Article Critique. attached is article to use
BDR
poverty.pdf
The impact of government funding of poverty reduction programmes
Suhyun Jung1, Seong-Hoon Cho2, Roland K. Roberts2
1 Department of Applied Economics, University of Minnesota, 1994 Buford Ave, 337k Ruttan Hall Saint Paul, MN 55108, United States (e-mail: [email protected])
2 University of Tennessee – Agricultural and Resource Economics, Knoxville, Tennessee, United States (e-mail: [email protected], [email protected])
Received: 29 December 2011 / Accepted: 13 October 2013
Abstract. This research evaluates the impacts on poverty rates of government funds for edu- cation, health and hospitals, and public welfare allocated to poverty reduction for counties with persistently high poverty in the Southern United States. Our analysis found that increases in education funding in a poverty hot-spot county reduce the poverty rates of that county and its neighbouring hot-spot counties. We also found that higher health and hospital funding in a hot-spot county is associated with higher poverty rates in neighbouring hot-spot counties and that public welfare funding is not effective in mitigating poverty either within or outside of poverty hot-spots.
JEL classification: H75, I32, R58
Key words: Government funding, spatial panel, poverty, southern United States
1 Introduction
1.1 Background and objective
Since President Lyndon Johnson declared war on American poverty in 1964, policy-makers have struggled to develop programmes to reduce poverty and researchers have striven to determine the effectiveness of those programmes. Through their efforts, significant amounts of research and government funding have been directed towards the poverty issue. For example, the Appalachian Regional Commission (ARC) was formed by the federal government in 1964 to improve the standard of living in the Appalachian region. This programme included grants, direct loans, guaranteed loans, and direct payments for retirees, the unemployed and the poor (Reeder and Calhoun 2002). Despite government spending for poverty reduction, the poverty rate in the United States has remained consistently above 10 per cent over the last four decades. For instance, the poverty rate rose from 11.3 per cent in 2000 to 15.9 per cent in 2011, its highest level since 1993, partially due to the recession that began at the end of 2007 (DeNavas-Walt et al. 2007; Bishaw 2009, 2012). Poverty rates have been persistently higher for decades in the
doi:10.1111/pirs.12089
© 2014 The Author(s). Papers in Regional Science © 2014 RSAI
Papers in Regional Science, Volume 94 Number 3 August 2015.
Southern United States, particularly in the borderland of Texas and Lower Mississippi Delta (hereafter referred to as ‘poverty hot-spots’), compared to other parts of the United States (Poston et al. 2010). The recent increase in the US poverty rate and the persistently higher poverty rates in the Southern United States have revitalized interest in understanding whether government programmes have been effective in reducing the poverty rate.
In our research, we analyse the government’s role in poverty-rate reduction by evaluating the hypotheses that government funds budgeted for health and hospital, education, and public welfare reduce the poverty rate. We particularly investigate the effects of government funding on poverty rates in poverty hot-spots using three-decade panel data, which is unique in this type of study. Our hypotheses were conceptually motivated and tested using a spatial Durbin model for panel data (Elhorst 2003, 2010). The model was estimated using county-level data from 16 states in the US Census Bureau’s South Division (referred to as ‘the South’). The data were for 1990, 2000, and 2010. Ex ante impact analysis was performed for persistent poverty areas in the South. The model was used to estimate direct, indirect, and total marginal poverty-reducing effects of government funds. We predicted poverty rates and marginal poverty-reducing effects of significant government funding categories in the South and in three poverty hot-spots (spatial clusters of counties with persistent poverty). Our study pro- vides important information to help policy-makers in developing regional poverty reduction strategies.
1.2 Review of empirical literature on poverty
Many researchers have investigated the effects of specific government funding categories on poverty and economic status. A number of researches have focused on government funds to
Fig. 1. Poverty hot-spots (high-poverty counties surrounded by high-poverty counties) in 2010 based on local indicators of spatial association (LISA) using the poverty rate
S. Jung et al.654
Papers in Regional Science, Volume 94 Number 3 August 2015.
improve health and education because health improvements and increased education were found to be highly correlated with economic growth (e.g., Triest 1997; Bloom and Canning 2000; Waidmann and Rajan 2000; Bhargava et al. 2001; Beale Fan et al. 2002; Jung and Thorbecke 2003; Probst et al. 2004). A large body of evidence shows that poverty is correlated with alcohol and drug abuse and mental health issues, and thus the funding of programmes that support people with these problems should impact poverty (Baingana et al. 2006). In other branches of the literature, public welfare programmes targeting low-income families with children, their parents, and caregivers (e.g., the Temporary Assistance to Needy Families (TANF) programme) were found to affect poverty (Lower-Basch 2011). This finding suggests that funding pro- grammes through the public welfare budget should address poverty issues.
Numerous studies have focused on the quantitative impact of government spending on poverty reduction. Fan et al. (2002) examined the effects on China’s rural poverty rate of government expenditures on rural education and infrastructure and found positive poverty- reducing impacts. Jung and Thorbecke (2003) explored the impacts of increased education expenditures, and the resulting excess supply of educated and skilled labour, on poverty alle- viation in Tanzania and Zambia. Afonso and Aubyn (2004) evaluated the efficiency of govern- ment spending on education and health among Organization for Economic Cooperation and Development (OECD) countries and suggested possible causes (i.e., different resource prices, public sector inefficiency) for varying government expenditure outcomes in terms of poverty indicators such as literacy, life expectancy, and infant mortality. Glennerster (2002) reviewed poverty measures in the United States and emphasized the need for a variety of poverty measures. He identified health expenditures as a crucial element in explaining the basic neces- sities of the poor. Smeeding (2006) compared government spending and poverty trends in 11 developed countries and emphasized the importance of creating incentives for low-wage workers when increasing welfare benefits targeted at low-income families.
A number of studies have evaluated regional poverty reduction strategies. Triest (1997) examined how increased educational opportunity and increased employment of low-income populations narrow the interregional poverty gap. Swaminathan and Findeis (2004) found that welfare assistance to poor workers had a positive effect on reducing poverty in metropolitan areas. Rupasingha and Goetz (2007) suggested that increases in government investment in social capital can reduce the poverty rate by easing transaction costs paid by local associations. Allard et al. (2003) and Blank (2005) suggested that poverty reduction is more effective when spatially- targeted governmental policies are implemented. Levernier et al. (2000) found that developing educational programmes specifically targeted at minorities and residents in non-metropolitan statistical areas (MSA) is a key element for reducing poverty.
Notwithstanding the importance of regional targeting of poverty reduction policies found in the literature, poverty reduction has rarely been explored using a spatially-explicit framework. Partridge and Rickman (2005) discussed spatial dependencies in poverty rates and adjusted for spatial autocorrelation by including weighted averages of neighbouring-county characteristics. Partridge and Rickman (2006) explored the geographic disparities in poverty across the United States and drew implications for integrated national poverty reduction strategies that combine place-based and person-based policies. They considered interregional equilibrium and disequi- librium perspectives in which firms are attracted to low-wage areas and labour departs until poverty equilibrium is reached. Under the equilibrium perspective, local economic development policies are unlikely to increase the utilities of the original residents because new migration offsets any wage gains arising from increased labour demand. Alternatively, under the disequi- librium perspective, local economic growth can reduce local poverty rates because barriers to mobility (e.g., housing market constraints, transportation costs, migration costs, and imperfect information) can contribute to deviations from equilibrium poverty rates that can persist over time.
Government funding and poverty reduction in the Southern US 655
Papers in Regional Science, Volume 94 Number 3 August 2015.
Empirical evidence on whether poverty rates tend to follow the interregional equilibrium perspective (e.g., Blomquist et al. 1988; Beeson and Eberts 1989) or disequilibrium perspective (e.g., Kaldor 1970; Krugman 1991; Glaeser et al. 1992) is mixed. Nevertheless, spatial patterns of poverty commonly suggest that poverty rates are persistently unequal across regions (Friedman and Lichter 1998; DeNavas-Walt et al. 2007; Weber et al. 2005). For example, the ‘Southern Black Belt,’ extending from Southwest Tennessee to East-central Mississippi and then East through Alabama to the border with Georgia, has had persistently higher poverty rates than other regions within the South (Wimberley and Morris 1997, 2002).
1.3 Significance of this analysis
The aforementioned empirical literature provides insight into four distinct aspects of poverty: (i) the impact of government funding on poverty; (ii) the regional targeting of poverty reduction strategies; (iii) the spatial nature of poverty; and (iv) the persistent nature of poverty. Neverthe- less, research addressing all four aspects of poverty in one framework has not been undertaken mainly because an econometric framework accounting for both the spatial and persistent natures of poverty was not available until recently. By adjusting for spatial and temporal autocorrelations, our comprehensive econometric framework is used to evaluate the impacts of government poverty-related funding categories on poverty-rate reductions in several clusters of high-poverty counties in the South. This aspect of our study is an important advance in poverty-related research addressing the efficacious government poverty-reduction strategies that target regionally-persistent poverty.
