Exploring the Structural Determinants of Food Stamp Program Participation in the South: Does Place Matter? Principle Investigator: Tim Slack, Ph.D., Department of Sociology, Louisiana State University, 126 Stubbs Hall, Baton Rouge, LA 70803. Phone: (225) 578-1116. Fax: (225) 578-5102. Email: slack@lsu.edu. Graduate Research Assistant: Candice A. Myers, M.A., Department of Sociology, Louisiana State University.

2008 SRDC RIDGE Program Research Themes Addressed: 1. High Poverty Areas 2. Impact of Immigrants on Nutrition Programs 3. Use of Food/Nutrition Programs 4. Policy Choices

Acknowledgements: This project was supported by the Research Innovation and Development Grants in Economics (RIDGE) Program of the Economic Research Service (ERS), U.S. Department of Agriculture (USDA), in partnership with the Southern Rural Development Center (SRDC), Mississippi State University. Previous iterations of the analysis provided in this report were presented at the 2009 mid-year (Atlanta, GA) and annual meetings (Washington, DC) of the RIDGE Program as well as the 2009 annual meetings of the Mid-South Sociological Society (Lafayette, LA), Rural Sociological Society (Madison, WI), and Southern Sociological Society (New Orleans, LA). Special thanks to Huizhen Niu of the Agricultural Economics Geographic Information Systems (AEGIS) Lab in the Department of Agricultural Economics and Agribusiness, LSU AgCenter, for her assistance in developing the maps presented in this report and used to diagnose spatial effects.

Project Summary

This report extends prior research on food assistance in the U.S. South by exploring the

relationship between county-level Food Stamp Program (FSP) participation and the local social

structure. Our research design is comparative in focus. We examine how the structural

determinants of FSP participation in the South differ in comparison to the non-South. We also

assess whether such factors differ in high poverty areas of the South—namely, the Black Belt,

Central Appalachia, Mid-South Delta, and Texas Borderland—in comparison to other regions of

the country. In short, we find that geographic space and place matter. Our results show that the

South—and high poverty areas in the South, in particular—are home to high regional

concentrations of FSP participation. Further, our results show that the structural determinants of

FSP participation differ between the South and non-South, and between high poverty areas and

the rest of the country, in significant ways. This research contributes to the scientific literature

on food assistance by: 1) bringing attention to the importance of local space and place by

building on the limited number of FSP studies that have focused on a scale between that of

individuals/households and states; and 2) utilizing unique data from the USDA Food and

Nutrition Service (FNS) that have yet to be assessed in the research literature. This study

addresses four of the seven 2008 SRDC RIDGE Program research themes, including: High

Poverty Areas; Impact of Immigrants on Nutrition Programs; Use of Food/Nutrition Programs;

and Policy Choices. All of these themes are of importance to public policy aimed at combating

food insecurity and better serving vulnerable populations in the South. It is our hope that the

information presented in this report will prove useful to policy makers and practitioners by

helping to identify communities that are likely to have special food assistance needs and by

helping to anticipate how local social change may influence such needs in the future.

Introduction and Problem Statement

This report extends prior research on food assistance in the U.S. South by exploring the

relationship between county-level Food Stamp Program (FSP) participation and the local social

structure. To do so, we take an explicitly comparative focus. We examine how the structural

determinants of FSP participation in the South differ in comparison to the non-South. We also

assess whether such factors differ in high poverty areas of the South—namely, the Black Belt,

Central Appalachia, Mid-South Delta, and Texas Borderland—in comparison to other regions of

the country. Our goal is to contribute to the scientific literature on food assistance by: 1)

bringing attention to the importance of local space and place by building on the limited number

of FSP studies that have focused on a scale between that of individuals/households and states;

and 2) utilizing unique data from the USDA Food and Nutrition Service (FNS) that have yet to

be assessed in the research literature. Our approach addresses four of the seven 2008 SRDC

RIDGE Program research themes, including: High Poverty Areas; Impact of Immigrants on

Nutrition Programs; Use of Food/Nutrition Programs; and Policy Choices. All of these themes

are of importance to public policy aimed at combating food insecurity and better serving

vulnerable populations in the South. It is our hope that the information presented in this report

will prove useful to policy makers and practitioners by helping to identify communities that are

likely to have special food assistance needs and by helping to anticipate how local social change

may influence such needs in the future.

Geographic space and place are increasingly being recognized as pivotal axes of social

inequality (Gans 2002; Gieryn 2000; Lobao 2004; Lobao and Saenz 2002; Lobao, Hooks, and

Tickamyer 2007; Massey 1996; Tickamyer 2000). This is a promising development because of

the potential it allows for drawing out “both the importance of where actors are located in 1

geographic space and how geographic entities themselves are molded by and mold stratification”

(Lobao et al. 2007: 3). Scholars concerned with spatial inequality have called for special

attention to issues of comparative advantage (and disadvantage) across space as well as to the

consideration of the subnational scale (Lobao 2004; Lobao and Hooks 2007; Tickamyer 2000). This

study speaks directly to these ideas.

Research has clearly demonstrated enduring relationships between geographic location

and economic hardship (Lichter and Johnson 2007; Lyson and Falk 1993; Massey and Denton

1993; Rural Sociological Society Task Force on Persistent Rural Poverty 1993; Slack et al. 2009;

Tickamyer and Duncan 1990; Wilson 1980). A prime example of this is the high and persistent

poverty that characterizes the South in comparison to other regions of the U.S., and areas such as

the Black Belt, Central Appalachia, Mid-South Delta, and Texas Borderland, in particular.

Related to the acute economic distress faced by many communities across the South is the fact that

the use of the FSP is much higher in the region compared to other areas of the country (Goetz,

Rupasingha, and Zimmerman 2004).

