services (Corbin 1999; Grønbjerg and Paarlberg 2001).

Previous research has also identified other variables that should be included as controls. The percentage of the population living in metropolitan areas might have demand- and supply-side effects. One one hand, there may be different levels of demand for certain kinds of services in metropolitan areas; on the other hand, the number of organizations or employees per capita required to provide services may be less in densely populated areas (Gamm and Putnam 1999; Lowry 2005). The number of nonprofit political and civic organizations in a county or state has also been found to depend on human capital as measured by the share of college graduates (Grønbjerg and Paarlberg 2001; Lowry 2005), and ideology (Lowry 2005). In the case of private social services organizations, ideology could cut different ways. Whereas liberals might be more likely to support some kinds of social services, conservatives might prefer supporting private nonprofit organizations over paying taxes.

Finally, religious organizations and motives have played major roles in the founding of nonprofit organizations in general and social services organizations in particular (James 1987; Smith 2002), although again there are a variety of explanations. Corbin (1999) finds that religious diversity and religious adherents per capita have positive associations with nonprofit organizations per capita in metropolitan areas, but Grønbjerg and Paarlberg (2001) cannot replicate his results for Indiana counties. Wuthnow (1999) analyzes survey responses and finds that volunteering for nonreligious nonprofit organizations differs by denomination. Mainline Protestants engage in more of this activity than do Catholics and much more than evangelical Protestants. Wuthnow argues that this is because evangelicals are more likely to participate in activities that provide services through their church. Lowry (2005) obtains consistent results using state aggregate data on political and civic nonprofit organizations. Unfortunately, relatively comprehensive data on the number of persons affiliated with different religious denominations in each state are only collected at 10-year intervals (Association of Religion Data Archives 2011). I therefore must omit measures of religious affiliation from my primary dynamic analysis.

Data and Measurement

Empirical analysis is complicated by the fact that there is no single data source that includes public and private sectors, and the classification schemes used for public spending and private organizations do not match up perfectly. Data on public welfare spending come from U.S. Census Bureau reports for state and local government finances (U.S. Census Bureau Various-a). Public welfare is divided into six functions: federal categorical assistance programs (AFDC or TANF), other cash assistance programs, medical vendor payments (Medicaid), nonmedical vendor payments, public welfare institutions, and Other Public Welfare.³ Federal categorical assistance, payments to medical vendors, and Other Public Welfare account for more than 95% of the total for the years in my data set.

I created three variables to measure public welfare spending: Federal categorical assistance, other cash assistance, and welfare institutions are combined to form transfers and direct services (“transfers” for short); medical vendor payments are simply labeled Medicaid; and nonmedical vendor payments are combined with Other Public Welfare. The last category is my key spending variable of interest. Examples of Other Public Welfare include the following:

Administration of: medical and cash assistance, general relief, vendor, and other welfare programs; regulation and support of private welfare institutions and activities; all intergovernmental payments for welfare other than for cash assistance programs; children services, such as foster care, adoption, day care, nonresidential shelters, and the like; activities supported by Federal Social Services Block Grant (Title XX) funds; low-income energy assistance and weatherization (note—administrative expenditure only. . .); welfare-related community action programs; social services to the physically disabled, such as transportation; temporary shelters and other services for the homeless; intergovernmental payments to public hospitals for medical assistance other than under the Medicaid program. (U.S. Census Bureau 2010)

Although this category obviously is not limited to programs that fund third-party providers, it does include those programs, and it excludes spending on transfers, direct services provided by the government, and health insurance. Health insurance also involves payments to private third-party providers, but Medicaid operates as an entitlement program, and the main policy instruments are eligibility criteria and levels of benefits, not aggregate spending (Bailey and Rom 2004). An analysis of the influence of private health care providers on Medicaid and vice versa might well be worth doing, but that is a separate analysis.

I combine spending by state and local governments similar to Peterson (1981) and Primo (2007) because the division of responsibility for public welfare spending varies greatly across states. I omit Alaska because of large swings in both Other Public Welfare spending and the division between state and local governments in the early 1990s.⁴

Data on private social services employees come from County Business Patterns (U.S. Census Bureau Various-c). Data are classified by industry using Standard Industrial Classification Codes (SICC) through 1997 and the North American Industrial Classification System (NAICS) beginning in 1998. The main difference is that the newer scheme has more detail, and the Census Bureau provides a table allowing them to be cross-walked.

I created three variables for social services employees in organizations providing IFS, residential care services, and child day care.⁵ IFS appear to be the most closely related to redistributive public welfare programs, whereas child day care services are the least closely related. In 1997, more than 90% of IFS employees worked for nonprofit organizations compared with 58% of residential care employees and just 38% of child day care employees (Smith 2002). Although Title XX block grants include funding for child care, many child care providers do not receive government funding, and Head Start programs not affiliated with schools are included in child care but their public funding comes under education.