We also emphasize that our study examines the impact of government funding on poverty over 20 years (i.e., 1990, 2000, and 2010) and few, if any, previous studies explicitly considered the spatial aspects of poverty in the context of its long-term persistence. Combining spatial and long-term analyses is particularly important because persistent poverty takes considerable time to address, and shorter-term temporal and spatial analyses performed separately or in combi- nation may not fully identify government’s effectiveness and role in poverty alleviation.
Another of our innovations is to evaluate the effectiveness of government funds allocated to poverty reduction in clusters of high-poverty counties, defined as poverty hot-spots. Identifying poverty hot-spots using a spatial statistical tool (see Section 2.7) allows us to be more focused and systematic. Given that persistent poverty is a serious problem and a daunting challenge in some areas, focusing resources on poverty hot-spots is being actively pushed (Duncan 1992; Lyson and Falk 1993; Wimberley and Morris 2002). Under increasingly tight budgets, states could take a more targeted approach by concentrating funding for poverty reduction on local ‘hot-spots’. Thus, our focus on poverty hot-spots is even more meaningful and sought after than ever.1
2 Methods and procedures
2.1 Conceptual motivation for the empirical specification
Conceptual hypotheses about the relationships between government funding for health care, education, or public welfare programmes and the poverty rate provide motivation for an empirical model to test those hypotheses. First, government funding for health-care programmes
1 We appreciate an anonymous referee for pointing out the following information. The targeting of impoverished hot-spots has been promoted by policy-makers in the health care debate over the Affordable Care Act (Blumberg 2012; HHS 2013; Manchikanti et al. 2011).
S. Jung et al.656
Papers in Regional Science, Volume 94 Number 3 August 2015.
is a crucial element in explaining the basic necessities of the poor (Glennerster 2002; Meyer and Sullivan 2012). Poor health has contributed to US poverty because medical expenses can exhaust family resources. For example, about half of bankruptcies in the United States in 2001 involved medical debt (Himmelstein et al. 2005). The strong correlation between health-related expenses and wealth loss is well documented (Smith 1999; Cook et al. 2010). Consequently, we hypothesize that an increase in government funding of health-care programmes decreases the poverty rate.
Second, government funding for education is correlated with earnings. In 2006, 23 per cent of the US population without a high school diploma had incomes below the poverty level while fewer than 4 per cent of the population with a college degree had incomes below the poverty level (Rynell 2008). Given the statistical evidence, many believe that the persistent cycle of poverty can be broken through education (Bhola 2006; Perry 2006; Rodgers and Rodgers 1993). Based on the premise that education can alleviate poverty, the US government has initiated programmes aimed at improving education with a major purpose of alleviating poverty. For example, the Individuals with Disabilities Education Act (IDEA) and the Child Nutrition Act (CNA), funded through the education budget, are weighted heavily towards the poor in their decision-making formulas (Fujiura and Yamaki 2000; Cook and Frank 2008). Thus, our hypothesis is that increased government funding of education programmes decreases the poverty rate.
The hypothesis about the relationship between government funding of public welfare pro- grammes and the poverty rate is based on two conflicting views. Proponents of public welfare spending to reduce poverty argue that reduced income inequality through welfare spending can increase family expenditures on education and increase incentives to work (Kenworthy 1999). Critics argue that such programmes fail to reduce poverty because: (i) too little money reaches the poor (Stigler 1970; Friedman and Friedman 1979; Crook 1997); (ii) the programmes undermine the intrinsic motivation of the poor (Murray 1984; Butler and Kondratas 1987; Lee 1987); and (iii) the programmes reduce incentives to invest and to work (Arrow 1979; Lindbeck et al. 1994; Okun 1975). Our testable hypothesis is the proponents’ view–welfare spending reduces poverty.
2.2 Empirical model specification
For panel data, the poverty rate equation is:
P Xit it i t it= + + + +α β μ λ ε , (1)
where i represents the ith county (i = 1, 2, . . . , 1,420); t denotes 1990, 2000, and 2010; P is the poverty rate;α is a constant parameter; X is a vector of explanatory variables including demo- graphic, employment and environmental characteristics, per capita government funding for education (E), health and hospitals (H), and public welfare (F), a dummy variable for whether or not the ith county is within a poverty hot-spot (S), and the interaction of the dummy variable with per capita government funding (i.e., E × S, H × S, F × S); β is a parameter vector; and ε is an error term. The terms μ and λ respectively denote unobserved spatial and time specific effects. The interaction variables capture differences in the impacts of per capita government funding by the counties that are or are not within poverty hot-spots.
Equation (1) allows the poverty rate to be both spatially and time persistent, generating spatial and temporal dependencies among the observations. We used a spatial Durbin model for panel data (Elhorst 2003, 2010) to estimate equation (1) because the model controls for both spatial and temporal dependencies. We employed a ‘specific-to-general’ approach to compare the aspatial poverty-rate equation with the spatial poverty-rate equation, and a
Government funding and poverty reduction in the Southern US 657
Papers in Regional Science, Volume 94 Number 3 August 2015.
‘general-to-specific’ approach to compare the generalized spatial model with the spatial error and spatial lag models (Elhorst 2010). We also compared fixed and random effects models.
2.3 Endogeneity issues of explanatory variables
Because per capita government funding for education, health and hospitals, and public welfare are largely determined by economic conditions, they are closely associated with the poverty rate, potentially introducing endogeneity into the poverty-rate equation (e.g., Fan and Chan-Kang 2008). In our estimates, endogeneity is not a concern because we use exogenous predetermined budget allocations for 1987, 1997 and 2007 as explanatory variables to explain the poverty rates in 1990, 2000 and 2010, respectively (Wooldridge 2009, p. 562). Other variables, such as employment variables, that are contemporanenous to the poverty-rate variable may also be endogenous. We performed a robustness analysis by estimating a reduced-form model that included only the most exogenous variables to see if the results changed significantly.2
2.4 Tests for model specifications
We conducted a spatial Lagrange multiplier (LM) test (Burridge 1980; Anselin 1988) and a robust LM test (Anselin et al. 1996) to compare the aspatial and spatial models in the context of the specific-to-general approach. The robust and non-robust LM statistics (248 and 11 respec- tively) for the spatial lag model and the corresponding LM statistics (597 and 360 respectively) for spatial error model indicated that the aspatial model was rejected at the 5 per cent level (hereafter referred to as ‘significant’) in favour of the spatial lag model or the spatial error model. The spatial model estimates were based on 1,420 × 3 observations assuming a hybrid of the first-order Queen continguity weight matrix and the inverse distance weight matrix (here- after ‘the hybrid weight matrix’).
In the context of the general-to-specific approach, we performed Wald and likelihood ratio (LR) tests using the framework of a spatial Durbin model for panel data (SDMP) that employs both spatial lag and spatial error components (LeSage and Pace 2009) with incorporation of temporally lagged dependent variable.
P P w P w P X w Xit it ij jt j
N
ij jt j
N
it ij jt j
N
= + + + + +− − = = =
∑ ∑ ∑α γ δ ρ β φ1 1 1 1 1
++ + +μ λ εi t it, (2)
where j represents the jth county, γ is a parameter of poverty rates for the lagged time period
(Pit‒1), wij is element (i, j) of the N × N spatial weight matrix W, w Pij jt j
N
= ∑
1
is poverty rates within
the neighbours defined by the hybrid weight matrix, δ is a parameter for spatially lagged poverty rate for the lagged time period (WPjt‒1), ρ is a parameter for spatially lagged poverty rates for the present time period (WPjt), ϕ is a parameter vector of spatially lagged independent variables for the present time period (WXjt), and μi and λt represent the spatial-specific time-invariant effect and time-specific spatial-invariant effect, respectively. The Wald and LR test statistics compar- ing the spatial Durbin model against the spatial lag model were 515 and 468, respectively, and the Wald and LR statistics comparing the spatial Durbin model against the spatial error model
2 Alternatively, endogeneity tests could be done; however, because of the burden of choosing valid instrumental variables, we chose the robustness test. We particularly thank an anonymous referee for bringing this issue to the authors’ attention.
S. Jung et al.658
Papers in Regional Science, Volume 94 Number 3 August 2015.
were 215 and 225, respectively. These statistics rejected the hypothesis that the SDMP can be simplified to the spatial lag model or spatial error model. The same tests were performed with consistent results using hybrid weight matrices that included second- and third-order Queen contiguity weight matrices. Based on these test results, the poverty-rate equation was estimated using the SDMP expressed in Equation (2).
2.5 Panel data model specification
The panel data model can be specified in two ways. One way deals with whether to include time- and/or spatial-specific effects, and the other is whether to treat unobserved effects as random or fixed effects. First, we perform an LR test to investigate whether to include a time period fixed effect and a spatial fixed effect. An LR statistic of 134 using estimates from the SDMP rejected the hypothesis that the temporal effects are jointly insignificant. An LR statistic of 3,219 using estimates from the SDMP without the time-invariant variables (i.e., urban influence codes and natural amenity scales) rejected the hypothesis that the spatial effects are jointly insignificant. As a result, we included a time-specific effect and a spatial-specific effect in the model.