The FSP—more recently referred to as the Supplemental Nutrition Assistance Program or

SNAP—is the nation’s largest food assistance program and a critical component of the social safety

net for its least advantaged citizens (Nord 2001). The reach of the program is substantial, helping to

put “healthy food on the table for over 35 million people each month” (Food and

Nutrition Service 2009), a figure that equates to more than one-in-10 Americans. As such, the FSP

plays a critical role in addressing food assistance and nutrition issues faced by vulnerable populations, a

fact that is especially true in the South.

To date, the bulk of research on FSP dynamics has either focused on individual and household-level data drawn from large national panel surveys (Bhattarai, Duffy, and Raymond

2

2005; Frongillo, Jyoti, and Jones 2006; Gibson 2004, 2006; Gunderson and Oliveira 2001; Nord

2001; Van Hook and Balistreri 2006) or on state-level data (Figlio, Gunderson, and Ziliak 2000;

Tapogna et al. 2004; Wallace and Blank 1999; Wilde et al. 2000). Individual and household-

level studies provide important information on how the characteristics of individuals and

households influence FSP participation, but do not allow for a determination of the role of social

structure in influencing program outcomes. State-level studies, on the other hand, allow for the

identification of structural mechanisms that operate across geographic space, but potentially

mask important intra-state variation in FSP dynamics. A few recent studies have brought

attention to the importance of using county-level data to assess structural influences on the FSP

(Goetz et al. 2004; Hoyt and Scott 2004; Wilde and Dicken 2003). These studies have

highlighted significant spatial variation in FSP-related outcomes at the county level as well as

important aggregate mechanisms that shape these outcomes. For example, Goetz et al. (2004)

use FSP expenditure data to show that counties characterized by better labor market conditions

and higher per capita reductions in AFDC/TANF expenditures experienced the greatest declines

in FSP expenditures between 1995 and 1999. In a related vein, drawing on data for a sample of

U.S. states, Hoyt and Scott (2004) show strong relationships between county-level participation in

cash assistance programs (i.e., AFDC/TANF and SSI) and FSP participation. Last, Wilde and

Dicken (2003) use county-level administrative data from California to explore FSP participation by

race/ethnicity, showing Hispanics, in particular, to be a growing share of the FSP caseload in that

state. We use these studies as a point of departure in the analysis that follows. 3

Research Objectives

Our overarching aim is to assess how FSP participation varies across space and establish how

other place-based (i.e., county-level) characteristics serve to influence existing differences.

More specifically, we are interested in establishing how FSP participation in the South and high

poverty areas is related to five broad sets of county-level characteristics: labor market conditions,

population structure, human capital, residential context and inequality, and expenditures on cash

assistance programs (we elaborate on the independent variables that make-up each of these

dimensions in the following section). Accordingly, we examine three specific research

questions:

Q1: Does the prevalence of FSP participation differ between the South and non-South?

Q2: Does the prevalence of FSP participation differ between high poverty areas and other

regions of the country?

Q3: Do the aggregate mechanisms related to FSP participation differ between the regions of

interest?

Research Methods

Data and Measures

To address these questions we analyze data drawn from the Food and Nutrition Service (FNS)

and the Economic Research Service (ERS) of the USDA, and the U.S. Census Bureau. Our units of

analysis are counties. We elaborate on the definition of our measures below. 4

Dependent Variable

Our dependent variable—FSP participation—is based on data obtained from the FNS circa 2005.

The FNS collects FSP participation data twice a year for the report months of January and July.

We define FSP participation as the average percentage of a county’s total population receiving

FSP benefits in these two time periods across the years 2004, 2005, and 2006. While this

approach necessarily masks intra-county variation in FSP participation over the period, it has the

benefit of providing a very stable assessment of the typical level of participation in the program

at mid-decade.

The FNS collects and organizes its data by project area. A project area refers to the

political entity designated by the state as the administrative unit for program operations. In most states

project areas are counties, but in some states project areas range from entities such as cities to the

state as a whole. We exclude states (N=15) from our analysis that do not organize their FSP program

operations at the county level. These states are primarily located in the Northeast and Northwest.1 In

total, our sample is comprised of 2561 counties in 34 states. Figure 1

provides an illustration of our data coverage.

<Figure 1 about here> Regional Definitions

We begin by defining the South according to the geography delineated by the Census Bureau2

and then define the non-South as the remainder of the contiguous U.S. Given available data, the 1 Specifically, these states include Connecticut, Idaho, Maine, Massachusetts, Montana, Nebraska, New Hampshire, New York, Oregon, Rhode Island, Utah, Vermont, Washington, West Virginia, and Wyoming.

2 The U.S. Census Bureau defines the South as including the following states: Alabama, Arkansas, Delaware, Washington D.C., Georgia, Kentucky, Louisiana, Maryland, Mississippi, North Caroline, Oklahoma, South Carolina, Tennessee, Texas, Virginia, and West Virginia.

5

South includes at total of 1,330 counties in 16 states (West Virginia is the only state in the South

not included in the analysis) and the non-South includes a total of 1,231 counties in 18 states.3

We define the Texas Borderland as all Texas counties whose largest city is within 100 miles of

the U.S.-Mexico border (Saenz 1997; Slack et al. 2009). The Borderland stretches along the Rio

Grande River and includes a total of 41 counties. We define Central Appalachia according to the

geography delineated by the Appalachian Regional Commission (ARC).4 This region includes a

total of 78 counties in eastern Kentucky, northern Tennessee, and southwest Virginia.5 We

define the Delta by starting with the geography delineated by the Delta Regional Authority, but

further restrict our analysis to the core Delta counties in Arkansas, Louisiana, and Mississippi

(Slack et al. 2009).6 This region straddles the southern portion of the Mississippi River and

includes a total of 133 counties. Last, following Wimberley and Morris (1997), we define the

Black Belt as the crescent of counties spanning through the South from North Carolina down

through Texas, minus those counties included in our Delta definition. This region includes a total

of 494 counties in Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, North Carolina,

South Carolina, Tennessee, Texas, and Virginia. 3 Specifically, the non-South includes counties in the following states: Arizona, California, Colorado, Illinois, Indiana, Iowa, Kansas, Michigan, Minnesota, Missouri, Nevada, New Jersey, New Mexico, North Dakota, Ohio, Pennsylvania, South Dakota, and Wisconsin.