Many other studies measure variation in organized private interests across subnational jurisdictions by the number of organizations, perhaps because these are the only available data (e.g., Corbin 1999; Gamm and Putnam 1999; Gray and Lowery 1996; Grønbjerg and Paarlberg 2001; Lowry 2005; Lowry and Potoski 2004; Luksetich 2008). However, the number of organizations may not be the most appropriate measure of the aggregate level of private sector services or their capacity to influence government decisions (Lowry and Potoski 2004). For example, it seems likely that several small private organizations would have less influence on government policy than a single large organization, due to coordination problems and free riding among the private organizations. In addition, the number of “establishments” reported in County Business Patterns is not the same thing as the number of organizations. Payroll data are also available, but they do not fully capture the aggregate budgets of these organizations and they reflect differences in pay scales across both states and types of social services.

I therefore use the number of employees per 1,000 state residents to measure the size of the private social services sector, and spending per capita adjusted for inflation to measure the public sector. Data are also available on public sector employees (U.S. Census Bureau 2000b), but they are aggregated by all public welfare functions. Moreover, spending is a better measure of the overall size of the public sector and its potential influence on private sector organizations that may be partially funded by government grants or contracts.

Models are estimated for two separate time periods: 1989 or 1990–96 and 1999–2000. The break coincides with the implementation of federal welfare reform in 1997 and the change in private employment classification systems in 1998. Data for the “after” period are limited to just 2 years because my models employ lagged values and the Census Bureau did not conduct its survey of local government finances in 2001 and 2003, so data on combined state and local government spending are not available for those years.

Table 1 presents summary statistics for my dependent and independent variables. For 1990–96, the bivariate correlations between spending on Other Public Welfare and social services employees are .607 for IFS, .565 for residential care services, and just .121 for child day care. For 1999–2000, the comparable correlations are .555, .601, and .109, respectively. While all these correlations are positive, they suggest that the connection between public spending and private services is strongest for IFS and residential care.

Table 1.

Summary Statistics

	1990–96 (N = 343)		1999–2000 (N = 98)
	M	SD	M	SD
Spending on Other Public Welfare and nonmedical vendors	151.8	58.3	186.0	75.3
IFS employees	2.0	0.8	3.5	1.3
Residential care employees	2.0	0.8	3.9	1.5
Day care employees	1.9	0.5	2.5	0.6
Transfer spending	99.6	54.3	68.6	43.2
Medicaid spending	434.3	153.0	533.2	153.1
Per capita income	24,122	3,605	28,186	4,298
Pct. old or young	39.2	2.1	38.4	1.4
Pct. non-Hispanic white	80.7	13.8	76.7	14.8
Pct. college graduates	21.4	4.3	24.8	4.5
Pct. metropolitan population	67.6	21.0	68.4	20.4
Federal HHS transfers	98.6	44.0	108.4	29.0
Citizen liberalism	48.1	13.5	47.3	14.5
Government liberalism	50.8	24.1	42.4	26.8

Notes: IFS = individual and family services. Pct. = Percentage. HHS = Health and Human Services. Alaska is omitted. State and local government spending, per capita income, and federal HHS transfers are measured in real (2,000) dollars per capita. Employees are per 1,000 state residents. Pct. old or young includes those ages 65 and above or 0 to 17.

Empirical Analysis

My primary empirical analysis employs two different types of dynamic models estimated for two different time periods. I estimate models of state and local government spending on Other Public Welfare as a function of social services employment in IFS, residential care, and child day care plus control variables.⁶ I estimate models of private social services employment as a function of state and local government spending on Other Public Welfare, transfers and direct services, and Medicaid plus control variables. For child day care services I use the percentage of the population younger than 18 instead of younger than 18 or older than 64. The two estimation techniques are Arellano–Bond models and error correction models.

Arellano–Bond Models

Arellano–Bond model is a generalized method of moments approach that was developed to address estimation issues raised by dynamic panel-data models with fixed unit effects, possible endogeneity, and a limited number of time periods (Arellano and Bond 1991). All these elements are present in my data. It assumes that the relationship to be modeled may be written in levels as

Y_{i t} = α_{0} + α_{1} Y_{i t - 1} + X_{i t} β + w_{t} + v_{i} + ε_{i t} \cdot

Subscripts i and t denote cross-sectional units and time periods, respectively; X_it and β may be vectors; w_t is a fixed time-period effect; v _i is a fixed unit effect; and ϵ_it is the residual. Subtracting Y_it−1 from both sides gives an expression in first differences where v_i has dropped out:

Δ Y_{i t} = α_{1} Δ Y_{i t - 1} + Δ X_{i t} β + Δ w_{t} + Δ ε_{i t} \cdot

This is the equation that is estimated. Endogeneity and autoregressive residuals are dealt with by using lagged levels of the dependent variable and predetermined variables, and lagged differences in strictly exogenous variables as instruments (see Wawro 2002).

Table 2 shows the results for state and local government spending on Other Public Welfare. Private social services employment is treated as endogenous; excluded instruments include the first differences of college graduates, percentage metropolitan population, citizen liberalism, and state and local government spending on transfers and Medicaid. For the early period (1990–96), the lagged changes in spending, IFS employees, and state government liberalism all have positive coefficients that are significant at the 99% confidence level. The coefficients on racial homogeneity and federal transfers are positive and not quite significant at the 90% level, whereas the coefficient on the percentage of the population that is young or old is (surprisingly) negative and significant at the 95% level. The coefficients on day care employees and per capita income are positive and exceed their standard errors. For the later period (1999–2000), the coefficients on lagged spending, IFS employees, and federal transfers are positive and significant at the 95% level. The coefficients on residential care employees and government liberalism are positive and exceed their standard errors.