Second, we tested the hypothesis that the unobserved effects can be treated as random effects using Hausman’s specification test based on the SDMP without the time-invariant variables (Hausman 1978; Lee and Yu 2010a). A Hausman statistic of 136 rejected the hypothesis, suggesting the estimation of the SDMP with fixed effects. Consequently, the poverty-rate equation was estimated using a fixed-effect SDMP model. Time-invariant variables were excluded from both tests because the coefficients of the time-invariant variables cannot be estimated for spatial fixed-effect models using panel data (Greene 2010).
2.6 Estimation and marginal effects
The SDMP poverty-rate equation specified with time period and spatial fixed effects using the hybrid weight matrix was estimated by maximum likelihood following Elhorst (2003, 2010). Once the parameters were estimated, the total marginal effect of a change in the kth explanatory variable xk on the South’s average poverty rate in county i = 1 up to N at a given time t was estimated by:
∂ ∂
∂ ∂
⎡ ⎣⎢
⎤ ⎦⎥
= −( ) +[ ]− P
x
P
x W I W
k Nk t
k N k 1
1� I ρ β φ . (3)
Alternatively, the marginal effect that incorporates γ and δ (i.e., 1 1−( ) − +( )[ ]−γ ρ δI W +[ ]β φI Wk N k ) (Debarsy et al. 2012; Elhorst 2012) could be used. By ignoring γ and δ, our
marginal effects estimated by Equation (3) are perceived as a short-term effect because they do not account for the temporal dynamics of the poverty rate and spatially lagged poverty rate. Despite the potential for obtaining long-term effects, we chose to use the short-term effects because of three observations made by Elhorst (2012) about the dynamic spatial Durbin model’s long-term effects (i.e., identification problem, lack of empirical evidence, and overfitting due to model complexity).
Because spatial spillover effects play a significant role in the total marginal effect, the total marginal effect was decomposed into a direct marginal effect (hereafter, referred to as direct effect) and an indirect marginal effect (hereafter, referred to as indirect effect) (LeSage and Pace 2009). The direct effect is the effect on county i’s poverty rate and the indirect effect is the effect on the poverty rate outside county i. We followed LeSage and Pace (2009) to calculate the direct and indrect effects (The details are laid out in Appendix 2).
Government funding and poverty reduction in the Southern US 659
Papers in Regional Science, Volume 94 Number 3 August 2015.
We identified poverty ‘hot-spots’ using local indicators of spatial association (LISA) analy- sis (see the following subsection for the details), and summarized the marginal effects of significant government funding categories across the poverty hot-spots. These summaries quan- tify the relative importance of the funding categories in alleviating poverty, and they also provide a means to examine how these effects changed over time. We used the estimated parameters of the SDMP model expressed in Equation (2) to predict poverty rates for the entire South and across the three poverty hot-spots for 1990, 2000, and 2010.
2.7 Identifying clusters of high-poverty counties
We used 2010 average county poverty rates to identify clusters of high-poverty counties. LISA values (Anselin 1995) were estimated to identify spatial clusters of poverty in the South. The LISA values indicate the extent of spatial autocorrelation between the poverty rate in a particular county and the poverty rates of the counties surrounding it. Poverty hot-spots were identified through plotting sets of contiguous locations for which LISA values were significant (Anselin 1995). These clusters can include a single county and its con- tiguous neighbours or a larger set of contiguous counties for which the LISA values are significant. The county LISA values for poverty rates for 2010 were defined as: LISA y y y w y yi i i
n i j
n ij j= −( ) ∑[ ]⋅ ∑ −( )= =1 2 1 , where n is the sample size, yi is the poverty rate in
county i with sample mean , and wij is an n × n contiguity weight matrix with diagonal elements of 0 and off-diagonal elements of 1 for all counties j that are contiguous to county i. The LISA clustering was done for poverty rates in 1990 and 2000 as well, but the poverty hot-spots did not change appreciably because of the persistent temporal and spatial clustering of poverty rates. See Figure 1 for identified three poverty hot-spots.
3 Study area and data description
3.1 Study area
This study focuses on 1,420 counties in 16 states (i.e., Arkansas, Alabama, Delaware, Florida, Georgia, Kentucky, Louisiana, Maryland, Mississippi, North Carolina, Oklahoma, South Caro- lina, Tennessee, Texas, Virginia, and West Virginia) in the South. The South was selected because its poverty rates have been persistently higher than other regions, making poverty an important issue in the South compared with other US regions. In 2010, the South had the highest poverty rate at 16.9 per cent, compared with 12.8 per cent in the Northeast, 13.9 per cent in the Midwest, and 15.3 per cent in the West (US Census Bureau 2011a). In addition, the South was the only region with a significant increase in its poverty rate (about 10 per cent increase from 15.7 per cent to 16.9 per cent) between 2009 and 2010 (US Census Bureau 2011a).
3.2 Data
The study employs five county-level geographical information system (GIS) datasets: (i) poverty data for 1989, 1999, and 2009 (i.e., the percentage of individuals with incomes below the US Census Bureau’s 1991, 2001, 2011 poverty threshold (US Census Bureau 2011d) based on family size and the ages of its members, adjusted for inflation using the consumer price index) to represent poverty rates in 1990, 2000, and 2010, respectively; (ii) demographic data (i.e., population percentages of Whites, Asian-Pacific Islanders, Blacks and other races, with the
S. Jung et al.660
Papers in Regional Science, Volume 94 Number 3 August 2015.
population percentage of Blacks as the reference category; population percentages below 18 years of age, between 18 and 24 years of age, 65 years of age or older, with the population percentage between 25 and 64 years of age as the reference category; female-headed house- holds; difficulty speaking English; at least some college education; and living in households with three or more workers) for 1990, 2000, and 2010 (US Census Bureau 2001a, 2011b); (iii) employment data (i.e., employment variables that include population percentages of unem- ployed workers 16 years of age or older; employment in agriculture, forestry and fisheries; employment in manufacture, mining and construction; employment in transportation, commu- nication and other public utilities; employment in wholesale and retail activities; and employ- ment in finance, insurance and real estate) for 1990, 2000, and 2010 (US Census Bureau 2001b, 2011c; US Department of Labour 2011); (iv) employment in the arts and environmental data (i.e., natural amenity scales and urban influence codes) for 1993 and 2003 (ERS USDA 2004, 2007); and (v) funds from federal and state governments allocated by state governments to county budget categories for education, health and hospitals, and public welfare – budget codes of C21, C42 and C79 for 1987, 1997 and 2007, respectively (US Census Bureau 2008).
County budget categories for revenues allocated by state governments were used because they include all government funds flowing into the budgets of municipalities, townships, special districts, and independent school districts (US Census Bureau 2008). Because the US Census Bureau does not publish spending from these budget categories, in using these data, we assume these funds are spent as intended. As can be seen from their descriptions below, these county funding categories are heavily weighted toward programmes directly affecting the poor.
The county budget category for education (hereafter, ‘education funding’) includes funds received by the counties from the state to support local schools and state redistribution of federal aid for education; handicapped, special and vocational education and rehabilitation; student transportation; equalization aid; school health; local community colleges; adult education; school buildings; and property tax relief related strictly to school funding. The IDEA and CNA programmes are examples of programmes funded through county education budgets (Salisbury 2004; Trohanis 2008;). These budget categories for education can help increase the income levels of individuals, which may affect the poverty rate of a county. For example, federal aid for handicapped, special and vocational education and rehabilitation can provide incentives for disabled and low income people to work and earn more income to lift themselves out of poverty. A significant marginal effect implies that an increase in education funding reduces the poverty rate.
The county budget category for health and hospitals (hereafter, ‘health and hospital funding’) includes funds received from the state to support county programmes for local health; maternal and child health; alcohol, drug abuse and mental health; environmental health; superfunds; nursing aid; hospital financing; and hospitalization of patients in local government hospitals. Health and hospital funding does not include Medicaid spending, which is included in the budget category for public welfare. Health and hospital funding can promote a better environment for the poor or help increase the ability of individuals to earn income. For instance, funding for alcohol, drug abuse and mental health can help drug or alcohol addicts and people with mental disease to overcome addictive behaviour, increasing their ability to earn income and lift themselves out of poverty.
The county budget category for public welfare (hereafter, ‘public welfare funding’) includes funds received from the state for public welfare purposes; medical care including Medicaid and related administration under public assistance programmes; care in nursing homes not associ- ated with hospitals; federal categorical assistance; non-categorical assistance and administration of local welfare programmes. Public welfare funding is more directly related to increasing the income of the poor than the other two funding categories. The TANF programme is an example of a programme funded through the public welfare budget.
Government funding and poverty reduction in the Southern US 661
Papers in Regional Science, Volume 94 Number 3 August 2015.
Urban influence codes for 1993 were used as proxies for rurality in 1990 and urban influence codes for 2003 were used as proxies for rurality in 2000 and 2010. Natural amenity scales for 1993 were used as proxies for the physical characteristics of counties that enhance their appeal as places to live (ERS USDA 2004). Variable names, definitions, and descriptive statistics for the variables used in the analysis are presented in Table 1.