4 The complete Appalachian geography defined by the ARC includes a total of 420 counties in 13 states. We chose to focus on the central region because it is where economic hardship is most concentrated. 5 The ARC definition of Central Appalachia also includes 9 counties in West Virginia which are not represented in our analysis. 6 The complete Delta geography defined by the Delta Regional Authority includes a total of 240 counties in 8 states. We chose to focus on the core of this region because it better conforms to social, rather than political, definitions of the “Delta” and it is also where poverty is most concentrated

6

Additional Independent Variables

We assess the influence of a total of 20 additional independent variables tapping five dimensions

of place-based characteristics in relation to FSP participation. The five dimensions of

independent variables include county-level: labor market conditions, population structure, human

capital, residential context and inequality, and expenditures on cash assistance programs. All

independent variables are drawn from the U.S. Census Bureau or ERS circa 2000.7 We outline

and motivate our choice of independent variables below. Given the dearth of literature on the

structural determinants of FSP dynamics, we rely heavily on the poverty literature as a guide to

our variable choices with the expectation that aggregate factors associated with higher poverty

rates will also be associated with higher FSP participation and vice versa.

Labor Market Conditions. We draw on a range of variables that tap local labor market

conditions. Research has shown that labor market hardship is strongly related to poverty. We

therefore include the county-level unemployment rate with the expectation that higher

unemployment will be related to higher FSP participation. Consonant with McLaughlin,

Gardner, and Lichter (1999), we include a measure of secondary/peripheral sector employment

that captures the percentage of the civilian labor force employed in retail trade and particular

segments of the service sector (i.e., arts, entertainment, and recreation services; accommodation

and food services; and other services [primarily personal services and business and repair

services]), as well as agriculture, forestry, fishing, and hunting.8 We also include median

household income, an important measure of local economic development and well-being (Lobao

and Hooks 2003). In sum, we examine three variables that tap local labor market conditions: the 7 We rely on 2000 as the reference year for our independent variables because of the wealth of information made available by the decennial census.

8 Notably, this definition excludes higher skilled service work, including employment in educational services, and health care or social support services.

7

percentage unemployed, the percentage employed in secondary/peripheral sector jobs, and

median household income.

Population Structure. Several variables related to the population structure of places have

been demonstrated to be important predictors of poverty, and thus are expected to be associated

with FSP participation. For example, larger percentages of female-headed households are often

associated with lower income-levels and higher poverty (Lichter and McLaughlin 1995; Lobao

and Hooks 2003). Migration is also important, in that positive net migration can serve as a proxy

for a strong economy because migrants are drawn to areas with stronger job opportunities (Frey

and Liaw 2005), while areas characterized by population stagnation or loss tend to be associated

with higher poverty rates (Rupasingha and Goetz 2007). Age structure has also been shown to

be an important correlate of poverty, with places characterized by larger proportions of non-

working age populations tending to have higher poverty (Lobao and Hooks 2003). Immigration

is another factor that influences poverty. While research has shown poverty among immigrants

tends to exceed that of their native-born counterparts at the individual level (Crowley, Lichter,

and Qian 2006), at the aggregate level the percentage of the population that is foreign-born has

been shown to be associated with lower levels of poverty (Rupasingha and Goetz 2007), likely

reflecting the “pull” of immigrants to more attractive labor markets. Last, the literature reveals

that places characterized by higher minority concentrations tend to be characterized by higher

poverty rates (Friedman and Lichter 1998; Rupasingha and Goetz 2007; Saenz 1997; Voss et al.

2006). For these reasons, we examine six variables related to county-level population structure:

the percentage of households headed by single females with children, net migration, the

percentage of the population that is under 15 years and over 65 years of age (non-working age), the

percentage foreign-born, the percentage black, and the percentage Latino. 8

Human Capital. A large body of literature has demonstrated significant linkages between

aggregate-level human capital and poverty. Places with lower aggregate educational levels tend

to be home to higher poverty rates (Friedman and Lichter 1998; Rupasingha and Goetz 2007;

Voss et al. 2006). English-language fluency has also been identified as an important determinant

of poverty among immigrant populations (Crowley et al. 2006; Davila and Mora 2000; Davila,

Bohara, and Saenz 1993). We therefore examine two variables related to county-level human

capital in relation to FSP participation: the percentage of the population 25 years and older with

less than a high school degree (or equivalent) and the percentage of the population that does not

speak English well or at all.

Residential Context and Inequality. An extensive literature has examined the higher

incidence of poverty in rural as compared to urban areas (see Jensen, McLaughlin, and Slack

2003), though research has also documented lower Public Assistance receipt among the poor in

rural areas (Jensen and Eggebeen 1994). We therefore consider whether places are metropolitan,

micropolitan, or noncore in character (Economic Research Service 2007). Dummy variables are

employed for micropolitan and noncore areas with metropolitan serving as the reference

category. There is also an established literature that points to important relationships between

residential segregation and spatially concentrated hardship (Lichter et al. 2008; Massey and

Denton 1993). We therefore examine four different measures of county-level inequality: indexes

of dissimilarity for the poor versus non-poor, black versus white, and Latino versus white

populations, as well as the Gini coefficient of household income inequality. 9 We also consider

whether an area suffers from housing stress (yes/no) as defined by the ERS.10

9 The index of dissimilarity measures the evenness with which two groups are distributed across the component

geographic areas (i.e., census tracts) that make up a larger area (i.e., counties). In cases where two groups are perfectly segregated, the index of dissimilarity would take the maximum value of 100. In cases where two groups are perfectly integrated, the index of dissimilarity would take the minimum value of zero (Racial Residential Segregation Measurement Project 2008). The Gini coefficient of household income inequality is a ratio measure of income dispersion. In cases

9

Expenditures on Cash Assistance. Of the few county-level studies of FSP dynamics, both

Goetz et al. (2004) and Hoyt and Scott (2004) have shown significant positive relationships

between expenditures on means-tested cash assistance programs and participation in the FSP.