Table 2.

Change in State and Local Government Spending per Capita on Nonmedical Vendors and Other Public Welfare: Arellano–Bond Models

	1990–96	1999–2000
Δ Spending_t−1	0.384** (6.39)	0.183** (2.17)
Δ IFS employees	18.25** (4.20)	35.16** (3.92)
Δ Residential care employees	−4.98 (0.87)	14.00 (1.56)
Δ Day care employees	9.79 (1.20)	−8.72 (0.57)
Δ Per capita income	0.002 (1.19)	7.3e–4 (0.20)
Δ Pct. old or young	−2.27** (2.22)	3.40 (0.70)
Δ Pct. non-Hispanic white	1.86 (1.46)	0.63 (0.29)
Δ State government liberalism	0.200** (2.63)	0.166 (1.22)
Δ Federal non-Medicaid transfers	0.082 (1.62)	0.160** (2.33)
Fixed year effects	Yes	Yes
Observations	343	98
States	49	49
Wald χ²	328.5	49.2
Degrees of freedom	15	10

Notes: IFS = individual and family services. Figures in parentheses are absolute t-ratios. Changes in IFS, residential care, and day care employees are treated as endogenous. Additional instruments include the first differences in pct. college graduates, pct. living in metropolitan areas, citizen liberalism, transfers, and Medicaid spending.

p < .10. **p < .05 (two-tailed test).

Table 3 shows the Arellano–Bond results for private social services employees, with public spending on Other Public Welfare treated as endogenous. Excluded instruments include changes in government liberalism and federal transfers. The coefficients on Other Public Welfare spending are uniformly positive; they are significant at the 99% level in both time periods for IFS employees, greater than their standard errors in both time periods for residential care employees, and significant at the 95% level in 1990–96 for day care employees. The coefficients on Medicaid spending are positive and significant at the 90% level or better for IFS and residential care employees during 1990–96. The other consistent result is that the coefficients on citizen liberalism are all negative; they are statistically significant at the 90% level in three of six equations and marginally significant (|t| > 1.5) in two others.

Table 3.

Change in Private Social Services Employees per Capita: Arellano–Bond Models

	IFS		Residential care		Day care
	1990–96	1999–2000	1990–96	1999–2000	1990–96	1999–2000
Δ Employees_t−1	0.713** (12.58)	−0.145 (1.00)	0.508** (7.92)	0.037 (1.47)	0.724** (12.70)	0.121 (0.91)
Δ Other Public Welfare spending	1.5e–3** (2.57)	3.7e–3** (3.00)	6.5e–4 (1.27)	1.5e–3 (1.28)	6.9e–4** (2.21)	4.5e–4 (0.60)
Δ Transfers	−1.6e–3 (1.60)	−2.1e–3 (1.29)	1.9e–5 (0.02)	9.5e–4 (0.60)	−8.2e–5 (0.15)	−1.4e–3 (1.37)
Δ Medicaid spending	4.6e–4** (2.77)	−4.5e–4 (0.69)	2.5e–4* (1.66)	−4.8e–4 (0.84)	−3.8e–5 (0.42)	5.5e–5 (0.17)
Δ Per capita income	1.3e–5 (0.61)	3.6e–5 (1.31)	9.2e–6 (0.48)	5.9e–5* (1.78)	5.0e–5** (4.49)	9.2e-5** (3.98)
Δ Pct. ages 0–17	—	—	—	—	038** (2.88)	−0.008 (0.18)
Δ Pct. old or young	0.017 (1.50)	0.066 (1.13)	0.010 (1.01)	0.129** (2.25)	—	—
Δ Pct. college graduates	−8.8e–4 (0.11)	0.028* (1.83)	0.020** (2.76)	0.009 (0.57)	0.004 (0.90)	0.003 (0.34)
Δ Pct. metropolitan population	2.5e–3 (0.23)	−0.101 (1.31)	0.010 (1.01)	0.135* (1.78)	0.006 (0.94)	−0.055 (1.15)
Δ Pct. non-Hispanic white	−0.022 (1.45)	−0.019 (0.69)	−0.011 (0.79)	−0.026 (0.93)	−0.009 (1.15)	0.033** (2.59)
Δ Citizen liberalism	−0.004 (1.53)	−0.002 (0.41)	−0.007** (2.94)	−0.008 (2.94)	−0.004** (2.64)	−0.005* (1.74)
Fixed year effects	Yes	Yes	Yes	Yes	Yes	Yes
Observations	343	98	343	98	343	98
States	49	49	49	49	49	49
Wald χ²	1,609.2	93.4	927.5	111.3	2,943.1	50.5
Degrees of freedom	16	11	16	11	16	11

Notes: IFS = individual and family services. Figures in parentheses are absolute t-ratios. The change in Other Public Welfare spending is treated as endogenous. Additional instruments include the first differences in government liberalism and non-Medicaid federal transfers.

p < .10. **p < .05 (two-tailed test).