Table 1. Variable names, descriptions, and statistics
Variable Description 1990 mean (SD)
2000 mean (SD)
2010 mean (SD)
Poverty variable
Individual poverty rate Percentage of individuals with incomes below the US Census Bureau poverty threshold based on the family size and the age of its member adjusted for inflation using the consumer price index in 1989, 1999, and 2009 (%)
20.04 16.94 18.26 (8.47) (6.74) (6.61)
Spatial lag of individual poverty rate
Spatial lag of individual poverty rate in 1989, 1999, and 2009 (%)
20.01 16.89 18.13
(6.92) (5.33) (5.02)
Demographic variable
White White population alone divided by total population (%)
79.89 77.52 76.11 (16.95) (17.26) (17.65)
Asia-Pacific Asia–Pacific population divided by total population (%)
0.44 0.65 0.96 (0.74) (1.05) (1.52)
Other races Native Americans and other races excluding white, Asian-Pacific, and Black population divided by total population (%)
3.05 5.08 6.24 (6.11) (6.43) (5.98)
Age 0–17 years Population aged below 18 years divided by total population (%)
26.61 25.27 23.53 (3.50) (3.15) (3.10)
Age 18–24 years Population aged between 18 and 24 divided by total population (%)
9.85 9.25 9.10 (3.53) (3.55) (3.59)
Age over 65 years Population aged 65 or older divided by total population (%)
14.33 14.14 15.27 (4.18) (3.83) (3.89)
Female-headed Population living in female-headed households divided by total households (%)
15.17 17.04 13.60 (5.67) (6.22) (4.22)
English speaking Population aged between 16 and 64 with difficulty speaking English divided by total population (%)
1.30 1.97 3.70 (2.97) (3.10) (5.22)
Some college education Population with at least some college education divided by total population aged 25 or older (%)
31.15 35.73 42.94 (10.49) (10.00) (10.71)
Three or more workers Population living in households with 3 or more workers divided by total population (%)
10.80 9.27 4.54 (2.80) (2.17) (1.74)
Employment variable
Unemployment rate Unemployed workers aged 16 years or older divided by total population aged 16 years or older (%)
6.62 4.58 9.82 (3.01) (1.66) (2.85)
Agriculture Population of employment in agriculture, forestry, and fisheries divided by total employed population (%)
6.24 4.54 5.78 (5.94) (4.85) (6.30)
Manufacturing Population of employment in manufacture, mining, and construction divided by total employed population (%)
30.98 27.06 21.54 (10.31) (8.61) (7.20)
Public utility Population of employment in transportation, communications, and other public utility divided by total employed population (%)
6.66 7.16 7.12 (2.06) (3.95) (2.25)
Wholesale and retail trade Population of employment in wholesale and retail trade divided by total employed population (%)
19.17 13.01 14.17 (3.47) (4.15) (2.70)
S. Jung et al.662
Papers in Regional Science, Volume 94 Number 3 August 2015.
Table 1. Continued
Variable Description 1990 mean (SD)
2000 mean (SD)
2010 mean (SD)
Finance and insurance Population of employment in finance, insurance, and real estate divided by total employed population (%)
4.18 4.44 4.67 (1.67) (1.74) (1.90)
Arts Population of employment in art, design, entertainment, performance, sports, and related workers divided by total employment (arts occupation in 1990 for 1990 and 2000 for 2000 and 2010) (%) (ERS USDA 2007)
0.60 0.62 0.62 (0.37) (0.38) (0.38)
Environmental variable
Urban influence code Measure of rurality ranges between 1 and 12, 1 being large metro area of 1+ million residents and 12 being noncore not adjacent to metro or micro area and does not contain a town of at least 2,500 residents (urban influence code in 1993 for 1990 and 2003 for 2000 and 2010) (ERS USDA 2007)
5.33 4.89 4.89 (2.65) (3.21) (3.21)
Natural amenity scale Measure of physical characteristics of a county area that enhance the location as a place to live, which is constructed by combining six measures of climate, typography, and water area that reflect environmental qualities most people prefer: warm winter, winter sun, temperate summer, low summer humidity, topographic variation, and water area (Natural amenity scale measured over the years between 1970 and 1996 used for 1990, 2000, and 2010) (ERS USDA 2004)
0.37 0.37 0.37 (1.37) (1.37) (1.37)
Governmental funding variable
Health and hospitals State aid for local health programmes; maternal and child health; alcohol, drug abuse, and mental health; environmental health; superfunds; nursing aid; hospital financing (including construction); and hospitalization of patients in local government hospitals in 1987, 1997, and 2007, respectively, for 1990, 2000, and 2010 ($/capita)
6.67 15.65 20.13 (13.67) (41.96) (57.03)
Education State aid for support of local schools; redistribution of Federal aid for education; handicapped, special, and vocational education and rehabilitation; student transportation; equalization aid; school health; local community colleges; adult education; school buildings; and property tax relief related strictly to school funding. in 1987, 1997, and 2007, respectively, for 1990, 2000, and 2010 ($/capita)
395.25 684.91 1,046.33 (131.66) (217.37) (349.44)
Public welfare State aid for public welfare purposes; medical care and related administration under public assistance programmes (including Medicaid) even if received by a public hospital; care in nursing homes not associated with hospitals; Federal categorical assistance; non-categorical assistance (e.g., home relief, emergency assistance); and administration of local welfare programmes in 1987, 1997, and 2007, respectively, for 1990, 2000, and 2010 ($/capita)
5.32 12.67 16.71 (13.14) (30.52) (52.87)
Notes: The data are at the county level for 1990, 2000, and 2010 unless indicated otherwise.
Government funding and poverty reduction in the Southern US 663
Papers in Regional Science, Volume 94 Number 3 August 2015.
4 Empirical results
4.1 Regression results
The results of the SDMP with spatial- and time-fixed effects using the hybrid weight matrix are presented in Table 2.3 The positive effect of the spatially lagged poverty rate indicates that a 1 per cent increase in the poverty rate in neighbouring counties increases the own poverty rate by 0.27 per cent. This finding reaffirms the spatial clustering of poverty in the South. Conversly, the coefficient for the time lagged poverty rate is not significant while the coefficient for the spatially lagged time lagged poverty rate is positive and significant, which means that a 1 per cent increase in the time lagged poverty rate in surrounding counties increased the own poverty rate by 0.06 per cent. The insignificance of the coefficent for the time lagged poverty rate variable is unexpected given the persistent nature of the poverty. This result may be explained by the persistent spatial clustering of poverty, captured by the spatially lagged poverty rate, overshadowing the time-lagged effect of poverty. Because the spatial spillover effects make interpretation of parameters difficult, we focus below on interpreting the direct, indirect, and total marginl effects of the variables.
The direct effects of all demographic variables except other races variable were significant and the signs were as expected. The direct effects indicate that counties with less White and Asian-Pacific population, higher percentages of youth, seniors, people living in female-headed households, people with difficulty speaking English, people without any college education, and people living in households with less than three workers had higher poverty rates. These results suggest that, in addition to having an economically active population, a county’s human-capital capacity was important in explaining lower poverty rates. The total effects differ from the direct effects for some of the variables. For example, a 1 percentage point increase in the Asian-Pacific population in a county decreases the poverty rate of the own county by 0.27 per cent, while it decreases the poverty rate in the entire South by 0.56 per cent, when summing the direct and spillover (indirect) effects. As another example, a 1 percentage point increase in people living in female-headed households increases the poverty rate of the own county by 0.06 per cent, but has no effect on the poverty rate in the entire South, after accounting for the spatial spillover. These differences in direct and total effects reaffirm the importance of understanding and accounting for spatial spillover effects.
The direct effects were significant for four of seven employment variables. Increases in the percentages of employments in the manufacturing sector and the finance and insurance sector of a county decrease the poverty rate in the own county. In contrast, increases in the percentages of the population unemployed and employment in agriculture increase the poverty rate in the own county. Again, these direct effects are quite different from the total effects that take into account of spatial spillover, reflecting the broader impacts on the South as a whole. The total effects indicate that 1 percentage point increases in employment in the agriculture sector, the manufactor sector, and art sector decrease the poverty rate in the entire South by 0.19 per cent, 0.20 per cent, and 2.11 per cent, respectively. The higher total effect of employment in the art sector reaffirms the positive effect of growth in creative employment that is highly associated with poverty mitigation and rural economic development (Cho et al. 2007). On the other hand, employment in the finance and insurance sector did not have a significant total effect on the poverty rate. The differences between the direct and total effects of the employment rates in different sectors indicate differences between the local and global effects of the employment structure on poverty.
3 The side-by-side presentation of the parameters for the full and reduced-form models is in Appendix 1 (robustness analysis). The comparison between the two models shows no significant change in terms of sign and significance of the non-spatially lagged parameters, suggesting no substantial endogeniety bias caused by the employment variables.
S. Jung et al.664
Papers in Regional Science, Volume 94 Number 3 August 2015.
