20
1 Disparities in accessibility to food stores Disparities in accessibility to healthy food have received growing attention as a public health concern. A large body of literature has shown that poor access to reasonably priced, nutritious, and good-quality food sources (eg grocery stores and restaurants) may lead to a poor diet such as low consumption of fruits and vegetables and large consumption of sugary or high-fat foods (Helling and Sawicki, 2003; Larson et al, 2009; Pearce et al, 2006; Wang et al, 2007). Poor diet quality may increase the risks of health problems such as obesity, diabetes, and cardiovascular diseases (Darmon et al, 2002; Edelstein et al, 1997; Epstein et al, 2001; Kyle and Blair, 2007; Larsen and Gilliand, 2008; Morland et al, 2006; Must et al, 1991; Zenk et al, 2005). Previous studies (Larson et al, 2009; Morland et al, 2006; Wang et al, 2007) reveal that better access to stores providing fresh, nutritious, and good-quality food is linked to a lower risk of obesity. Despite efforts made to improve access to healthy food, inequalities in food access persist and remain a major public health issue (Algert et al, 2006; Helling and Sawicki, 2003; Moore et al, 2008b). Food access refers to one’s ability to obtain the services of food providers. Food access varies across space. Uneven geographic distributions of providers and consum- ers lead to inequality in spatial accessibility (Wang and Luo, 2005). Food access also varies in nonspatial dimensions. Population groups differ widely in demographic and socioeconomic characteristics (eg income, age, sex, and ethnicity). Nonspatial factors also interact with spatial access. For example, residents who are transit-dependent are largely a nonspatial issue because of their age, medical condition, or lack of economic Geographic disparities in accessibility to food stores in southwest Mississippi Dajun Dai Department of Geosciences, Georgia State University, Atlanta, GA 30303, USA and Partnership for Urban Health Research, Institute of Public Health, Georgia State University, Atlanta, GA 30302, USA; e-mail: [email protected] Fahui Wang Department of Geography and Anthropology, Louisiana State University, Baton Rouge, LA 70803, USA and School of Urban Management, Resources and Environment, Yunnan University of Finance and Economics, Kunming, Yunnan 650221, China; e-mail: [email protected] Received 4 November 2009; in revised form 10 September 2010 Environment and Planning B: Planning and Design 2011, volume 38, pages 659 ^ 677 Abstract. Disparities in accessibility to healthy food are a critical public-health concern. Poor access to reasonably priced, nutritious, and good-quality food may lead to poor diet and increase the risks of health problems such as obesity, diabetes, and cardiovascular diseases. This research advances the popular two-step floating catchment area (2SFCA) method by incorporating a kernel density (KD) function to form the ‘KD2SFCA method’. The study applies the method to measure the spatial access to food stores in southwest Mississippi, and examines the interaction between the spatial access and nonspatial factors. The research shows that neighborhoods with higher scores of urban socioeconomic disadvantage actually have better spatial accessibility to food stores; but higher percentages of carless households and lower income in some neighborhoods may compromise overall accessibility. Neighbor- hoods with stronger cultural barriers tend to be associated with poorer spatial accessibility. The study clearly differentiates spatial and nonspatial factors in access inequalities, and thus helps policy makers to design corresponding remedial strategies. doi:10.1068/b36149

Geographic disparities in accessibility to food stores in southwest Mississippi

  • Upload
    fahui

  • View
    215

  • Download
    1

Embed Size (px)

Citation preview

Page 1: Geographic disparities in accessibility to food stores in southwest Mississippi

1 Disparities in accessibility to food storesDisparities in accessibility to healthy food have received growing attention as a publichealth concern. A large body of literature has shown that poor access to reasonablypriced, nutritious, and good-quality food sources (eg grocery stores and restaurants)may lead to a poor diet such as low consumption of fruits and vegetables and largeconsumption of sugary or high-fat foods (Helling and Sawicki, 2003; Larson et al,2009; Pearce et al, 2006; Wang et al, 2007). Poor diet quality may increase the risksof health problems such as obesity, diabetes, and cardiovascular diseases (Darmonet al, 2002; Edelstein et al, 1997; Epstein et al, 2001; Kyle and Blair, 2007; Larsenand Gilliand, 2008; Morland et al, 2006; Must et al, 1991; Zenk et al, 2005). Previousstudies (Larson et al, 2009; Morland et al, 2006; Wang et al, 2007) reveal that betteraccess to stores providing fresh, nutritious, and good-quality food is linked to a lowerrisk of obesity. Despite efforts made to improve access to healthy food, inequalities infood access persist and remain a major public health issue (Algert et al, 2006; Hellingand Sawicki, 2003; Moore et al, 2008b).