Several interesting results emerge from Tables 2 and 3. First, results for public spending and private employment are generally consistent across time periods, despite changes in federal welfare policy and the classification scheme for private employees. Second, the coefficients on residential care and child day care employees are not significant in the public spending equations, whereas the coefficient on spending on transfers and direct services is never significant in the private employment equations. Third, the results imply that Other Public Welfare spending and IFS employees are complements, with causation running both ways. Nonetheless, the two sectors appear to respond differently to ideology. Public sector spending is higher in states with more liberal political elites (at least during 1990–96), whereas private sector social services employment is higher in states with more conservative citizens. Finally, the only coefficient in any model that has an unexpected sign and is statistically significant is the percentage old or young in the public sector spending model for 1990–96.

Error Correction Models

The Arellano–Bond model addresses several of the technical issues raised by my data, but it also begins with certain assumptions about the dynamic relationships between explanatory and dependent variables. De Boef and Keele (2008) argue that substantive theory typically is not sufficient to determine the exact specification of a dynamic model, so it is better to use a general autoregressive distributed lag model. A simple case takes the form⁷:

Y_{i t} = α_{0} + α_{1} Y_{i t - 1} + X_{i t} β_{0} + X_{i t - 1} β_{1} + ε_{i t} \cdot

This can be algebraically manipulated to take the form of an error correction model:

Δ Y_{i t} = γ_{0} + γ_{1} Y_{i t - 1} + Δ X_{i t} δ_{0} + X_{i t - 1} δ_{1} + ρ_{i t} \cdot

In this model, the short-run effects of contemporaneous levels in X _it on Y_it are measured by the corresponding elements of δ₀, whereas the short-run effects of lagged levels in X_it−1 are measured by δ₁ − δ₀. If the two series are in equilibrium, then the long-run multiplier for a unit change in X is δ₁ / −γ₁ (De Boef and Keele 2008, 190–91).⁸

Contrary to some claims in the literature, error correction models are not limited to cases of two cointegrated time series (de Boef and Keele 2008). They can be estimated using time series cross-sectional data, and there is no fixed minimum number of time periods required (Beck 2001). However, more time periods are better, and Beck (2001, 274) suggests that we should be “suspicious” of results based on fewer than 10 time periods.

I therefore estimated error correction models for 1989–96 only. I cannot extend the data back farther in time because the subclassifications for social services employment reported in County Business Patterns were different prior to 1988, and I need one years’ worth of lagged data. Models were estimated with panel-corrected standard errors and panel-specific first-order autocorrelated errors.

The results are shown in Tables 4 and 5. Table 4 shows that the contemporaneous effects of IFS and residential care employees on public sector spending (δ₀) are positive and statistically significant at the 95% level. The contemporaneous effect of state government liberalism is also positive and significant at the 95% confidence level. The long-run multipliers for IFS and residential care employees are both large and greater than their standard errors, but not statistically significant at even the 90% level.

Table 4.

Change in State and Local Government Spending per Capita on Nonmedical Vendors and Other Public Welfare: Error Correction Model, 1989–96

Variable	Coefficient (\|t\|)
Spending_t−1	−0.106* (1.94)
Δ IFS employees	15.42** (2.07)
IFS employees_t−1	3.34* (1.70)
Δ Residential care employees	−5.05 (1.03)
Residential care	3.38* (1.91)
Δ Day care employees	27.12** (2.75)
Day care employees_t−1	−0.41 (0.19)
Δ Per capita income	0.003 (1.43)
Per capita income_t−1	1.1e–4 (0.38)
Δ Pct. old or young	0.448 (0.49)
Pct. old or young_t−1	−0.398 (0.76)
Δ Pct. non-Hispanic	1.491 (1.17)
Pct. non-Hispanic white_t−1	−0.013 (0.16)
Δ State government liberalism	0.134** (2.13)
State government	0.005 (0.15)
Δ Federal non-Medicaid transfers	0.106 (1.38)
Federal non-Medicaid transfers_t−1	0.032 (0.46)
Fixed year effects	Included
Observations	392
R ²	.226
Long-run multipliers
IFS employees	31.52 (1.17)
Residential care	31.85 (1.04)
Day care employees	−3.86 (0.19)

Notes: IFS = individual and family services. Prais–Winsten regression with panel-corrected standard errors and panel-specific AR(1) autocorrelation. Figures in parentheses are absolute t-ratios.

p < .10. **p < .05 (two-tailed test).

Table 5.