Table 2. Parameter estimates of the spatial Durbin panel model with spatial fixed and time fixed effects
Variables Parameter of non-lagged
variable
Parameter of spatially-lagged
variable
Direct effect
Indirect effect
Total effect
Poverty variable
Individual poverty rate 0.2684* (0.0202)
Time lag of individual poverty rate –0.0212 0.0631* –0.0178 0.0737* 0.0559 (0.0183) (0.0282) (0.0181) (0.0349) (0.0362)
Demographic variable
White –0.1231* 0.0306 –0.1230* –0.0022 –0.1252* (0.0216) (0.0354) (0.0207) (0.0427) (0.0452)
Asia–Pacific –0.2520* –0.1614 –0.2672* –0.2971 –0.5644* (0.1107) (0.1775) (0.1095) (0.2111) (0.2180)
Others –0.0356 0.1407* –0.0277 0.1734* 0.1457* (0.0306) (0.0479) (0.0290) (0.0581) (0.0571)
Age 0–17 years 0.5069* –0.0845 0.5096* 0.0684 0.5780* (0.0412) (0.0699) (0.0412) (0.0852) (0.0944)
Age 18–24 years 0.2865* –0.1988* 0.2762* –0.1568 0.1193 (0.0520) (0.0872) (0.0519) (0.1113) (0.1280)
Age over 65 years 0.2353* 0.1999* 0.2512* 0.3451* 0.5963* (0.0403) (0.0661) (0.0392) (0.0804) (0.0870)
Female-headed 0.0626* –0.1390* 0.0556* –0.1575* –0.1019 (0.0263) (0.0423) (0.0273) (0.0536) (0.0598)
English speaking 0.1926* –0.0770 0.1898* –0.0333 0.1566* (0.0300) (0.0438) (0.0280) (0.0517) (0.0512)
Some college education –0.1155* 0.0927* –0.1114* 0.0805* –0.0309 (0.0157) (0.0241) (0.0152) (0.0291) (0.0298)
Three or more workers –0.2741* –0.0611 –0.2822* –0.1750* –0.4572* (0.0304) (0.0503) (0.0301) (0.0621) (0.0678)
Employment variable
Unemployment rate 0.1839* –0.0378 0.1864* 0.0125 0.1989* (0.0322) (0.0457) (0.0320) (0.0531) (0.0467)
Agriculture 0.0743* –0.2133* 0.0613* –0.2518* –0.1905* (0.0229) (0.0374) (0.0229) (0.0457) (0.0456)
Manufacturing –0.0640* –0.0790* –0.0705* –0.1246* –0.1951* (0.0177) (0.0295) (0.0177) (0.0364) (0.0378)
Public utility –0.0039 –0.0194 –0.0061 –0.0227 –0.0287 (0.0275) (0.0461) (0.0272) (0.0574) (0.0615)
Wholesale and retail trade –0.0195 –0.0492 –0.0236 –0.0685 –0.0920 (0.0218) (0.0383) (0.0215) (0.0493) (0.0530)
Finance and insurance –0.2670* 0.1226 –0.2631* 0.0708 –0.1923 (0.0474) (0.0936) (0.0479) (0.1237) (0.1423)
Arts –0.2244 −1.3036* –0.3214 −1.7880* −2.1094* (0.2128) (0.4382) (0.2232) (0.5663) (0.6520)
Time period fixed effects variable
1990 2.4721 (4.3698)
2000 –0.4658 (4.1937)
2010 −2.0063 (4.1941)
Government funding and poverty reduction in the Southern US 665
Papers in Regional Science, Volume 94 Number 3 August 2015.
None of the time period fixed effects (i.e., 1990, 2000, 2010 dummy variables) are signifi- cant, indicating that poverty rates do not have any significant temporal pattern. The lack of temporal variation in poverty rates reaffirms the persistence of poverty in the South over the 1990–2010 period. The significant direct, indirect, and total effects of the poverty hot-spot dummy variable indicates that poverty rates are higher in counties within and around poverty hot-spots compared with other counties in the South.
The education funding variable was significant, as were the interactions with the poverty hot-spot dummy variable for two government funding variables (i.e., health and hospital funding and education funding). The significant direct effect and non-significant total effect of the education funding variable show that education funding in a county is a significant factor in poverty mitigation in that county but not in the South as a whole. The significant interactions suggest that health and hospital funding and education funding have significant effects on poverty rates within the poverty hot-spots.
A $1 increase in per capita education funding in a hot-spot county decreases poverty rates by 0.0018 per cent and 0.0022 per cent within the own county and within all hot-spot counties, respectively. On the other hand, a $1 increase in a county in the South decreases poverty rates by 0.0006 per cent in the own county but has no effect on the poverty rate in the entire South, after accounting for the spatial spillover effects. These results support the hypothesis that education funding effectively reduced poverty rates within poverty hot-spots as a whole, and confirm the findings of other researchers that education expenditures are correlated with earn- ings and can decrease the poverty rate (Rodgers and Rodgers 1993; Bhola 2006; Perry 2006; Rynell 2008). In addition, the indirect effect of education funding is not significant. The implication is that most of the total effect occurs within the county where the funds are allocated.
Table 2. Continued
Variables Parameter of non-lagged
variable
Parameter of spatially-lagged
variable
Direct effect
Indirect effect
Total effect
Poverty hot-spot
Poverty hot-spot 2.7294* 0.2068 2.7888* 1.1945* 3.9833* (0.4090) (0.5443) (0.4029) (0.5991) (0.5335)
Governmental funding variable
Health and hospitals –0.0003 –0.0031 –0.0004 –0.0038 –0.0043 (0.0013) (0.0026) (0.0011) (0.0034) (0.0039)
Education –0.0006* 0.0004 –0.0006* 0.0003 0.0003 (0.0003) (0.0004) (0.0003) (0.0005) (0.0006)
Public welfare 0.0019 0.0020 0.0021 0.0032 0.0052 (0.0014) (0.0024) (0.0014) (0.0031) (0.0035)
Governmental funding variable – in poverty hot-spots
Health and hospitals 0.0028 0.0129* 0.0037 0.0179* 0.0216* (0.0025) (0.0053) (0.0026) (0.0069) (0.0080)
Education –0.0018* 0.0001 –0.0018* –0.0004 –0.0022* (0.0004) (0.0006) (0.0004) (0.0006) (0.0006)
Public welfare –0.0005 –0.0046 –0.0010 –0.0062 –0.0072 (0.0048) (0.0102) (0.0047) (0.0128) (0.0136)
Notes: Spatial fixed effects are not presented (40 county fixed effect variables were significant out of 1420 counties). * p < 0.05.
S. Jung et al.666
Papers in Regional Science, Volume 94 Number 3 August 2015.
A $1 increase in per capita health and hospital funding in a hot-spot county increases the poverty rates by 0.0179 per cent and 0.0216 per cent in neighbouring counties and in all hot-spot counties, respectively. The positive sign for the indirect effect implies that health and hospital funding in a hot-spot county is associated with higher poverty rates in neighbouring hot-spot counties. The results may indicate that counties with more heavily subsidized health and hospital systems are regional healthcare centres within poverty hot-spots that are surrounded by more rural, less populated counties with lower costs of living. Every county within a poverty hot-spot cannot have a well-developed healthcare and hospital system because they do not have sufficient population to efficiently support one. Nevertheless, those counties may have lower costs of living, providing the poor with opportunities to live more comfortably in surrounding counties and still be within driving distance of a good healthcare system. This finding coincides with previous research. Andrulis and Duchon (2007) found disparities in the numbers of hospitals between urban and suburban areas and the number of suburban hospitals in high poverty areas had been decreasing the most. Felland et al. (2009) found that there is an increased demand from suburban low-income people for urban healthcare facilities because of several barriers such as limited transportation and insufficient awareness.
The insignificance of both the public welfare variable and the interaction of the poverty hot-spot dummy variable with public welfare funding suggests rejection of the hypothesis that public welfare funding provides a poverty-reducing effect. This finding implies a critical view of public welfare spending as a poverty-reducing measure in the South both within and outside of poverty hot-spots, and supports the view of critics of the poverty-reducing effects of public welfare programmes (Stigler 1970; Okun 1975; Arrow 1979; Friedman and Friedman 1979; Murray 1984; Butler and Kondratas 1987; Lee 1987; Lindbeck et al. 1994; Crook 1997). Our results are worthy of consideration given recent attention to the Medicaid programme in debates about budget reductions and the Affordable Care Act (Manchikanti et al. 2011; Sommers and Epstein 2011; Blumberg 2012; HHS 2013). Even though our public welfare variable is not designed to isolate the effect of Medicaid spending, the information is worthy of attention in reference to policy-making for poverty reduction.
4.2 Predicted poverty rates and average marginal effect of education funding
The predicted poverty rates and average marginal effect of education funding inside and outside of poverty hot-spots are presented in Table 3. The predicted poverty rate, averaged across the three years and the three poverty hot-spots (28.72%), is higher than the average outside the poverty hot-spots (16.89%) and in the entire South (18.42%). In particular, the average across
Table 3. Average predicted poverty rates inside and outside of poverty hot-spots
Predicted poverty rate (%) Average marginal effect
Regions No. of Obs.
1990 2000 2010 Average of 1990, 2000, 2010
Education funding
Poverty hot-spots Texas 15 40.17 34.55 32.15 35.62 –0.00219 Mississippi-Delta 128 30.31 26.18 26.83 27.77 –0.00223 Central- Appalachia 40 32.75 27.12 27.60 29.16 –0.00223 Sum of all hot-spots 183 31.65 27.07 27.43 28.72 –0.00223
Outside of poverty hot-spots 1,237 18.32 15.44 16.91 16.89 –0.00031
The entire South 1,420 20.04 16.94 18.27 18.42 –0.00031
Government funding and poverty reduction in the Southern US 667
Papers in Regional Science, Volume 94 Number 3 August 2015.
the three years is the highest in the Texas poverty hot-spot (35.62%). The higher predicted poverty rates in the Texas poverty hot-spot may be explained by Texas having the highest percentage of immigrants among the states in the South. These immigrants lack health insurance and their unemployment rate is higher compared with other residents, leading to higher poverty rates (Rector 2006; Camarota 2012).