Food access refers to one's ability to obtain the services of food providers. Foodaccess varies across space. Uneven geographic distributions of providers and consum-ers lead to inequality in spatial accessibility (Wang and Luo, 2005). Food access alsovaries in nonspatial dimensions. Population groups differ widely in demographic andsocioeconomic characteristics (eg income, age, sex, and ethnicity). Nonspatial factorsalso interact with spatial access. For example, residents who are transit-dependent arelargely a nonspatial issue because of their age, medical condition, or lack of economic

Geographic disparities in accessibility to food storesin southwest Mississippi

Dajun DaiDepartment of Geosciences, Georgia State University, Atlanta, GA 30303, USA and Partnershipfor Urban Health Research, Institute of Public Health, Georgia State University, Atlanta, GA30302, USA; e-mail: [email protected]

Fahui WangDepartment of Geography and Anthropology, Louisiana State University, Baton Rouge,LA 70803, USA and School of Urban Management, Resources and Environment, YunnanUniversity of Finance and Economics, Kunming, Yunnan 650221, China; e-mail: [email protected] 4 November 2009; in revised form 10 September 2010

Environment and Planning B: Planning and Design 2011, volume 38, pages 659 ^ 677

Abstract. Disparities in accessibility to healthy food are a critical public-health concern. Poor accessto reasonably priced, nutritious, and good-quality food may lead to poor diet and increase the risks ofhealth problems such as obesity, diabetes, and cardiovascular diseases. This research advances thepopular two-step floating catchment area (2SFCA) method by incorporating a kernel density (KD)function to form the `KD2SFCA method'. The study applies the method to measure the spatial accessto food stores in southwest Mississippi, and examines the interaction between the spatial access andnonspatial factors. The research shows that neighborhoods with higher scores of urban socioeconomicdisadvantage actually have better spatial accessibility to food stores; but higher percentages of carlesshouseholds and lower income in some neighborhoods may compromise overall accessibility. Neighbor-hoods with stronger cultural barriers tend to be associated with poorer spatial accessibility. The studyclearly differentiates spatial and nonspatial factors in access inequalities, and thus helps policy makersto design corresponding remedial strategies.

doi:10.1068/b36149

Page 2: Geographic disparities in accessibility to food stores in southwest Mississippi

means, but they also tend to have to travel for a longer time to access food sources,which affects their spatial access. Understanding food access requires consideration ofboth spatial and nonspatial factors and their interactions. Previous studies (Hellingand Sawicki, 2003; Pearce et al, 2006; Zenk et al, 2005) have recognized that all thesefactors are important, but their roles differ among places and population groups.Evaluating disparities of food access begins with the development of sound methodsof accessibility measures.

In section 2 the methods for measuring accessibility and identifying areas for improve-ment are reviewed. The study area and data used in the analysis are described in section 3.In section 4 we present the proposed spatial accessibility measure and the method isapplied to the study area. In section 5 we evaluate the socioeconomic disparities in spatialaccessibility. We conclude with a brief summary and some discussion of directions forfuture work.

2 Measures of accessibilityVarious measures have been proposed to analyze spatial accessibility to food resources.One measure focuses on resource availabilityöthe ratio of the number of food providersto population within the same geographic unit. The unit is usually an administrative unit,such as a county. This measure is recommended for geographic analysis of a food envi-ronment by the Risk Factor Monitoring and Methods, National Cancer Institute (NCI)(https://riskfactor.cancer.gov/mfe/defining-measures-instruments-and-methodologies), and hasbeen used in many existing studies (Austin et al, 2005; Bingham and Zhang, 1997; Hellingand Sawicki, 2003; Immergluck, 1999; Moore et al, 2008b; Raja et al, 2008). Similarapproaches have been used in other studies, such as retail grocery square feet per house-hold in postal sectors in Leeds and Bradford, UK (Clarke et al, 2002) or the number ofsupermarkets within one kilometer around census tracts in Montreal, Canada (Apparicioet al, 2007).While the method is straightforward and reflects the balance between popula-tion and food resources, it has several limitations. First, it does not reveal the spatialvariation of food resources within the analysis unit. Second, it does not differentiate thestores in size or capacity. Third, and most importantly, it does not account for people'smobilityöspatial interaction between food providers and customersöacross residentialboundaries. The spatial interaction has two aspects: supply competition between foodproviders for a customer and demand competition between customers for a food provider.Spatial interaction is of more importance in rural areas where limited numbers of foodproviders force people to travel far for shopping (Burns et al, 2004; Sharkey and Horel,2008). When the choice of supermarkets in a neighborhood is limited, residents travel toneighborhoods nearby with a high density of food resources (Moore et al, 2008b). Luo andWang (2003) pointed out that within-area variations depend on the scale or geographicunit. The larger the unit of analysis is, the less problematic the spatial interaction, but themore problematic the internal variation, and vice versa. The challenge is to reconcilethe two.