Change in Private Social Services Employees per Capita: Error Correction Models, 1989–96

	IFS	Residential care	Day care
Social services	−0.034 (0.80)	−0.010 (0.34)	−0.011 (0.47)
Δ Other Public Welfare spending	0.002** (2.65)	−4.5e–4 (1.01)	7.7e–4** (3.20)
Other Public Welfare spending_t−1	−3.4e–4* (1.62)	1.4e–4 (0.74)	1.4e–4* (1.90)
Δ Transfers	−6.9e–4 (0.97)	3.5e–4 (0.58)	8.0e–5 (0.22)
Transfers_t−1	−5.0e–5 (0.21)	−1.4e–4 (1.17)	−2.5e–4** (2.64)
Δ Medicaid spending	7.6e–5 (0.48)	2.7e–4** (2.25)	−1.3e–4** (2.21)
Medicaid spending_t−1	3.3e–4** (2.79)	−1.2e–4 (1.46)	5.3e–5 (1.44)
Δ Per capita income	2.4e–7 (0.01)	−1.4e–4 (0.91)	1.8e–5 (1.62)
Per capita income_t−1	3.1e–6 (0.68)	−3.9e–7 (0.08)	4.21e–6 (1.21)
Δ Pct. ages 0–17	—	—	0.008 (0.46)
Pct. ages 0–17_t−1	—	—	−8.7e–4 (0.28)
Δ Pct. old or young	0.001 (0.10)	0.008 (1.44)	—
Pct. old or young_t−1	0.008 (1.48)	0.009** (2.44)	—
Δ Pct. college graduates	−0.006 (1.04)	0.012* (1.78)	0.005 (1.39)
Pct. college graduates	0.003 (0.99)	0.003 (1.01)	−9.8e–4 (0.69)
Δ Pct. non-Hispanic white	0.020 (1.05)	0.006 (0.38)	0.010 (1.03)
Pct. non-Hispanic white	0.001 (1.67)	9.5e–4 (0.85)	2.2e–4 (0.58)
Δ Pct. metropolitan population	0.003 (0.69)	0.007 (0.89)	0.003 (0.76)
Pct. metropolitan population	−0.001** (2.04)	−2.5e–4 (0.41)	−1.2e–4 (0.36)
Δ Citizen liberalism	−1.7e–4 (0.08)	−0.002 (0.86)	−0.002*** (2.15)
Citizen liberalism_t−1	0.002 (2.13)	0.002 (1.28)	5.8e–5 (0.14)
Fixed year effects	Yes	Yes	Yes
Observations	392	392	392
R ²	.233	.251	.583
Long-run multiplier	0.010 (0.75)	0.014 (0.25)	0.012 (0.54)

Notes: IFS = individual and family services. Prais–Winsten regression with panel-corrected standard errors and panel-specific AR(1) autocorrelation. Figures in parentheses are absolute t-ratios.

p < .10. **p < .05 (two-tailed test), ***p < .01.

Table 5 shows that the contemporaneous effect of public sector spending on IFS and day care employees is positive and significant at the 95% level. Once again, spending on transfers is not significant at even the 90% level in any equation. The coefficient on Medicaid spending is positive and significant for residential care employees but negative and significant for day care employees.

There are relatively few significant coefficients for the control variables. The coefficients on the change in citizen liberalism are negative in all three equations and significant at the 95% level for day care employees. The coefficient on the change in percentage college graduates is positive and significant at the 90% level for residential care employees. The long-run multiplier for Other Public Welfare spending is smaller in absolute value than its standard error in every equation.

Overall, the results tend to reinforce the results of the Arellano–Bond models, despite the assumption that all explanatory variables are exogenous. I attempted to address endogeneity by estimating two- and three-stage least squares models with four equations. For both models, the R² on the public spending equation is positive but extremely low,⁹ and none of the endogenous changes in spending or employees are statistically significant at the 90% level.

Summary and Predicted Effects

The large number of results concerning the relationship between public and private sectors presented in Tables 2 to 5 can be summarized fairly easily. For each dependent variable, there are three models: Arellano–Bond models for 1990–96 and 1999–2000, and an error correction model for 1989–96. In the models for the change in spending on Other Public Welfare, the coefficient on the change in IFS employees is positive and significant at the 95% level or better all 3 times. The coefficients on the change in residential care employees are not significant at the 90% level in any model, and the coefficient on the change in day care employees is positive and significant at the 95% level in one model.

In the models of the change in IFS employees, the coefficient on the change in Other Public Welfare spending is positive and significant at the 95% level all 3 times, none of the coefficients on the change in transfers are significant at the 90% level although all of them are negative, and one coefficient on the change in Medicaid is significant at the 95% level. In the residential care employees equations, none of the coefficients on Other Public Welfare spending or transfers are significant at the 90% level, and Medicaid is significant at the 95% level once. In the child day care equations, the coefficients on Other Public Welfare spending are positive and significant at the 95% level twice, Medicaid is negative and significant once at the 95% level, and transfer spending is not significant.

Overall, the results imply that Other Public Welfare spending and IFS employees are complements with causation running both ways; the same may hold for day care employees in 1989–96. There is some evidence that Other Public Welfare spending and residential care employees are complements in 1999–2000 and that transfer spending and IFS employees are substitutes, but these coefficients almost all fall short of statistical significance at conventional levels. These results are robust to a number of changes in specification. I tried including unemployment and poverty rates as measures of need. I also included a folded Ranney index and the population-weighted mean of Other Public Welfare spending in contiguous states in the public spending equation to capture the effects of interparty and interstate competition, respectively. None of these had significant coefficients.

We can also consider the magnitudes of the estimated effects. Table 6 shows the predicted cumulative effects after 4 years (the length of a gubernatorial administration in most states) of assumed 1-time changes in selected public spending or private employment variables. To isolate the effect of the assumed change, projections are based on a single equation from Tables 2 to 5 and do not include any feedback effects. The assumed changes are the mean absolute values of annual changes in the data, expressed as percentages of the mean levels. Similarly, the predicted effects are expressed as percentages of mean levels of the dependent variables. Elasticities are equal to the percentage change in the dependent variable divided by the percentage change in the explanatory variable.