County-average predicted poverty rates decreased by 5.62 per cent, 4.13 per cent, 5.63 per cent, 2.88 per cent, and 3.10 per cent between 1990 and 2000 in the Texas, Mississippi Delta, and Central Appalachia poverty hot-spots, outside of the three poverty hot-spots, and in the entire South, respectively. The decrease in the entire South corresponds with US poverty trends of the 1990s (Lichter and Campbell 2005). Between 2000 and 2010, the county-average predicted poverty rate increased slightly by 0.65 per cent and 0.48 per cent in the Mississippi- Delta and Central-Appalachia poverty hot-spot, respectively, while the average poverty rate decreased by 2.4 per cent in the Texas poverty hot-spot.
The marginal effect of education funding differs slightly across poverty hot-spots. The counties in the Texas poverty hot-spots had an average marginal poverty-reducing effect of 0.00219 per cent for a $1 per capita increase in funds allocated to education. The other poverty hot-spots had an average marginal poverty-reducing effect of 0.00223 per cent, slightly higher than the Texas poverty hot-spot. The smaller marginal effect of education funding in the Texas hot-spot reflects the highest government education funding and poverty rate among the three poverty hot-spots for all three time periods. This might result from Texas having a higher rate of immigration (i.e., 14% of undocumented immigrants in the nation lived in Texas in 1996; HRO 2001) and being located on the border with Mexico. The higher rate of immigration likely increased the poverty rate in the region, triggering higher spending for education because both legal and illegal immigrants’ children are entitled for education unlike other government expenditures (Bernsen 2006).
5 Conclusions
This research evaluated the effects on poverty rates of state and federal government funding allocations to county governments for education, health and hospitals, and public welfare. We employed a recently developed econometric framework that incorporates both the spatial and persistent natures of poverty and used data from 1990, 2000, and 2010 for 1,420 counties across 16 states in the South. Based on the estimates from the empirical model, we discussed marginal effects of government funding allocations on poverty rates in a subset of counties identified through LISA analysis as poverty hot-spots. Our use of LISA statistics is an advantage of our research in that prior knowledge of large poverty hot-spots allowed us to focus more efficiently on the spatial relationship between government funding and poverty reduction in large multi- county poverty hot-spots where persistent poverty is an issue.
Our analysis found that increases in education funding in a hot-spot county reduce the poverty rates of the county and the neighbouring hot-spot counties. We also found that higher health and hospital funding in a hot-spot county is associated with higher poverty rates in neighbouring hot-spot counties and that public welfare funding is not effective in mitigating poverty either within or outside of poverty hot-spots. These findings suggest that education is the only budget category with significant poverty reducing effects within poverty hot-spots while health and hospital funding in a hot-spot county is associated with higher poverty rates in neighbouring hot-spot counties. Our findings also show that the marginal effects of education funding differ across the three poverty hot-spots.
We learn from these implications that: (i) providing education funding specifically targeting impoverished hot-spots can reduce poverty; (ii) funding state-supported regional healthcare and
S. Jung et al.668
Papers in Regional Science, Volume 94 Number 3 August 2015.
hospital systems within a poverty hot-spot can provide a benefit to poor families in lower-cost- of-living counties surrounding the major healthcare hub; (iii) redesigning of public welfare programmes aimed at mitigating poverty may be needed if poverty reduction is truly their objective; and (iv) using site-specific information may be useful in prioritizing funding decisions among the three poverty hot-spots.
Future research could incorporate the bias-corrected maximum likelihood estimator or the quasi-maximum likelihood estimator for spatial dynamic panel data proposed by Lee and Yu (2010b) and Yu et al. (2008). The estimator of the SDMP with fixed effects based on Elhorst (2003, 2010) we used in our study is time-consistent and asympototically normal when time t is large for given number of cross-sectional units n (Lee and Yu 2010b). However, the approach of estimating the SDMP with fixed effects applied in our case may yield inconsistent parameter estimates when n is large and t is small (e.g., n = 1,420 and t = 3 in our case) (Lee and Yu 2010b; Lee and Yu 2010c). Thus, the bias correction procedure may provide consistent parameter estimates.
Also, future research could focus more attention on smaller clusters of counties with persistent poverty. For example, of the 320 counties in the 16 states that were classified as persistent-poverty counties by ERS in 2010 (ERS USDA 2008), 156 counties were included in the poverty hot-spots identified by the LISA analysis. Thus, more detailed regional poverty reduction strategies can be evaluated in future research by closely examining smaller poverty clusters. Likewise, the effectiveness of more detailed government funding for poverty-related programmes could be investigated in future research. Although isolating the effect of funding for a single programme such as Medicaid is beyond the scope of our research, our modelling framework could be used for such analysis. Medicaid funding was not included in our analysis because data were not available for 1987 (CMS 2013), and we were interested in the effects of broad categories of funding directed at poverty reduction. As more consistent historical data become available, such analysis can be undertaken.
Government funding and poverty reduction in the Southern US 669
Papers in Regional Science, Volume 94 Number 3 August 2015.
Appendix 1
Table A1. Comparison of parameter estimates of the spatial Durbin panel model with and without the employment and demographic data variables
Variables Full model Reduced-form model
No employment data No employment, race, and age
Parameter of non-lagged
variable
Parameter of spatially-lagged
variable
Parameter of non-lagged
variable
Parameter of spatially-lagged
variable
Parameter of non-lagged
variable
Parameter of spatially-lagged
variable
Poverty variable
Individual poverty rate 0.2684* 0.3314* 0.3865* (0.0202) (0.0196) (0.0186)
Time lag of individual poverty rate
–0.0212 0.0631* –0.0281 0.0533 0.0253 0.0137 (0.0183) (0.0282) (0.0186) (0.0280) (0.0189) (0.0278)
Demographic variable
White –0.1231* 0.0306 –0.1386* 0.0433 (0.0216) (0.0354) (0.0219) (0.0361)
Asia–Pacific –0.2520* –0.1614 –0.3125* –0.1099 (0.1107) (0.1775) (0.1132) (0.1787)
Others –0.0356 0.1407* –0.0393 0.1626* (0.0306) (0.0479) (0.0312) (0.0489)
Age 0–17 years 0.5069* –0.0845 0.5217* –0.0420 (0.0412) (0.0699) (0.0414) (0.0695)
Age 18–24 years 0.2865* –0.1988* 0.3210* –0.1879* (0.0520) (0.0872) (0.0527) (0.0871)
Age over 65 years 0.2353* 0.1999* 0.2598* 0.2780* (0.0403) (0.0661) (0.0410) (0.0661)
Female-headed 0.0626* –0.1390* 0.0621* –0.1646* 0.1349* –0.1700* (0.0263) (0.0423) (0.0268) (0.0427) (0.0265) (0.0412)
English speaking 0.1926* –0.0770 0.1981* –0.1657* 0.2437* –0.2676* (0.0300) (0.0438) (0.0303) (0.0431) (0.0295) (0.0410)
Some college education –0.1155* 0.0927* –0.1232* 0.0952* –0.1329* 0.1104* (0.0157) (0.0241) (0.0154) (0.0184) (0.0159) (0.0191)
Three or more workers –0.2741* –0.0611 –0.2914* –0.1661* –0.2942* –0.2438* (0.0304) (0.0503) (0.0310) (0.0493) (0.0323) (0.0498)
Employment variable
Unemployment rate 0.1839* –0.0378 (0.0322) (0.0457)
Agriculture 0.0743* –0.2133* (0.0229) (0.0374)
Manufacturing –0.0640* –0.0790* (0.0177) (0.0295)
Public utility –0.0039 –0.0194 (0.0275) (0.0461)
Wholesale and retail trade –0.0195 –0.0492 (0.0218) (0.0383)
Finance and insurance –0.2670* 0.1226 (0.0474) (0.0936)
Arts –0.2244 −1.3036* (0.2128) (0.4382)
Time period fixed effects variable
1990 2.4721 1.6270 1.6872 (4.3698) (3.8951) (0.8686)
2000 –0.4658 –0.3559 –0.4146 (4.1937) (3.8438) (0.9582)
2010 −2.0063 −1.2711 −1.2725 (4.1941) (3.7991) (0.9214)
S. Jung et al.670
Papers in Regional Science, Volume 94 Number 3 August 2015.