Several methods have been developed to address the issue. For example, a fewstudies (Pearce et al, 2006; Sharkey and Horel, 2008; Smoyer-Tomic et al, 2006; Zenket al, 2005) considered people's mobility across boundaries by using the nearest neigh-bor method to compute the distance (eg the Euclidean distance, Manhattan distance,or travel distance through the road network) or travel time from a residential locationto the closest food provider (eg Apparicio et al, 2007). However, people may shop in anumber of stores. Some researchers (eg Wang et al, 2007) used the food-to-populationratio in a buffer around each residential location. The fallacy remains since the ratiocannot reveal the spatial variability of accessibility within the buffer. The gravity-basedindex (Clarke et al, 2002; Guy, 1997; 1983; Helling and Sawicki, 2003) accounts for

660 D Dai, F Wang

Page 3: Geographic disparities in accessibility to food stores in southwest Mississippi

spatial variability of resources and people's mobility, but it does not account forcompetition for food providers by residents. The two-step floating catchment area(2SFCA) method, commonly used in health care accessibility studies (eg Cervigniet al, 2008; Guagliardo, 2004; Langford and Higgs, 2006; Luo and Wang, 2003;Wang and Luo, 2005) has great potential for use in food-access evaluation. Thismethod addresses the aforementioned three limitations and outperforms other meth-ods such as the kernel density (KD) estimation (Yang et al, 2006). However, access isassigned equally within a catchment. A recent study (Luo and Qi, 2009) divided acatchment area into multiple travel-time zones and assigned each zone a weight. Yet itstill assumed equal accessibility within each zone.

In addition to the challenges to improve the measure of spatial accessibility to foodresources, the interaction between spatial access and nonspatial factors warrantsfurther investigation. Nonspatial factors in the literature include race, income, educa-tional attainment, and unemployment rate (Algert et al, 2006; Donkin et al, 1999; Guyand David, 2004; Helling and Sawicki, 2003; Larsen and Gilliand, 2008; Raja et al,2008). Poor access to food sources has been found in concentrated areas of minoritypopulations and economically disadvantaged neighborhoods (Algert et al, 2006; Guyand David, 2004; Larsen and Gilliand, 2008; Raja et al, 2008). Others disagree. InAtlanta, GA, if driving trips longer than 5 minutes were considered acceptable, theAfrican American neighborhoods were found to have similar or better access to chainsupermarkets when compared with access for the predominantly white neighborhoods(Helling and Sawicki, 2003). Other researchers (Apparicio et al, 2007; Donkin et al,1999) also reported that socially deprived or minority neighborhoods in inner citieshave short distances to food stores. One issue is that these nonspatial factors are oftenhighly correlated. For example, neighborhoods with low educational attainment tendto have high unemployment rates and low income levels. There is a need to reveal thetrue dimensions of nonspatial factors and to evaluate the interactions between spatialaccessibility and these nonspatial factors.

This study has dual objectives: (1) to propose a new measure of spatial accessibilityaccounting for spatial interactions between supplies and demands and distance decayof attraction within a catchment range, and (2) to examine the disparities of spatialaccess and its relationship with nonspatial factors through a case study. The accessi-bility measure is termed `KD2SFCA' as it improves the existing 2SFCA method byadding a KD function to capture variation within a catchment area. The method isimplemented in a GIS environment and is applied to studying accessibility to foodstores in southwest Mississippi. According to the Centers for Disease Control andPrevention (2006), Mississippi has the highest obesity rate in the United States inrecent years, which makes it a suitable area for this study.

The contributions of this study may be summarized as follows:1. in the methodological realm, it proposes an advanced measure to quantify spatialaccessibility, which is an important dimension of food-access evaluation;2. it employs factor analysis to consolidate nonspatial factors, and examines thedisparities in spatial accessibility among various population groups captured by thesenonspatial factors;3. it has important implications for public policy, clearly differentiating spatial andnonspatial factors in access inequalities; thus it has the potential to help policymakers identify neighborhoods that are short of food access leading to the designof corresponding remedial strategies.

Geographic disparities in accessibility to food stores in southwest Mississippi 661

Page 4: Geographic disparities in accessibility to food stores in southwest Mississippi

3 Study area and dataThe study area includes thirteen counties in southwest Mississippi with a total popula-tion of 517 991 in 2000 (figure 1). There are 121 census tracts in the study area. City ofJackson, the state capital and the largest city in Mississippi, is located northeast of thestudy area with a population of 217304 in 2000.