Table 6.

Cumulative Predicted Effects after 4 Years

Time period	Model	Assumed change (%)	Predicted effect (%)	Elasticity
Effect of IFS employees on Other Public Welfare spending
1989–96	ECM	7.5	2.0**	.26
1990–96	A-B	7.7	3.0**	.38
1999–2000	A-B	5.5	4.5**	.81
Effect of Other Public Welfare spending on IFS employees
1989–96	ECM	8.5	0.5**	.06
1990–96	A-B	8.7	2.6**	.30
1999–2000	A-B	8.4	1.4**	.17
Effect of residential care employees on Other Public Welfare spending
1999–2000	A-B	5.3	1.9	.36
Effect of Other Public Welfare spending on residential care employees
1999–2000	A-B	8.4	0.6	.07
Effect of day care employees on Other Public Welfare spending
1989–96	ECM	5.7	1.3**	.23
1990–96	A-B	5.8	1.1	.19
Effect of Other Public Welfare spending on day care employees
1989–96	ECM	8.5	1.6**	.19
1990–96	A-B	8.7	1.3**	.15
Effect of transfer spending on IFS employees
1989–96	ECM	7.5	−0.3	−.04
1990–96	A-B	8.0	−1.6	−.18
1999–2000	A-B	13.8	−0.5	−.06

Notes: IFS = individual and family services. ECM = Error Correction Model. Assumed changes in explanatory variables are equal to the mean absolute values of changes in the data, shown as a percentage of the mean level. Predicted effects are shown as percentages of mean levels. Results are calculated using the coefficients in Tables 2 to 5.

p < .10. **p < .05 (contemporaneous effect; two-tailed test).

Note first that the elasticities for the effect of private sector employees on Other Public Welfare spending are always larger than the corresponding elasticity for the reverse effect. This pattern is consistent with crowd-out effects, where a change in public spending on grants and contracts is partially offset by a shift in other private organization revenues. Second, although the coefficients for the contemporaneous effects in Tables 3 and 5 are similar for the Arellano–Bond and error correction models, the cumulative effects after 4 years of public spending on IFS employees are quite different. This is because the lagged level of public spending in the error correction models has a negative coefficient. Third, the elasticities for the effect of changes in transfer spending on IFS employees are all quite small.

What about Religion?

Several previous studies have found associations between organized religious activity and private nonprofit organizations or social services (Corbin 1999; Lowry 2005; Wuthnow 1999). Because data on religious affiliation aggregated by state are only collected at 10-year intervals (Association of Religion Data Archives 2011), it was not possible to include them in dynamic models using first differences and lagged annual levels.

Preliminary analysis using data on religious affiliation for 1990 and 2000 casts doubt on previous findings. I first calculated bivariate correlations between social services employees per capita and the number of “adherents”¹⁰ per capita to mainline Protestant, Catholic, or evangelical Protestant and Mormon denominations, and private social services employees per capita in 1990 and 2000. IFS and residential care employees are positively correlated with mainline Protestants and Catholics and negatively correlated with other Protestants and Mormons in both years, and all but one of the correlations are statistically significant at the 95% level. This is consistent with previous research suggesting that members of evangelical denominations are less likely to participate in voluntary and nonprofit activities outside the church (Lowry 2005; Wuthnow 1999).

However, correlations between the average annual changes in religious adherents and actual changes in private social services employees, as well as between changes over the 10-year period, tell a different story. Although there were more residential care employees in states with more mainline Protestants in 1990 and 2000, the changes are negatively correlated. Similarly, although there were fewer IFS and residential care employees in states with many evangelicals and Mormons in 1990 and 2000, the changes during the interim are positively correlated. See the supplementary appendix for additional details.

Discussion

Whereas most prior political science research on state public welfare spending focuses on means-tested transfer programs or Medicaid, this article has focused on that part of public welfare spending that includes grants and contracts to private social services organizations. Dynamic models estimated with multiple techniques over two different time periods imply that Other Public Welfare spending and IFS employment are complements, and an increase in either one leads in the short run to an increase in the other. There is also some evidence implying a similar relationship between Other Public Welfare spending and residential care or day care employees in one of my two time periods. Conversely, private social services employment is largely independent of public spending on transfers, and although the coefficient on Medicaid spending is significant in some models, there is no consistent pattern.

One implication of these findings is that if public spending on transfers or Medicaid was to be reduced, we could not count on the private social services sector to compensate unless some spending was shifted over to programs that financed private sector providers. A reduction in Other Public Welfare spending, however, would lead to a less-than-proportionate decrease in IFS employees, at least for the short term.

The finding that an increase in private IFS employees leads to an increase in Other Public Welfare spending implies that organized interests have influence on state and local public welfare programs. This is a variable that has been omitted from previous studies that concentrate on transfer programs and Medicaid.