Therefore, we examine county per capita expenditures on TANF and SSI combined.

In sum, we plan to assess the influence of five different realms of variables on county-

level FSP participation: labor market conditions, population structure, human capital, residential

context and inequality, and expenditures on cash assistance. Table 1 provides descriptive

statistics for all of the variables employed in the analysis.

<Table 1 about here> Analytic Strategy

The analysis that follows is carried out using both descriptive statistics and Ordinary Least

Squares (OLS) regression. Our modeling strategy uses a lagged panel design, in which

independent variables are measured at an earlier point in time compared to the dependent

variable. As outlined above, our independent variables represent county-level measures circa

2000, while our dependent variable measures FSP participation circa 2005. An advantage of

lagged panel designs compared to simultaneous cross-sectional analysis (i.e., when all variables are

measured at the same point in time) is that the approach addresses problems associated with

endogeneity and simultaneity bias. where household incomes are perfectly unequal (i.e., one household accounts for all of the household income in a county, while all others have no income), the Gini coefficient would take the maximum value of 1. In cases where household incomes are perfectly equal (i.e., all households have exactly the same income), the Gini coefficient would take the minimum value of zero (Jones and Weinberg 2000).

10 A county is defined as experiencing housing stress if 30% or more of the households within a county had one or more of the following housing conditions in 2000: 1) lacked complete plumbing; 2) lacked complete kitchen; 3) paid 30% of more of income for owner costs or rent; or 4) had more than one person per room.

10

Before finalizing our models, we examined a series of regression diagnostics to assess

statistical threats to the accuracy of our models. Based upon these diagnostics, we use a natural

logarithm transformation to normalize the substantial skew in the distribution of county per

capita expenditures on TANF and SSI. To address multicollinearity we use factor analysis to

combine percent foreign-born, percent Latino, and percent with limited English proficiency into

a single factor score.11 We also tested for spatial effects in our data, an issue that has proven an

important consideration in aggregate analyses of poverty (Rupasingha and Goetz 2007; Voss et

al. 2006) and FSP expenditures (Goetz et al. 2004). These diagnostics revealed significant

regional clusters of counties with high and low FSP participation, necessitating the consideration

of spatial effects in our models (this is discussed further in the results section). Last, because

states vary in their approaches to FSP administration specifically and addressing poverty and

inequality more generally (Lobao and Hooks 2003), we control for state fixed effects.

Specifically, our models include 33 dummy variables (California serves as the reference group)

to control for state-specific county-invariant factors.

Results

Prevalence of FSP Participation

The geographic distribution of FSP participation is illustrated in Figure 2. Notable here are those

counties one standard deviation above the mean—“high” FSP participation—and those counties

one standard deviation below the mean—“low” FSP participation. Figure 2 clearly illustrates a

higher prevalence of FSP participation in the South and a lower prevalence of FSP participation

in the non-South. 11 We used principal component factor analysis with varimax rotation to combine these three variables into a single factor. All variables loaded on the same component and explain 86.9 percent of the variance in these variables among counties.

11

<Figure 2 about here>

Table 2 provides a description of FSP participation by region in tabular form. The data

show that county-level FSP participation rates are 5.0 percent higher in the South (13.2%)

compared to the non-South (8.2%) on average. The data also show that county-level FSP

participation rates in high poverty areas are well above the national average (10.4%), reaching

13.4 percent in the Black Belt, 18.5 percent in the Delta, 19.0 percent in the Borderland, and 21.9

percent in Appalachia, respectively.

Models of FSP Participation Spatial Effects As shown in Figure 2, county-level FSP participation is not randomly distributed across

geographic space. Rather there are clear regional concentrations of counties with both high and

low levels of participation. This clustering effect is quantified by the Moran’s I statistic, which

measures the degree to which the characteristics of spatial units are correlated with those of their

neighbors. In the absence of any spatial relationship the Moran’s I statistic has a mean of zero,

while a perfect positive correlation has a mean of 1, and a perfect negative correlation has a

mean of -1. As shown in Figure 3, the Moran’s I value for county-level FSP participation is

0.61, indicating significant positive spatial clustering—significant clustering of like values—

among counties.

Also shown in Figure 3 is the Moran scatterplot, which plots county-level FSP

participation (x-axis) against its spatial lag (y-axis) for each county. The data are standardized so

that units on the graph represent standard deviations from the mean. The spatial lag is calculated

as the standardized level of FSP participation averaged across a given county’s contiguous 12

neighbors. The quadrants of the scatterplot correspond to four types of spatial association: 1) the

upper right quadrant shows those counties with above average values on the dependent variable

that are neighbored by counties that also have above average values (high-high); 2) the lower left

quadrant shows those counties with below average values on the dependent variable that are

neighbored by counties that also have below average values (low-low); 3) the upper left quadrant

shows those counties with below average values on the dependent variable that are neighbored

by counties with above average values (low-high); and 4) the lower right quadrant shows those

counties with above average values on the dependent variable that are neighbored by counties

with below average values (high-low). The Moran’s I statistic discussed above (0.61) is the

slope of the line fitted to the data distributed across these quadrants. Substantively, what Figure

3 shows is that counties with high FSP participation tend to be surrounded by neighboring

counties that also have high FSP participation, while counties characterized by low FSP

participation levels tend to neighbored by counties that also exhibit low FSP participation.