According to the Reference USA database (http://www.referenceusa.com/), the studyarea has 907 food stores, excluding restaurants, school or workplace cafeterias, andother food providers (eg farmer's markets), different from the `food-store environment'defined by the NCI (http://riskfactor.cancer.gov/mfe/categorizing-the-food-environment).The Reference USA database reports individual businesses from a telephone surveyconducted by the vendor. Each record includes the business name, the primaryStandard Industrial Classification (SIC) code, its employment size, and its locationby longitude and latitude. Using the standard four-digit SIC system, the food storesare classified into the following store categories: supermarket, grocery, convenience,meat and fish, fruit and vegetable, candy and nut, dairy, bakery, natural food, andspecialty.

To differentiate the capacities between stores, a weight scaling from 1 ^ 10 isassigned to each store on the basis of its name recognition, employment size, andprimary SIC-code description (see table 1). Drawn from earlier studies (Moore et al,2008a; 2008b; Raja et al, 2008), supermarkets and other grocery stores with SIC codes

0 5 10 20

miles

Figure 1. Study area.

662 D Dai, F Wang

Page 5: Geographic disparities in accessibility to food stores in southwest Mississippi

of 541101 and 541104 ^ 541106 and more than 50 employees receive higher scores thanother stores. Small fruit and vegetable stores receive lower weights than regular grocerystores because of their smaller employment sizes on average (2 versus 9). Thisunidimensional-weighting scheme is adopted to differentiate between various types ofstores in terms of their employment sizes since detailed information such as food itemsand their freshness and quality are not available in the database. It reflects concep-tually the differences in their capacities to carry healthy food items. Studies by Ashmanet al (1993) and Johnson et al (1996) report that wilted, damaged, or spoiled produceis common in small stores while the quality of produce in medium and large stores isgenerally high. Consequently, neighborhoods near a large supermarket providing avariety of food items have higher accessibility than those near a gas station. In future,surveys of the range and freshness of food items in stores will validate and improvethis weighting scheme.

Census data at the census block and tract levels are from the 2000 CensusSummary File 3 (SF3) (http://www.census.gov/census2000/sumfile3.html). Our primaryanalysis unit is the census tract since it is the smallest area unit with all the requiredsocioeconomic data. Census tracts are commonly used in similar studies as a proxyfor neighborhoods (eg Helling and Sawicki, 2003; Larsen and Gilliand, 2008). Theroad network database was obtained from the Environment Systems Research Institute(ESRI) (http://www.esri.com) data CDs. The road network includes all levels of roadsand streets with information such as speed limits.

Several technical issues in data preparation warrant some discussion. Population-weighted centroids based on block-level census data were calibrated to representcensus-tract locations. Population-weighted centroids are more accurate than simplegeographic centroids, especially in rural areas where census tracts are large and thepopulation usually clusters in limited spaces (Hwang and Rollow, 2000; Wang and Luo,2005). The shortest travel times through road networks between population centroidsand food stores were computed using the Network Analyst Tool in ArcGIS 9.3 (ESRI).It is assumed that drivers abide by speed limits. Possible edge effect was examined byexpanding the study area to include census tracts with their centroids within a 15-milebuffer zone around the study area. We found that these areas are mostly rural with nocities or towns across nearby borders. That is to say, residents mostly rely on storeswithin the study area for food purchase, and the edge effect is minimal.

Table 1. Food-store category and weighting scores.

Food-store category Number Employment Weight Examplesof stores size

Supermarkets 31 >50 10 Kroger, Walmart,Sam's Club, NatchezMarket

Groceries 133 <50 3 Center Grocery,Dexter Grocery,H&L Grocery

Convenience, meat and 457 <50 2 Dollar General,fish, candy and nut, dairy, Dirt Cheap,bakery, fruit and vegetable, Fred's Storenatural food, and specialtyGas stations 286 <50 1 Kangaroo Express,

Quickstop, Texaco

Geographic disparities in accessibility to food stores in southwest Mississippi 663

Page 6: Geographic disparities in accessibility to food stores in southwest Mississippi

4 Measuring spatial accessibility to food stores by the KD2SFCA methodThis study proposes the KD2SFCA by incorporating a kernel function into the 2SFCAmethod to capture variation within each catchment area. The original 2SFCA methodworks as follows.

At step 1, for each food-store location j, search all population locations (k) within athreshold travel distance or travel time (d 0 ) from j to form the catchment area forfood-store location j, and sum the population within this catchment as the customerbase for this food store at j. The ratio of food-store weight Sj to its correspondingcustomer base measures the availability of this store at j (Rj ), written as

Rj �SjX

k2fdk j

4 d0gPk

, (1)

where Pk is the population at location k whose centroid is within the catchment(ie dkj 4 d0 ) for food-store location j ; dkj is the travel distance or travel time betweenpopulation location k and food-store location j ; Sj is the weight score of the food store at j.