There is also evidence that citizens’ and political elites’ preferences, as measured by W. D. Berry et al.’s (1998) ideology indexes, matter. Controlling for the size of the private social services sector, states with more liberal political elites tend to spend more on Other Public Welfare programs; but controlling for state and local government public welfare spending, states with more liberal citizens tend to have fewer employees in residential care and child day care. This implies that the effect of ideology on private social services affects composition more than scale: IFS should constitute a larger share of private social services in more liberal states, and government funding should constitute a larger share of revenue for IFS providers in those states.

There are not a lot of significant results for control variables other than ideology. Non-Medicaid federal transfers have a positive, sometimes significant, effect on state and local Other Public Welfare spending, but the percentage of the population in vulnerable age groups does not. This is consistent with Peterson’s (1981) argument that redistributive spending by subnational governments should depend on resources but not need.

There are also some questions that remain unanswered. The consistent implication of my models that the magnitude of the effect of an increase in private social services employment on public spending is larger than the reverse is surprising, as the theoretical argument for causation running from public spending to private employment seems to be stronger. One possibility is that increased public funding partially crowds out other revenue sources (Brooks 2000; Lowry 1997), but a more definitive answer requires before and after data on revenues from all sources for specific private organizations that receive grants or contracts. There is also the question of whether influence by private social services organizations on public welfare spending is better interpreted as self-interested rent seeking or policy advocacy on behalf of their unorganized clients. Finally, the role of religion in the private social services sector remains in question. Data are lacking to do a dynamic analysis of changes in state aggregates, but the rough analysis presented here does not support earlier findings. Given the historical importance of religious organizations in supplying private social services, this is an issue that merits further investigation.

Footnotes

Acknowledgements

I thank Scott Allard, Jim Alt, Patrick Brandt, Curt Childress, Harold Clarke, Jeff Frieden, Sunshine Hillygus, Neil Malhotra, Matt Potoski, Kevin Quinn, Beth Simmons, and Margy Waller for helpful comments.

Author’s Note

Previous versions were presented at the 2010 State Politics and Policy Conference, Harvard University and the University of Texas at Dallas.

Declaration of Conflicting Interests

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

Funding

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

Notes

Bio

Robert C. Lowry is a professor of political science in the School of Economic, Political and Policy Sciences at the University of Texas at Dallas. His research interests include state fiscal policy and political and civic organizations.

References

Alesina

Alberto

Baqir

Reza

Easterly

William

. 1999. “Public Goods and Ethnic Divisions.” Quarterly Journal of Economics 114(November): 1243–84.

Arellano

Manuel

Bond

Stephen

. 1991. “Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations.” The Review of Economic Studies 58(April): 277–97.

Association of Religion Data Archives. 2011. U.S. Congregational Membership: State Reports. http://www.thearda.com/mapsReports/reports/selectState.asp (Accessed June 29, 2012).

Bailey

Michael A.

Rom

Mark Carl

. 2004. “A Wider Race? Interstate Competition across Health and Welfare Programs.” The Journal of Politics 66(May): 326–47.

Barrilleaux

Charles

Holbrook

Thomas

Langer

Laura

. 2002. “Electoral Competition, Legislative Balance, and American State Welfare Policy.” American Journal of Political Science 46(April): 415–27.

Beck

Nathaniel

. 2001. “Time-Series-Cross-Section Data: What Have We Learned in the Past Few Years?” Annual Review of Political Science 4:271–93.

Berry

Jeffrey M.

(with Arons

David F.

). 2003. A Voice for Nonprofits. Washington, DC: Brookings Institution Press.

Berry

William D.

Baybeck

Brady

. 2005. “Using Geographic Information Systems to Study Interstate Competition.” American Political Science Review 99(November): 505–19.

Berry

William D.

Ringquist

Evan J.

Fording

Richard C.

Hanson

Russell L.

1998. “Measuring Citizen and Government Ideology in the American States, 1960-93.” American Journal of Political Science 42(January): 327–48.

10.

Brooks

Arthur C.

2000. “Public Subsidies and Charitable Giving: Crowding out, Crowding in, or Both?” Journal of Policy Analysis and Management 19(Summer): 451–64.

11.

Brown

Robert D.

1995. “Party Cleavages and Welfare Effort in the American States.” American Political Science Review 89(March): 23–33.

12.

Corbin

John J.

1999. “A Study of Factors Influencing the Growth of Nonprofits in Social Services.” Nonprofit and Voluntary Sector Quarterly 28(September): 296–314.

13.

De Boef

Suzanna

Keele

Luke

. 2008. “Taking Time Seriously.” American Journal of Political Science 52(January): 184–200.

14.

Faricy

Christopher

. 2011. “The Politics of Social Policy in America: The Causes and Effects of Indirect vs. Direct Social Spending.” The Journal of Politics 73:74–83.

15.

Fellowes

Matthew C

Rowe

Gretchen

. 2004. “Politics and the New American Welfare States.” American Journal of Political Science 48(April): 362–73.

16.

Gamm

Gerald

Putnam

Robert D.

1999. “The Growth of Voluntary Associations in America, 1840-1940.” Journal of Interdisciplinary History 29(Spring): 511–57.

17.

Gray

Virginia

Lowery

David

. 1996. The Population Ecology of Interest Representation: Lobbying Communities in the American States. Ann Arbor: University of Michigan Press.