<Figure 3 about here>

Figure 4 shows a Local Indicators of Spatial Association (LISA) map, which provides a

geographic illustration of the data presented in the Moran’s I scatterplot. The LISA map shades

significant clusters of counties falling into one of the four value dimensions displayed in the

scatterplot (high-high; low-low; low-high; and high-low), while counties that are not in

significant spatial clusters are not shaded. Notable here is that the regional clusters of high FSP

participation are all in the South and in high poverty areas in particular.

<Figure 4 about here>

13

Regression Analysis

In order to assess the relationship between FSP participation and regional location in the South, we

estimate region-specific models that regress the five dimensions of county characteristics— labor

market conditions, population structure, human capital, residential context and inequality, and

expenditures on cash assistance programs—on county-level FSP participation. Table 3

shows the results of these models.

<Table 3 about here>

In both the South and non-South, percent unemployed, percent households headed by

single mothers, percent non-working age, percent less than high school, and per capita

expenditures on cash assistance show significant positive relationships with FSP participation,

while the factor score tapping the presence of Latino immigrant populations shows a significant

negative relationship with FSP participation. It is also notable that the spatial lag term shows a

significant positive effect in both the South and non-South, reflecting the importance of

accounting for regional concentrations of high and low FSP participation.

Tests for statistical significance between the corresponding coefficients for the South and

non-South models (see Clogg, Petkova, and Haritou 1995; Paternoster et al. 1998) indicate a

number of significant regional differences in FSP participation. In the South, percent

unemployed, percent non-working age, percent less than high school, and black/white

segregation all exert significantly greater upward pressure on FSP participation than is true in the

non-South, while in small towns (micropolitan areas) in the South FSP participation is

significantly lower than in the non-South.

In Table 4 we turn our attention to FSP participation in high poverty areas. Model 1 shows coefficients for models assessing the association between regional dummy variables and

14

county-level FSP participation. Each of the four coefficients for the high poverty areas are

positive and significant compared to the rest of the U.S., reflecting higher than average FSP

participation in these regions. Again, the spatial lag term is positive and significant.

<Table 4 about here>

In Model 2 we assess whether these regional differences in FSP participation hold net of

other important predictors. The results reveal that Appalachia and the Borderland continue to

show significantly higher FSP participation even with the full range of predictors in the model.

In contrast, net of other factors the Delta shows no significant difference from other regions of

the country, while the Black Belt actually shows significantly lower FSP participation once all of

the predictors are considered. Again, the spatial lag remains positive and significant in the

expanded model.

Finally, we focus on understanding differences in the effects of the predictors of FSP

participation in each of the four high poverty areas. Table 5 shows coefficients from a single

regression model that examines interaction terms between each high poverty area dummy

variable and the full range of predictors. It is important to note that while the effects for each

high poverty area are displayed in separate columns in Table 5 to facilitate the presentation, the

coefficients are the product of a single model that includes all main effects, interactions, and

controls. The results show notable differences in the structural determinants of FSP participation

in each high poverty area. In the Borderland, percent female-headed households and percent

non-working age show particularly strong positive relationships with FSP participation, while

income inequality shows a particularly strong negative relationship. In Appalachia, percent

female-headed households also shows a particularly strong positive effect, while percent black is

associated with a particularly strong negative effect on FSP participation. In the Delta, the 15

results show percent non-working age holds a particularly strong positive relationship with FSP

participation, while median household income and percent less than high school show

particularly strong negative associations. Last, in the Black Belt percent non-working age and

black/white segregation show particularly strong positive relationships with FSP participation,

while percent female-headed households and poor/non-poor segregation are associated with

particularly strong negative effects. Again, even considering the effects of all of the variables

included in this model, the spatial lag continues to show a significant positive effect.

<Insert Table 5 about here>

Discussion and Implications Researchers are increasingly recognizing geographic space as a key dimension of social

inequality (Gans 2002; Gieryn 2000; Lobao 2004; Lobao and Saenz 2002; Lobao, Hooks, and

Tickamyer 2007; Massey 1996; Tickamyer 2000). In particular, scholars concerned with spatial

inequality have called for special attention to issues of comparative advantage (and

disadvantage) across space as well as to the consideration of the subnational scale (Lobao 2004;

Lobao and Hooks 2007; Tickamyer 2000). In addition, research has clearly demonstrated

enduring relationships between geographic location and economic hardship (Lichter and Johnson

2007; Lyson and Falk 1993; Massey and Denton 1993; Rural Sociological Society Task Force on

Persistent Rural Poverty 1993; Slack et al. 2009; Tickamyer and Duncan 1990; Wilson 1980). A

prime example of this is the high and persistent poverty that characterizes the South in

comparison to other regions of the United States, and areas such as the Appalachia, the Black

Belt, the Borderland, and the Delta, in particular. 16

In this study we speak to these issues by examining the spatial contours and structural

determinants of FSP participation at the county level. In short, we show that geographic space

and place matter. The U.S. South and high poverty in areas in the South, in particular, are home

to especially high regional concentrations of FSP participation. We show that even when

considering a full range of additional predictors suggested by the literature, the regional

clustering of high and low FSP participation rates continues to be a significant consideration.

Further, our results show that the structural determinants of FSP participation differ between the

South and non-South, and between high poverty areas and the rest of the country, in significant

ways.

This research contributes to the scientific literature on food assistance by: 1) bringing

attention to the importance of local space and place by building on the limited number of FSP

studies that have focused on a scale between that of individuals/households and states; and 2)

utilizing unique data from the USDA Food and Nutrition Service (FNS) that have yet to be

assessed in the research literature.