At step 2, for each population location i, search all food-store locations l within thethreshold distance or time d0 from i to form the catchment area for the population at i,and sum R within the catchment area at i to obtain the spatial accessibility at i, written as

Ai �X

l2fdi l4 d

0gRl �

Xl2fd

i l4 d

0g

SlXk2fd

k l4 d

0gPk

, (2)

where l denotes all store locations within the catchment of the ith population location,and all other notations are the same as in equation (1).

To account for distance decay between food providers and residents, the KD2SFCAintegrates a kernel function in each of the aforementioned steps. The bandwidth (h) in theKD function is identical to the catchment range (it also refers to the threshold traveldistance or travel time, ie d0 ) in the 2SFCA. That is to say, accessibility is discounted bytravel distance or travel time in a KD function towards the edge of the kernel, andbecomes 0 beyond the kernel. Specifically, in the first step, the KD function (f) is usedto rescale the population at each location (k) according to its travel distance or travel timefrom a food-store location ( l ), and everything else remains the same. In the second step,the KD function (f) is applied to rescale Rl according to the travel distance or travel timebetween a food-store location and a population location. The KD2SFCA is written as:

Ai �X

l2fdi l4 d

0gRl f�dil , h� �

Xl2fd

i l4 d

0g

Sl f�dil , h�Xk2fd

k l4 d

0gPk f�dkl ,h�

, (3)

where f represents the KD function and all other parameters are the same as inequations (1) and (2).

The kernel function (or kernel shape) and the bandwidth h (or the threshold thatdefines the catchment area) need to be determined prior to the analysis. Different kernelfunctions reflect various distance-decay patterns between food stores and customers.Several methods have been proposed to pick up the best kernel function (Fotheringhamet al, 2000, pages 155 ^ 157) or to optimize h (Cao et al, 1994) according to the globalstructure of the dataset. However, Epanechnikov (1969) found that the choiceamong the various kernel functions does not significantly affect the outcomes of theprocess. Studies by Levine (2009) and Williamson et al (1998) point out that the choiceof bandwidth is an important issue in any KDE (http://www.kde.org/) application. Somesuggest using an adaptive bandwidth h (ie a larger h in areas where events are sparser, and a

664 D Dai, F Wang

Page 7: Geographic disparities in accessibility to food stores in southwest Mississippi

smaller h where they are denser) (Brunsdon, 1995; Fotheringham et al, 2002). Ideally, thekernel function and bandwidth should be determined by customers' perception of travelimpedance (eg time, monetary cost, and convenience). Such a perception varies acrossages, races, and socioeconomic factors, and its determination would require extensivesurveys. For convenience, in this study we used the Epanechnikov function, which is thedefault function in ArcGIS 9.3. This research used a 30-minute threshold in the initialanalysis.We also experimented with various bandwidths ranging from 15 ^ 90 minutes with5-minute increments to investigate the impact of the bandwidths. The Epanechnikovfunction is written as follows where i and j are the same as in equations (1) and (2):

f�dij , h� �3

4

"1ÿ

�dijh

�2#, if dij 4 h;

f�dij , h� � 0, if dij 4 h:

8>><>>: (4)

Figure 2(a) presents the variation in spatial accessibility measured by the KD2SFCA.For comparison, we also replicated the work using the conventional 2SFCA method,as shown in figure 2(b). In general, results from the two methods are consistent with thehigher spatial accessibility scores observed in cities and major towns. The KD2SFCAmethod shows a clearer rural ^ urban gradient in spatial accessibility to food storesthan the conventional 2SFCA method. Taking the City of Jackson and its surroundingregion as an example, the KD2SFCA reveals that the food-store accessibility is greatestin the central city, decreases in the suburbs, and becomes poor in the surroundingrural areas. In contrast, the 2SFCA method suggests better access in the rural areas

Spatial accessibility

0.84 ^ 2.55

2.56 ^ 3.51

3.52 ^ 4.21

4.22 ^ 5.34

5.35 ^ 6.55

0 10 20 40 miles

(a) (b)(a) (b)

Figure 2. Spatial accessibility to food stores in southwest Mississippi. (a) By the KD2SFCAmethod, (b) by the 2-step floating catchment area method, for 30-minute travel time.