18.

Grønbjerg

Kirsten A.

2001. “The U.S. Nonprofit Human Service Sector: A Creeping Revolution.” Nonprofit and Voluntary Sector Quarterly 30(June): 276–97.

19.

Grønbjerg

Kirsten A

Paarlberg

Laurie

. 2001. “Community Variations in the Size and Scope of the Nonprofit Sector: Theory and Preliminary Findings.” Nonprofit and Voluntary Sector Quarterly 30(December): 684–706.

20.

Hacker

Jacob S.

2002. The Divided Welfare State: The Battle over Public and Private Social Benefits in the United States. New York: Cambridge University Press.

21.

James

Estelle

. 1987. “The Public/Private Division of Responsibility for Education: An International Comparison.” Economics of Education Review 6(1): 1–14.

22.

Kramer

Ralph M.

1987. “Voluntary Agencies and the Personal Social Services.” In The Nonprofit Sector: A Research Handbook, ed. Powell

Walter W.

New Haven, CT: Yale University Press, 240–57.

23.

Lowry

Robert C.

1997. “The Private Production of Public Goods: Organizational Maintenance, Managers’ Objectives, and Collective Goods.” American Political Science Review 91(June): 308–23.

24.

Lowry

Robert C.

2005. “Explaining the Variation in Organized Civil Society across States and Time.” The Journal of Politics 67(May): 574–94.

25.

Lowry

Robert C

Potoski

Matthew

. 2004. “Organized Interests and the Politics of Discretionary Federal Grants.” The Journal of Politics 66(May): 513–33.

26.

Luksetich

William

. 2008. “Government Funding and Nonprofit Organizations.” Nonprofit and Voluntary Sector Quarterly 37(September): 434–42.

27.

Mueller

Dennis C.

2003. Public Choice III. New York: Cambridge University Press.

28.

Percival

Garrick L.

Johnson

Martin

Nieman

Max

. 2009. “Representation and Local Policy: Relating County-Level Public Opinion to Policy Outputs.” Political Research Quarterly 62(March): 164–77.

29.

Peterson

Paul E.

1981. City Limits. Chicago: The University of Chicago Press.

30.

Primo

David M.

2007. Rules and Restraint: Government Spending and the Design of Institutions. Chicago: University of Chicago Press.

31.

Reid

Elizabeth J.

1999. “Nonprofit Advocacy and Political Participation.” In Nonprofits and Government: Collaboration and Conflict, eds. Boris

Elizabeth T.

Steuerle

C. Eugene

. Washington, DC: Urban Institute Press, 291–325.

32.

Salamon

Lester M.

1995. Partners in Public Service: Government-Nonprofit Relations in the Modern Welfare State. Baltimore, MD: The Johns Hopkins University Press.

33.

Smith

Steven Rathgeb

Lipsky

Michael

. 1993. Nonprofits for Hire: The Welfare State in the Age of Contracting. Cambridge, MA: Harvard University Press.

34.

Smith

Steven Rathgeb

. 2002. “Social Services.” In The State of Nonprofit America, ed. Salamon

Lester M.

Washington, DC: Brookings Institution Press, 149–86.

35.

U.S. Census Bureau. 2000a. Federal Aid to States for Fiscal Year 1999. Washington, DC: Government Printing Office.

36.

U.S. Census Bureau. 2000b. “Government Employment and Payroll.” http://www.census.gov/govs/apes/historical_data_2000.html (Accessed June 29, 2012).

37.

U.S. Census Bureau. 2001. Federal Aid to States for Fiscal Year 2000. Washington, DC: Government Printing Office.

38.

U.S. Census Bureau. 2010. “Government Finance and Employment Classification Manual.” http://www.census.gov/govs/www/class.html (Accessed June 29, 2012).

39.

U.S. Census Bureau. Various-a. Government Finances. Washington DC: Government Printing Office.

40.

U.S. Census Bureau. Various-b. Statistical Abstract of the United States. Washington DC: Government Printing Office.

41.

U.S. Census Bureau. Various-c. “County Business Patterns.” http://www.census.gov/econ/cbp/index.html (Accessed June 29, 2012).

42.

U.S. Congress. House of Representatives. Committee on Government Reform and Oversight. 1995. The Istook-McIntosh-Ehrlich Proposal. 104th Cong., 1st Sess., 28 September.

43.

Volden Craig. 2002. “The Politics of Competitive Federalism: A Race to the Bottom in Welfare Benefits?” American Journal of Political Science 46(April): 352–63.

44.

Walker

Jack L.

Jr . 1991. Mobilizing Interest Groups in America: Patrons, Professions, and Social Movements. Ann Arbor, MI: University of Michigan Press.

45.

Wawro

Gregory

. 2002. “Estimating Dynamic Panel Models in Political Science.” Political Analysis 10(Winter): 25–48.

46.

Wuthnow

Robert

. 1999. “Mobilizing Civic Engagement: The Changing Impact of Religious Involvement.” In Civic Engagement in American Democracy, eds. Skocpol

Theda

Fiorina

Morris P.

Washington, DC: The Brookings Institution Press, 331–63.