The consideration of geographic space and place also has important implications for

public policy. Indeed, this is underscored by the continued support of the Southern Rural

Development Center (SRDC), in partnership with the Economic Research Service (ERS) of the

USDA, for food and nutrition-related research focused on the U.S. South. The underlying

assumption of this programmatic focus is that the South faces unique challenges in terms of food

assistance and hunger and that research is needed to advance evidence-based public policy to

address these issues. Accordingly, this study addresses four of the seven 2008 RIDGE Program

research themes/topics, namely: High Poverty Areas; Impact of Immigrants on Nutrition

Programs; Use of Food/Nutrition Programs; and Policy Choices. All of these themes are of 17

importance to policy makers who seek to better understand the determinants of FSP use and how

to better serve vulnerable populations in the South, and beyond. It is our hope that the

information presented in this report will prove useful to policy makers and practitioners by

helping to identify communities that are likely to have special food assistance needs and by

helping to anticipate how local social change may influence such needs in the future. 18

References

Bhattarai, G.R., P.A. Duffy, and J. Raymond. 2005. “Use of Food Pantries and Food Stamps in Low-Income Households in the United States.” Journal of Consumer Affairs 39: 276- 298.

Clogg, C.C., E. Petkova, and A. Haritou. 1995. “Statistical Methods for Comparing Regression Coefficients between Models.” American Journal of Sociology 100:1261-1293. Crowley, M., D.T. Lichter, and Z. Qian. 2006. “Beyond Gateway Cities: Economic

Restructuring and Changing Poverty Among Mexican Immigrants.” Family Relations 55: 345-360.

Davila, A., and M.T. Mora. 2000. “English Skills, Earnings, and the Occupational Sorting of

Mexican Americans along the U.S.-Mexico Border.” International Migration Review 34: 133-157.

Davila, A., A.K. Bohara, and R. Saenz. 1993. “Accent Penalties and the Earnings of Mexican Americans.” Social Science Quarterly 74: 902-916. Economic Research Services. 2007. “Measuring Rurality: What is Rural?” United States

Department of Agriculture. Retrieved on June 8, 2008, from http://www.ers.usda.gov/ Briefing/Rurality/WhatisRural/.

Figlio, D.N., C. Gundersen, and J.P. Ziliak. 2000. “The Effects of the Macroeconomy and Welfare Reform on Food Stamp Caseloads.” American Journal of Agricultural Economics 82: 635-641. Food and Nutrition Service. 2009. “Food Stamp Program.” United States Department of Agriculture. Retrieved on June 8, 2008, from http://www.fns.usda.gov/FSP/.

Frey, W.H., and K..L Liaw. 2005. “Migration within the United States: Role of Race-Ethnicity.” Brookings-Wharton Papers on Urban Affairs 6: 207-262. Friedman, S., and D.T. Lichter. 1998. “Spatial Inequality and Poverty Among American Children.” Population Research and Policy Review 17: 91-109. Frongillo, E.A., D.F. Jyoti, and S.J. Jones. 2006. “Food Stamp Program Participation Is Associated with Better Academic Learning among School Children.” Journal of Nutrition 136: 1077-1080.

Gans, H.J. 2002. “The Sociology of Space: A Use-Centered View.” City and Community 1: 329- 339. Gibson, D. 2004. “Long-Term Food Stamp Program Participation is Differentially Related to Overweight in Young Girls and Boys.” Journal of Nutrition 134: 372-379.

19

Gibson, D. 2006. “Long-Term Food Stamp Program Participation Is Positively Related to Simultaneous Overweight in Young Daughters and Obesity in Mothers.” Journal of Nutrition 136: 1081-1085. Gieryn, T.F. 2000. “A Space for Place in Sociology.” Annual Review of Sociology 26: 463-496.

Goetz, S.J., A. Rupasingha, J.N. Zimmerman. 2004. “Spatial Food Stamp Participation Dynamics in U.S. Counties.” The Review of Regional Studies 34: 172-190.

Gundersen, C., and V. Oliveira. 2001. “The Food Stamp Program and Food Insufficiency." American Journal of Agricultural Economics 83: 875-887.

Hoyt, W.H., and F.A. Scott. 2004. “Geographic Variation in Food Stamp and Other Assistance Program Participation Rates: Identifying Poverty Pockets in the South.” Final report submitted to the Southern Regional Development Center, Mississippi State University, Mississippi State, MS.

Jensen, L., and D.J. Eggebeen. 1994. Nonmetropolitan poor children and reliance on public assistance. Rural Sociology 59: 45-65. Jensen, L., D.K. McLaughlin, and T. Slack. 2003. “Rural Poverty: The Persisting Challenge.” Pp. 118-131 in D.L. Brown and L.E. Swanson (eds.), Challenges for Rural America in the Twenty-First Century. University Park, PA: The Pennsylvania State University Press. Jones, A.F., and D.F. Weinberg. 2000. “The Changing Shape of the Nation’s Income Distribution.” Current Population Reports. U.S. Census Bureau.

Lichter, D.T., and K.M. Johnson. 2007. “The Changing Spatial Concentration of America’s Rural Poor Population.” Rural Sociology 72: 331-358.

Lichter, D.T., and D.K. McLaughlin. 1995. “Changing Economic Opportunities, Family Structure, and Poverty in Rural Areas.” Rural Sociology 60:688-706. Lichter, D.T., D. Parisi, M.C. Taquino, and B. Beaulieu. 2008. “Race and the Micro-Scale

Concentration of Poverty.” Cambridge Journal of Regions, Economy and Society 1: 51- 67.

Lobao, L.M. 2004. “Continuity and Change in Place Stratification: Spatial Inequality and Middle-Range Territorial Units.” Rural Sociology 69: 1-30.

Lobao, L.M., and G. Hooks. 2003. “Public Employment, Welfare Transfers, and Economic Well-Being across Local Populations: Does a Lean and Mean Government Benefit the Masses?” Social Forces 82:519-556.

Lobao, L.M., and R. Saenz. 2002. “Spatial Inequality and Diversity as an Emerging Research Area.” Rural Sociology 67: 497-511.