Geographic disparities in accessibility to food stores in southwest Mississippi 665

Page 8: Geographic disparities in accessibility to food stores in southwest Mississippi

between Jackson and Vicksburg and the southern suburbs, which might be related tothe smoothing effect without distance decay of supply ^ demand interactions. The com-parison between the two accessibility scores also shows the level of agreement betweenthe two methods. For example take the case of d0 � 45 minutes. Given this bandwidth,the two methods generate accessibility scores with an identical weighted mean and nearly

7

6

5

4

3

2

1

Spatialaccessibilityscore

by2SFCA

1 2 3 4 5 6 7

Spatial accessibility score by KD2SFCA

Figure 3. Spatial accessibility by the 2-step floating catchment method (2SFCA) and the KD2SFCAmethod (d0 � 45 minutes).

Standard

deviationofspatialaccessibility

2.5

2.0

1.5

1.0

0.5

0.0

KD2SFCA2SFCA

20 30 40 50 60 70 80 90Bandwidth (minutes)

Figure 4. Standard deviations of spatial accessibility by the 2-step floating catchment area (2SFCA)method and the KD2SFCA method.

666 D Dai, F Wang

Page 9: Geographic disparities in accessibility to food stores in southwest Mississippi

the same standard deviation. Figure 3 shows the distribution of accessibility scoresin census tracts by the two methods plotted against the 458 line. It shows that the twomethods are in general agreement. The 2SFCA method generates relatively higher scoresin poor-access areas and lower scores in good-access areas than the KD2SFCA method.

We examine further the sensitivity of the two methods to the choice of catchmentsize. The weighted mean of accessibility in the entire study area remains constantregardless of the change of catchment size (Wang, 2006, pages 95 ^ 96). Figure 4 showshow the standard deviations of the accessibility scores vary in response to differentcatchment sizes. By either method, a larger bandwidth leads to stronger smoothing andthus a lower standard deviation of the accessibility scores. For bandwidths less than80 minutes, the KD2SFCA method returns accessibility scores with an equal or higherstandard deviation than the 2SFCA method. This is likely to be attributable to theuneven distributions of supply and demand locations. In our case, there is a moreclustered pattern of the food stores (supply) than of the population (demand) locations.The KD2SFCA tends to inflate the terms in the numerator (relative to the denominator)more so than the 2SFCA does within a reasonable distance (or travel-time) threshold.Consequently, the KD2SFCA method yields higher accessibility scores in areas witha dense distribution of stores and lower accessibility scores in areas with a sparsedistribution, and thus a higher deviation for all accessibility scores than the 2SFCAmethod. Figures 2(a) and 2(b) are based on a bandwidth of 30 minutes, where theaccessibility scores by the KD2SFCA method have a higher variance than those by the2SFCA method. Note that figure 2(a) shows more variation in accessibility than is shownin figure 2(b) (eg the larger rural ^ urban differential in the northwest and southwest areas).When the bandwidth exceeds 80 minutes, the kernel function discounts remote stores moresignificantly, and therefore the KD2SFCA method leads to a stronger smoothing effectand accessibility scores with a smaller variance than the 2SFCA method.

5 Assessing socioeconomic disparities in spatial accessibility to food storesIn this section we examine the socioeconomic disparities in spatial access to foodstores; that is, how spatial access is related to neighborhood socioeconomic structure.On the basis of the literature, twelve variables are selected to capture neighborhoodcharacteristics. The year for which food retailers entered the SIC coding system in the

Table 2. Correlation coefficients between twelve socioeconomic variables and spatial accessibil-ity.

Variable Abbreviation Correlationcoefficient

Rural population Rural ÿ0.451**Female-headed household Female 0.444**Occupied house ownership HOwner ÿ0.323**Median household income (US $) MidInco ÿ0.239**Population (aged 17+ years) below poverty level Poverty 0.237**Carless occupied household NoCar 0.218*Linguistically isolated household LingH 0.184*Nonwhite population Minor 0.172Household lacking complete plumbing facilities NoPlm ÿ0.154Population (aged 25+ years) without high school diploma NoEdu 0.100Household lacking complete kitchen facilities NoKitch ÿ0.098Occupied house with >1 occupant per room Over1 0.002

*Significant at p � 0:05; **significant at p � 0:01.Note: All variables are measured as ratios except for the median household income;bandwidth is 30 minutes. Spatial accessibility is the logarithmic transformation of the

Geographic disparities in accessibility to food stores in southwest Mississippi 667

Page 10: Geographic disparities in accessibility to food stores in southwest Mississippi

database was, in general, 2000 and thus the assessment of food accessibility correlateswith the definition of socioeconomic variables derived from the 2000 census.