Table A1. Continued
Variables Full model Reduced-form model
No employment data No employment, race, and age
Parameter of non-lagged
variable
Parameter of spatially-lagged
variable
Parameter of non-lagged
variable
Parameter of spatially-lagged
variable
Parameter of non-lagged
variable
Parameter of spatially-lagged
variable
Poverty hot-spot
Poverty hot-spot 2.7294* 0.2068 2.6742* –0.1890 3.2504* 0.7977 (0.4090) (0.5443) (0.4192) (0.5561) (0.4395) (0.5579)
Governmental funding variable
Health and hospitals –0.0003 –0.0031 0.0024 –0.0013 0.0008 –0.0030 (0.0013) (0.0026) (0.0025) (0.0027) (0.0014) (0.0028)
Education –0.0006* 0.0004 –0.0008* 0.0005 –0.0006* –0.0002 (0.0003) (0.0004) (0.0003) (0.0004) (0.0003) (0.0005)
Public welfare 0.0019 0.0020 0.0014 0.0006 0.0019 0.0014 (0.0014) (0.0024) (0.0014) (0.0024) (0.0015) (0.0025)
Governmental funding variable – in poverty hot-spots
Health and hospitals 0.0028 0.0129* 0.0024 0.0097 0.0009 0.0052 (0.0025) (0.0053) (0.0025) (0.0054) (0.0027) (0.0056)
Education –0.0018* 0.0001 –0.0017* 0.0008 –0.0023* –0.0006 (0.0004) (0.0006) (0.0004) (0.0006) (0.0005) (0.0006)
Public welfare –0.0005 –0.0046 0.0008 –0.0042 0.0008 –0.0033 (0.0048) (0.0102) (0.0049) (0.0104) (0.0052) (0.0109)
Notes: Spatial fixed effects are not presented. * p < 0.05.
Appendix 2
From Equation (3), the total marginal effect of kth explanatory variable x in a given county (i = 1) on P is:
∂ ∂
= −( ) +[ ]− P
x W I W
k k N k
1
1 I ρ β φ , (A1)
which can be reexpressed as:
∂ ∂
= −( )
+⎡
⎣
⎢ ⎢ ⎢ ⎢
⎤
⎦
⎥ ⎥ ⎥ ⎥
−P
x W
w
w
w k
k k
N
1
1
11
21
1
I ρ
β φ φ
φ �
. (A2)
Government funding and poverty reduction in the Southern US 671
Papers in Regional Science, Volume 94 Number 3 August 2015.
Let sij be an (i, j) element of (I − ρW)−1, then Equation (A2) becomes:
∂ ∂
=
+
+
+
=
=
∑
∑P x
s s w
s s w
s s w
k
k k j j j
n
k k j j j
n
n k k nj
1
11 1 1 1
21 2 1 1
1
β φ
β φ
β φ
�
jj j
n
1 1=
∑
⎡
⎣
⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢
⎤
⎦
⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥
. (A3)
The first element of the matrix in Equation (A3) is the direct effect of the kth explanatory variable x on P in a given county (i = 1) while the other elements are the indirect effects of kth explanatory variable x on P in other counties. Therefore, the total marginal effect is the sum of all elements in the matrix.
References
Andrulis DP, Duchon LM (2007) The changing landscape of hospital capacity in large cities and suburbs: Implications for the safety net in metropolitan America. Journal of Urban Health 84: 400–414
Afonso A, Aubyn St M (2004) Non-parametric approaches to education and health expenditure efficiency in the OECD, ISEG/UTL-Technical University of Lisbon, Mimeo
Allard SW, Tolman R, Rosen D (2003) Proximity to service providers and service utilization among welfare recipients: The Interaction of Place and Race. Journal of Policy Analysis and Management 22: 599–613
Anselin L (1988) Lagrange multiplier test diagnostics for spatial dependence and spatial heterogeneity. Geographical Analysis 20: 1–17
Anselin L (1995) Local indicators of spatial association – LISA. Geographical Analysis 27: 93–115 Anselin L, Bera A, Florax RJGM, Yoon M (1996) Simple diagnostic tests for spatial dependence. Regional Science and
Urban Economics 26: 77–104 Arrow KJ (1979) The trade-off between growth and equity. In: Greenfield HI, Levenson AM, Hamovitch W, Rotwein
E (eds) Theory for economic efficiency: Essays in honor of Abba P. Lerner. MIT Press, London Baingana F, Alem A, Jenkins R (2006) Mental health and the abuse of alcohol and controlled substances. In: Jamison
DT, Feachem RG, Makgoba MW, Bos ER, Baingana FK, Hofman KJ, Rogo KO (eds) Disease and mortality in Sub-Saharan Africa. The World Bank, Washington, DC
Beale CL (2004) Anatomy of nonmetro high poverty areas: Common in plight distinctive in nature. Amber Waves 2, February: 21–22
Beeson P, Eberts R (1989) Identifying productivity and amenity effects in interurban wage differentials. Review of Economics and Statistics 71: 443–452
Bernsen JA (2006) The costs of illegal immigration to Texas. Lone Star Foundation, Austin, TX Bhargava A, Jamison D, Lau L, Murray C (2001) Modeling the effects of health on economic growth. Journal of Health
Economics 20: 423–440 Bhola HS (2006) Adult and lifelong education for poverty reduction: A critical analysis of contexts and conditions.
International Review of Education 52: 231–246 Bishaw A (2009) Poverty: 2007 and 2008 American Community Surveys. URL: http://www.census.gov/prod/2009pubs/
acsbr08-1.pdf Bishaw A (2012) Poverty: 2010 and 2011. American Survey Briefs. URL : http://www.census.gov/prod/2012pubs/
acsbr11–01.pdf Blank RM (2005) Poverty, policy and place: How poverty and policies to alleviate poverty are shaped by local
characteristics. International Regional Science Review 28: 441–464 Blomquist G, Berger M, Hoehn J (1988) New estimates of the quality of life in urban areas. American Economic Review
78: 89–107 Bloom D, Canning D (2000) Health and the wealth of nations. Science 287: 1207–1209 Blumberg LJ (2012) The affordable care act: What does it do for low-income families? Institute for Research on Poverty.
Fast Focus No. 15. University of Wisconsin, Madison
S. Jung et al.672
Papers in Regional Science, Volume 94 Number 3 August 2015.
Burridge P (1980). On the cliff-ord test for spatial correlation. Journal of the Royal Statistical Society B 42: 107–108 Butler S, Kondratas A (1987) Out of the poverty trap. Free Press, New York; Collier Macmillan, London Camarota SA (2012) Immigrants in the United States, 2010. A profile of America’s foreign-born population. Centre for
Immigration Studies. URL: http://www.cis.org/2012-profile-of-americas-foreign-born-population Cho S, Kim SG, Clark CD, Park WM (2007) Spatial analysis of rural economic development using a locally weighted
regression model. Agricultural and Resource Economics Review 36: 24–38 CMS (2013) CMS – 64 quarterly expense report. Centre for Medicare and Medicaid Services. URL: http://
medicaid.gov/Medicaid-CHIP-Programme-Information/By-Topics/Data-and-Systems/MBES/CMS-64-Quarterly- Expense-Report.html
Cook JT, Frank DA (2008) Food security, poverty, and human development in the United States. Annals of the New York Academy of Sciences 1136: 193–209
Cook K, Dranove D, Sfekas A (2010) Does major illness cause financial catastrophe? Health Services Research 45: 418–436
Crook C (1997) The future of the state. Economist, 20 September. URL: http://www.economist.com/node/850929 Debarsy N, Ertur C, LeSage JP (2012) Interpreting dynamic space-time panel data models. Statistical Methodology 9:
158–171 DeNavas-Walt C, Bernadette DP, Smith J (2007) Income, poverty, and health insurance coverage in the United States:
2006 US Census Bureau, Current Population Reports. US Government Printing Office, Washington DC Duncan CM (ed) (1992) Rural poverty in America. Auburn, New York Elhorst JP (2003) Specification and estimation of spatial panel data models International Regional Science Review 26:
244–268 Elhorst JP (2010) Spatial panel data models. In: Fischer MM, Getis A (eds) Handbook of applied spatial analysis.
Springer, Berlin Elhorst JP (2012) Dynamic spatial panels: Models, methods and inferences. Journal of Geographical Systems 14: 5-28 ERS USDA (2004) Natural amenity scale. URL: http://www.ers.usda.gov/data-products/natural-amenities-scale.aspx ERS USDA (2007) Measuring rurality: Urban influence codes. URL: http://www.ers.usda.gov/Briefing/rurality/
UrbanInf ERS USDA (2008) Persistent poverty counties, 1970–2000. URL: http://www.ers.usda.gov/data-products/county-
typology-codes.aspx Fan S, Chan-Kang C (2008) Regional road development, rural and urban poverty: Evidence from China. Transport
Policy 15: 305–314 Fan S, Zhang L, Zhang X (2002) Growth, inequality and poverty in rural China: The role of public investment.
International Food Policy Research Institute Policy Report 125 Felland LE, Lauer J, Cunningham PJ (2009) Suburban poverty and the health care safety net. HSC Research Brief No.