20

Lobao, L.M. and G. Hooks. 2007. “Advancing the Sociology of Spatial Inequality: Space, Places, and the Subnational Scale.” Pp. 29-61 in The Sociology of Spatial Inequality, edited by L.M. Lobao, G. Hooks, and A.R. Tickamyer. Albany, New York: State University of New York Press. Lobao, L.M., G. Hooks, and A.R. Tickamyer. 2007. The Sociology of Spatial Inequality. Albany, NY: State University of New York Press. Lyson, T.A., and W.W. Falk. 1993. Forgotten Places: Uneven Development in Rural America. Lawrence, KS: University Press of Kansas.

Massey, D.S. 1996. “The Age of Extremes: Concentrated Affluence and Poverty in the Twenty- First Century.” Demography 33: 395-412.

Massey, D.S, and N.A. Denton. 1993. American Apartheid: Segregation and the Making of the Underclass. Cambridge, MA: Harvard University Press.

McLaughlin, D.K., E. Gardner, and D.T. Lichter. 1999. “Economic Restructuring and the Changing Prevalence of Female-Headed Families in America.” Rural Sociology 64: 394- 416.

Nord, M. 2001. “Food Stamp Participation and Food Stamp Security.” Food Review 24: 13-19. Paternoster, R., R. Brame, P. Mazerolle, A. Piquero. 1998. “Using the Correct Statistical Test for the Equality of Regression Coefficients.” Criminology 36:859-866.

Racial Residential Segregation Project. 2008. “Residential Segregation: What It Is and How We Measure It.” Retrieved on June 8, 2008, from http://enceladus.isr.umich.edu/ race/seg.html.

Rupasingha, A., and S.J. Goetz. 2007. “Social and Political Forces as Determinants of Poverty: A Spatial Analysis.” The Journal of Socio-Economics 36: 650-671.

Rural Sociological Society Task Force on Persistent Rural Poverty. 1993. Persistent Poverty in Rural America. Boulder, CO: Westview Press.

Saenz, R. 1997. “Ethnic Concentration and Chicano Poverty: A Comparative Approach.” Social Science Research 26:205-28. Slack, T., J. Singelmann, K. Fontenot, D.L. Poston, Jr., R. Saenz, and C. Siordia. 2009. “Poverty in the Texas Borderland and Lower Mississippi Delta: A Comparative Analysis of Differences by Family Type.” Demographic Research 20: 353-376. Tapogna, J., A. Suter, M. Nord, M. Leachman. 2004. “Explaining Variations in State Hunger Rates.” Family Economics & Nutrition Review 16: 12-21.

21

Tickamyer, A.R. 2000. “Space Matters! Spatial Inequality in Future Sociology.” Contemporary Sociology 29: 805-813. Tickamyer, A.R., and C.M. Duncan. 1990. “Poverty and Opportunity Structure in Rural America.” Annual Review of Sociology 16: 67-86.

Van Hook, J., and K.S. Balistreri. 2006. “Ineligible Parents, Eligible Children: Food Stamps Receipt, Allotments, and Food Insecurity among Children of Immigrants.” Social Science Research 35: 228-251.

Voss, P.R., D.D. Long, R.B. Hammer, and S. Friedman. 2006. “County Child Poverty Rates in the U.S.: A Spatial Regression Approach.” Population Research and Development Review 25: 369-391. Wallace, G., and R.M. Blank. 1999. “What Goes Up Must Come Down? Explaining Recent

Changes in Public Assistance Case Loads.” Pp. 49-89 in S.H. Danziger (ed.), Economic Conditions and Welfare Reform. Kalamazoo, MI: W.E. Upjohn Institute for Employment Research.

Wilde, P., and C. Dicken. 2003. “Using County-Level Data to Study the Race and Ethnicity of Food Stamp Program Participants in California.” Paper prepared for the American Agricultural Economics Association Meeting, Montreal, Canada. Wilde, P., P. Cook, C. Gunderson, M. Nord, and L. Tiehen. 2000. “The Decline in Food Stamp Program Participation in the 1990’s.” Food Assistance and Nutrition Report No. 7. Washington, D.C.: USDA.

Wilson, W.J. 1980. The Declining Significance of Race: Blacks and Changing American Institutions, 2nd Edition. Chicago: University of Chicago Press.

Wimberley, R.C., and L.V. Morris. 1997. The Southern Black Belt: A National Perspective. The University of Kentucky: TVA Rural Studies.

22

Table 1. Descriptive statistics Variables Mean Minimum Maximum Dependent variable FSP participation 10.80 0.00 50.70 Independent variables Labor market conditions

Percent unemployed 3.35 0.00 17.30 Percent secondary/peripheral jobs 31.01 15.05 67.11 Median household income 35,086 15,805 82,929

Population structure Percent female-headed households 6.24 0.00 24.38 Net migration 5.70 -41.10 133.80 Percent black 10.02 0.00 86.07 Percent non-working age 35.63 17.91 49.29 Percent foreign-born¹ 3.30 0.00 50.90 Percent Latino¹ 6.52 0.00 98.10

Human capital Percent less than h.s. 23.73 3.04 65.29 Percent not English proficient¹ 1.64 0.00 28.60

Residential context and inequality Metropolitan 0.35 0.00 1.00 Micropolitan 0.22 0.00 1.00 Non-core 0.43 0.00 1.00 Gini 42.53 32.80 56.30 Housing stress 0.15 0.00 1.00 Index of dissimilarity: poor vs. non-poor 24.29 0.00 54.90 Index of dissimilarity: black vs. white 49.55 0.00 96.40 Index of dissimilarity: Latino vs. white 32.22 0.00 73.70

Expenditures on cash assistance Per capita expenditures on TANF and SSI² 174.97 0.00 3,920.32

N=2,561 ¹ Original distributions presented, but variables are included in a factor analysis score for analysis. ² Original distribution presented, but variable was transformed to its natural log for analysis.