The initial assessment is a bivariate correlation analysis between the spatial acces-sibility and each of the twelve individual variables. The results, shown in table 2,suggest significant disparities in spatial access to food retailers. Rural areas are clearlydisadvantaged. Poor accessibility is pervasive in rural areas, and improves as urban-ization increases (figure 5). Better spatial accessibility also tends to be associated witha higher female-headed-household rate, a higher poverty rate, a higher carless rate, andneighborhoods with more linguistic barriers. Conversely, lower accessibility is corre-lated with higher home ownership and higher median income levels. Also note thecorrelation between the variables as seen in table 3.

This research uses the principal components factor analysis method (Wang, 2009)to consolidate the twelve variables into three independent factors. The eigenvaluessuggest that three factors account for more than 80% of the total variance of thetwelve variables. Table 4 presents the loadings of each variable on the three factorsafter applying the varimax rotation technique. Note that the variable `rural population'

2.0

1.5

1.0

0.5

0.0

ÿ0.5

Log(spatialaccessibility)

0.0 0.2 0.4 0.6 0.8 1.0

Rural population ratio

R 2 � 0:203

Figure 5. Scatter plot of rural population ratio versus spatial accessibility.

Table 3. Correlation coefficients between the twelve socioeconomic variables (see table 2 forabbreviated variable names).

Variable Variable

Rural Minor Female LingH Poverty NoEdu

Rural 1Minor ÿ0.364** 1Female ÿ0.603** 0.558** 1LingH ÿ0.188* ÿ0.035 0.160 1Poverty ÿ0.136 0.767** 0.537** ÿ0.016 1NoEdu 0.141 0.636** 0.325** ÿ0.054 0.852** 1MidInco ÿ0.032 ÿ0.635** ÿ0.495** 0.008 ÿ0.852** ÿ0.822**HOwner 0.647** ÿ0.496** ÿ0.766** ÿ0.133 ÿ0.511** ÿ0.237**Over1 ÿ0.233** 0.800** 0.457** ÿ0.038 0.706** 0.590**NoPlm 0.282** 0.323** 0.010 ÿ0.094 0.498** 0.590**NoKitch 0.206* 0.297** 0.054 ÿ0.086 0.485** 0.569**NoCar ÿ0.264** 0.710** 0.598** ÿ0.013 0.823** 0.776**

*Significant at p � 0:05; **significant at p � 0:01.

668 D Dai, F Wang

Page 11: Geographic disparities in accessibility to food stores in southwest Mississippi

spreads its loadings among the three factors, and therefore is not captured by onesingle factor.

Factor 1 (urban socioeconomic disadvantages) captures mainly seven variables:female-headed household, nonwhite population, occupied house with >1 occupantper room, population (aged 17+ years) below poverty level, carless occupied household,ownership of occupied household, and median household income. In other words, ahigher factor 1 score in an area is associated with higher percentages of female-headedhouseholds and minorities, crowdedness, poverty, lack of transportation mobility, andlow levels of homeownership and income. These seven variables indicate socioeco-nomic disadvantage. By plotting the factor 1 scores against the rural population ratioas shown in figure 6(a), we can see that this indicator of socioeconomic disadvantagesis predominately high in urban areas. Figure 7 also shows the concentration of highfactor 1 scores in inner cities and towns (eg Jackson, Vicksburg, Natchez, McComb,

Table 3 (continued)

Variable Variable

MidInco HOwner Over1 NoPlm NoKitch NoCar

MidInco 1HOwner 0.444** 1Over1 ÿ0.586** ÿ0.513** 1NoPlm ÿ0.468** ÿ0.037 0.247** 1NoKitch ÿ0.453** ÿ0.130 0.208* 0.909** 1NoCar ÿ0.685** ÿ0.590** 0.654** 0.518** 0.516** 1

Table 4. Rotated factor structure of twelve variables on the three factors.

Factor 1: Factor 2: Factor 3:urban socio- rural socio- culturaleconomic economic barriersdisadvantages disadvantages

Female-headed household 0.924 0.066 ÿ0.127Nonwhite population 0.857 0.230 0.096Occupied house with >1 occupant per room 0.809 0.180 0.165Population (aged 17+ years) below poverty 0.784 0.522 0.067level

Carless occupied household 0.767 0.482 ÿ0.042Occupied house ownership ÿ0.784 0.089 0.327Median household income (US $) ÿ0.651 ÿ0.557 ÿ0.094Rural population ÿ0.593 0.497 0.413Household lacking complete plumbing facilities 0.098 0.914 0.005Household lacking complete kitchen facilities 0.115 0.885 ÿ0.060Population (aged 25+ years) without high 0.557 0.704 0.174school diploma

Linguistically isolated households ÿ0.021 0.025 ÿ0.895

Geographic disparities in accessibility to food stores in southwest Mississippi 669

Page 12: Geographic disparities in accessibility to food stores in southwest Mississippi