13 Friedman M, Friedman R (1979) Free to choose. Harcourt Brace Jovanovich, New York Friedman S, Lichter DT (1998) Spatial inequality and poverty among African American children. Population Research
and Policy Review 17: 91–109 Fujiura GT, Yamaki K (2000) Trends in demography of childhood poverty and disability. Exceptional Children 66:
187–199 Glaeser E, Kallal H, Scheinkman J, Shleifer A (1992) Growth of cities. Journal of Political Economy 100: 1126–1152 Greene WH (2010) Econometric Analysis (7th edn). Prentice Hall, Upper Saddle River Glennerster H (2002) United States poverty studies and poverty measurement: The past twenty-five years. Social Service
Review 76: 83–107 Hausman JA (1978) Specification Tests in Econometrics. Econometrica 46: 1251–1271 Himmelstein DU, Warren E, Thorne D, Woodhandler SJ (2005) Illness and injury as comtributors to bankruptcy.
Working paper. URL: http://ssrn.com/abstract=664565 HHS (2013) Healthcare: about the law. Department of Health and Human Services. URL: http://www.hhs.gov/
healthcare/rights/index.html HRO (2001) Health care for undocumented immigrants: Who pays? Austin: House Research Organization. Texas House
of Representatives. October. Report 77–13 Jung H, Thorbecke E (2003) The impact of public education expenditure on human capital, growth, and poverty in
Tanzania and Zambia: A general equilibrium approach. Journal of Policy Modeling 25: 701–725 Kaldor N (1970) The case for regional policies. Scottish Journal of Political Economy 17: 337–348 Kenworthy L (1999) Do social-welfare policies reduce poverty? A cross-national assessment. Social Forces 77:
1119–1139 Krugman P (1991) History versus expectations. Quarterly Journal of Economics 106: 651–667 Lee DR (1987) The tradeoff between equality and efficiency: Short-run politics and long-run realities. Public Choices
53: 149–165
Government funding and poverty reduction in the Southern US 673
Papers in Regional Science, Volume 94 Number 3 August 2015.
Lee L, Yu J (2010a) Some recent developments in spatial panel data models. Regional Science and Urban Economics 40: 255–271
Lee L, Yu J (2010b) A spatial dynamic panel data model with both time and individual fixed effects. Economic Theory 26: 564–597
Lee L, Yu, J. (2010c) Estimation of spatial autoregressive panel data models with fixed effects. Journal of Econometrics 154: 165–185
LeSage JP, Pace RK (2009) Introduction to Spatial Econometrics. CRC Press Boca Raton, FL Levernier W, Partridge MD, Rickman DS (2000) The causes of regional variation in US poverty: A cross-county
analysis. Journal of Regional Science 40: 473–497 Lindbeck A, Molander P, Persson T, Petersson O, Sandmo A, Swedenborg B, Thygesen N (1994) Turning Sweden
around. MIT Press, Cambridge, MA Lichter DT, Campbell LA (2005) Changing patterns of poverty and spatial inequality in Appalachia. Population
Reference Bureau, Appalachian Regional Commission Lower-Basch E (2011) Goals for TANF reauthorization. TANF Policy Brief. 24 January Lyson T, Falk W (1993) Forgotten places: Uneven development in rural America. University Press of Kansas,
Lawrence, KS Manchikanti L, Caraway DL, Parr AT, Fellows B, Hirsch JA (2011) Patient protection and affordable care act of 2010:
Reforming health care reform for the new decade. Pain Physician 14: E35–E67 Meyer BD, Sullivan JX (2012) Identifying the disadvantaged: Official poverty, consumption poverty, and the new
supplemental poverty measure. Journal of Economic Perspectives 26: 111–136 Murray C (1984) Losing ground: American social policy, 1950–1980. Basic Books, New York Okun AM (1975) Equality and efficiency: The big tradeoff. Brookings Institution, Washington Partridge MD, Rickman DS (2005) High-poverty non-metropolitan counties in America: Can economic development
help? International Regional Science Review 28: 415–40 Partridge MD, Rickman DS (2005) (2006) The geography of American poverty: Is there a role for place-based policies?
The W.E. Upjohn Institute for Employment Research: Kalamazoo, MI Poston DL Jr, Singelmann J, Siordia C, Slack T, Robertson BA, Saenz R (2010) Spatial context and poverty: Area-level
effects and micro-level effects of household poverty in the Texas borderland and lower Mississippi Delta: United States, 2006. Applied Spatial Analysis and Policy 3: 139–162
Perry G (2006) Poverty reduction and growth: virtuous and vicious circles. The World Bank, Washington, DC Probst JC, Moore CG, Glover SH, Samuels ME (2004) Person and place: The compounding effects of race/ethnicity and
rurality on health. American Journal of Public Health 94: 1695–1703 Rector R (2006) Importing poverty: Immigration and poverty in the United States: A book of Charts. Heritage
Foundation Special Report No. 9 Reeder RJ, Calhoun SJ (2002) Federal funds in rural America: Payments vary by region and type of county. Rural
America 17: 1–3 Rodgers JR, Rodgers JL (1993) Chronic poverty in the United States. The Journal of Human Resources 28: 25–54 Rupasingha A, Goetz SJ (2007) Social and political forces as determinants of poverty: A spatial analysis. Journal of
Socio-Economics 36: 650–671 Rynell A (2008) Causes of poverty: Findings from recent research. The Heartland Alliance Mid-America Institute on
Poverty. October Salisbury CG (2004) Make an investment in our school children: Increase the nutritional value of school lunch
programmes. Brigham Young University Education and Law Journal 2: 331–352 Smeeding T (2006) Poor people in rich nations: The United States in comparative perspective. Journal of Economic
Perspectives 20: 69–90 Smith JP (1999) Health bodies and thick wallets: The dual relation between health and economic status. The Journal of
Economic Perspectives 13: 145–166 Sommers BD, Epstein AM (2011) Why states are so miffed about Medicaid – economics, politics, and the ‘Woodwork
effect’. The New England Journal of Medicine 365: 100–102 Stigler G (1970) Director’s law of public income redistribution. Journal of Law and Economics 13: 1–10 Swaminathan H, Findeis J (2004) Policy intervention and poverty in rural America. American Journal of Agricultural
Economics 86: 1289–1296 Triest RK (1997) Regional differences in family poverty. New England Economic Review Jan/Feb: 3–17 Trohanis PL (2008) Progress in providing services to young children with special needs and their families: An overview
to and update on the implementation of the Individuals with Disabilities Education Act (IDEA). Journal of Early Intervention 30: 140–151
US Census Bureau (2001a) United States census 1990 and 2000. URL: http://www.census.gov/main/www/ cen2000.html
S. Jung et al.674
Papers in Regional Science, Volume 94 Number 3 August 2015.
US Census Bureau (2001b) Government finance and employment classification manual. URL: http://www2.census.gov/ govs/class/classfull.pdf
US Census Bureau (2008) Data base on historical finances of local governments, Census of Governments. URL: http://www.census.gov/govs/state
US Census Bureau (2011a) Income, poverty, and health insurance coverage in the United States: 2010. URL: http:// www.census.gov/newsroom/releases/archives/income_wealth/cb11-157.html#tableb
US Census Bureau (2011b) United States census 2010. URL: http://2010.census.gov/2010census/ US Census Bureau (2011c) American Community Survey. URL: http://www.census.gov/acs/www/ US Census Bureau (2011d) Poverty thresholds. URL: http://www.census.gov/hhes/www/poverty/data/threshld US Department of Labour (2011) Data bases, tables, and calculators by subject: Employment. URL: http://www.bls.gov/
data/#employment Waidmann TA, Rajan S (2000) Race and ethnic disparities in health care access and utilization: An examination of state
variation. Medical Care Research and Review 57: 55–84 Weber BA, Jensen L, Miller K, Mosely J, Fisher M (2005) Critical review of rural poverty literature: Is there truly a rural
effect? International Regional Science Review 28: 381–414 Wimberley RC, Morris LV (1997) The southern Black Belt: A national perspective. TVA Rural Studies and the
University of Kentucky, Lexington, KY Wimberley RC, Morris LV (2002) The regionalization of poverty: Assistance for the Black Belt South? Southern Rural
Sociology 18: 294–306 Wooldridge JM (2009) Introductory econometrics: A modern approach (4th edn). South-Western College, Mason, OH Yu J, de Jong R, Lee L (2008) Quasi-maximum likelihood estimators for spatial dynamic panel data with fixed effects
when both n and T are large. Journal of Econometrics 146: 118–134
Government funding and poverty reduction in the Southern US 675
Papers in Regional Science, Volume 94 Number 3 August 2015.
© 2015 The Author(s). Papers in Regional Science © 2015 RSAI
Papers in Regional Science, Volume 94 Number 3 August 2015.
doi:10.1111/pirs.12089
:
Resumen. Esta investigación evalúa los impactos en las tasas de pobreza de los fondos del go- bierno para la educación, la salud y los hospitales, y el bienestar público asignado a la reducción de la pobreza para los condados con una alta pobreza persistente en el sur de los Estados Unidos. Nuestro análisis encontró que el aumento de fondos para la educación en un condado candente en cuanto a la pobreza reduce las tasas de pobreza de ese condado y de los condados candentes vecinos. Asimismo, encontramos que el aumento de la fi nanciación para la salud y los hospitales en un condado candente en pobreza se asocia con mayores tasas de pobreza en los condados candentes vecinos y que la fi nanciación del bienestar público no es efi caz en la mitigación de la pobreza, ya sea dentro o fuera de los puntos candentes de pobreza.
Copyright of Papers in Regional Science is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