23

Table 2. County-level FSP participation by region, 2004-2006 Region N Mean Minimum Maximum Nation 2,561 10.8 0.0 50.7 South 1,330 13.2 0.0 45.9 Non-South 1,231 8.2 0.3 50.7 Appalachia 78 21.9 10.6 45.9 Black Belt 494 13.4 2.0 39.3 Borderland 41 19.0 2.3 35.9 Delta 133 18.5 4.6 40.1

24

Table 3. Unstandardized OLS regression coefficients from region-specific models of county-level FSP participation, 2004-2006 Independent variables South Non-South Labor market conditions

Percent unemployed 0.724***† 0.298** Percent secondary/peripheral jobs -0.030 0.009 Median household income -0.002 -0.003

Population structure Percent female-headed households 0.789*** 0.991*** Net migration 0.007 0.017 Percent black -0.024* 0.023 Percent non-working age 0.270***† 0.165*** Latino/immigrant/English -0.330** -0.357*

Human capital Percent less than h.s. 0.229***† 0.080***

Residential context and inequality Metro (ref.) Micro -0.514*† 0.246 Non-core -0.189 0.204 Gini 0.076 0.169*** Housing stress (1=yes) 1.311*** 0.781 Poor/non-poor D -0.009 -0.023 Black/white D 0.024***† 0.004 Latino/white D -0.004 0.002

Expenditures on cash assistance Ln per capita expenditures on TANF/SSI 1.719*** 1.336***

Spatial lag 0.293*** 0.433*** Intercept -25.508*** -25.592*** R-squared 0.837 0.743 N 1,330 1,231 Notes: Models include controls for state fixed effects. Median household income coefficient multiplied by 100. * p<.05; ** p<.01; *** p<.001; † Different from non-South model at p < .05.

25

Table 4. Unstandardized OLS regression coefficients of county-level FSP participation in high poverty regions, 2004-2006 Independent variables Model 1 Model 2 Borderland 2.655*** 1.155* Appalachia 2.282*** 2.194*** Delta 2.309*** -0.543 Black Belt 1.701*** -0.962*** Labor market conditions

Percent unemployed 0.508*** Percent secondary/peripheral jobs -0.021 Median household income -0.002

Population structure Percent female-headed households 0.843*** Net migration 0.010 Percent black -0.003 Percent non-working age 0.198*** Latino/immigrant/English -0.343**

Human capital Percent less than h.s. 0.162***

Residential context and inequality Metro (ref.) Micro -0.100 Non-core -0.013 Gini 0.131*** Housing stress (1=yes) 1.109*** Poor/non-poor D -0.017 Black/white D 0.010* Latino/white D -0.006

Expenditures on cash assistance Ln per capita expenditures on TANF/SSI 1.648***

Spatial lag 0.850*** 0.310*** Intercept 0.723 -27.356*** R-squared 0.602 0.823 N 2,561 2,561 Notes: Models include controls for state fixed effects. Median household income coefficient multiplied by 100. * p<.05; ** p<.01; *** p<.001.

26

Table 5. Unstandardized OLS regression regional interaction coefficients from models of county-level FSP participation in high poverty regions, 2004-2006 Independent variables Borderland Appalachia Delta Black Belt Labor market conditions

Percent unemployed 0.175 0.229 0.063 0.048 Percent secondary/peripheral jobs -0.130 -0.294 0.068 0.005 Median household income -0.028 -0.045 -0.025* -0.005

Population structure Percent female-headed households 1.134* 0.702* -0.077 -0.350** Net migration 0.026 0.080 -0.016 -0.024 Percent black -0.419 -0.686* -0.004 0.039 Percent non-working age 0.670** -0.160 0.296* 0.112* Latino/immigrant/English -0.525 -2.866 -0.810 0.139

Human capital Percent less than h.s. 0.106 0.028 -0.122* -0.016

Residential context and inequality Metro (ref.) Micro 1.151 -2.008 -1.365 -0.499 Non-core 0.602 -0.847 -0.848 -0.781 Gini -0.966*** -0.097 -0.188 -0.039 Housing stress (1=yes) -0.162 0.162 0.362 0.015 Poor/non-poor D -0.002 0.171 0.029 -0.055* Black/white D -0.064 0.016 -0.002 0.033* Latino/white D -0.018 0.048 0.011 -0.004

Expenditures on cash assistance Ln per capita expenditures on TANF/SSI -0.564 1.542 0.636 -0.213

Spatial lag 0.276*** Intercept -26.026 R-squared 0.840 N 2,561 Notes: Cell entries are coefficients produced by a single regression model including all main effects, all regional interactions, the spatial lag, and state fixed effects. Median household income coefficient multiplied by 100. * p<.05; ** p<.01; *** p<.001.

27

No county-level data

County-level data

Figure 1. County-level FSP data coverage 28

N

W E

No data S

Low participation (< 1 S.D.)

Average participation

High participation (> 1 S.D.)

Figure 2. County-level FSP participation, 2004-2006

29

Moran’s I = 0.6093 Figure 3. Moran scatterplot of county-level FSP participation, 2004-2006

30

Figure 4. LISA cluster map of county-level FSP participation, 2004-2006

31

About the Authors

Tim Slack, Principal Investigator: Tim Slack holds a Ph.D. in rural sociology from Penn State

University and is currently an Assistant Professor of Sociology at Louisiana State University.

Dr. Slack’s research interests are in the areas of poverty and inequality, rural sociology, social

demography, and work and labor markets. For further information visit www.soc.lsu.edu/slack.

Candice A. Myers, Graduate Research Assistant: Candice A. Myers holds a M.A. in

sociology from Louisiana State University. She is currently pursuing her Ph.D. in sociology at LSU

and plans to examine poverty and hardship-related issues for her dissertation research. For further

information visit www.soc.lsu.edu/myers. 32