(a)

(b)

(c)

3.0

2.0

1.0

0.0

ÿ1.0

ÿ2.0

4.0

2.0

0.0

ÿ2.0

2.0

1.0

0.0

ÿ1.0

ÿ2.0

ÿ3.0

ÿ4.0

ÿ5.0

R 2 � 0:352

R 2 � 0:247

R 2 � 0:171

Factor3

Factor2

Factor1

0.0 0.2 0.4 0.6 0.8 1.0Rural population ratio

Figure 6. Scatter plots of rural population ratio versus three factors. (a) Factor 1: urban socio-economic disadvantages; (b) factor 2: rural socioeconomic disadvantages; (c) factor 3: culturalbarriers.

670 D Dai, F Wang

Page 13: Geographic disparities in accessibility to food stores in southwest Mississippi

and Brookhaven). Factor 2 (rural socioeconomic disadvantages) captures mostly threevariables: household lacking complete plumbing facilities, household lacking completekitchen facilities, and population (aged 25+ years) without high-school diploma. Ahigher factor 2 score is linked to areas with low education attainment and poorhousing conditions. Figure 6(b) shows that its scores correlated with rural populationratios (also see figure 8). Factor 3 (cultural barriers) primarily captures the variable`linguistically isolated household' (household unable to speak English). Neighborhoodswith higher scores of cultural barriers also tend to be more rural [figure 6(c) andfigure 9]. More than half the linguistically isolated households (55%) in the study areaare HispanicöSpanish speakers, many of whom are immigrants engaged in farming andlumbering, among other jobs, in rural regions. The rest include Asian (20%), Indo-European (19%) (such as Russian and Serbo-Croatian), and others (6%) such as African,or Arab (speaking other languages) which scatter across the study area.

Multivariate regression analysis is used to examine the joint effects of the threefactors on spatial accessibility. In addition to the ordinary least squares model, thespatial lag model (Anselin, 1988) is used to control for spatial autocorrelation inthe data. The results are presented in table 5. Both models show that factor 1 andfactor 3 are statistically significant in affecting the variation of spatial accessibility.Areas with higher urban socioeconomic disadvantages actually enjoy better spatialaccessibility. This is not surprising as other studies reveal that inner-city neighborhoodswith socioeconomic disadvantages also tend to have better spatial accessibility to jobs

Factor 1 score

ÿ1.6739 toÿ0.9845ÿ0.9844 to ÿ0.4472ÿ0.4471 to 0.1751

0.1752 to 1.1516

1.1517 to 2.2697

0 5 10 20

miles

Figure 7. Urban socioeconomic disadvantages (factor 1) in southwest Mississippi.

Geographic disparities in accessibility to food stores in southwest Mississippi 671

Page 14: Geographic disparities in accessibility to food stores in southwest Mississippi

(Shen, 1998; Wang, 2003). However, this advantage may not transfer to true convenienceof food access since the socioeconomic disadvantages remain as major obstacles bytheir effect on mobility and affordability. Negative values of the cultural barrier factorindicate that neighborhoods with cultural barriers, represented by linguistic isolation(more so in rural regions in our study area), are linked to significantly poor foodaccessibility. Factor 3 is statistically significant but the significance is lower than forfactor 1. Factor 2 has no significant effect in spatial accessibility.

0 5 10 20

miles

Figure 8. Rural socioeconomic disadvantages (factor 2) in southwest Mississippi.

Table 5. Results from both OLS (ordinary least squares) model and the spatial lag model.

OLS model Spatial lag model

coefficients t-values coefficients t-values

Constant 1.259** 46.439 0.714 5.562Factor 1 0.110** 4.032 0.084** 3.251Factor 2 ÿ0.047 ÿ1.718 ÿ0.048 ÿ1.931Factor 3 ÿ0.085** ÿ3.119 ÿ0.054* ÿ2.180Spatial lag 0.439** 4.401

R 2 0.1996 0.338

*Significant at p � 0:05; **significant at p � 0:01.

672 D Dai, F Wang

Page 15: Geographic disparities in accessibility to food stores in southwest Mississippi

We also tested the catchment sizes of 15, 20, and 25 minutes travel time. The resultsshow that the factor 1 was consistently significant in the regression models, but notfactors 2 and 3. That is to say, areas with urban socioeconomic disadvantages remainat an advantageous position with regard to spatial access to food stores.

6 Discussion and conclusionIn planning research growing attention has been focused on examining inequality inspatial accessibility to resources such as food access (Algert et al, 2006; Donkin et al,1999; Raja et al, 2008), health care access (Dai, 2010; Langford and Higgs, 2006; Luoand Qi, 2009), and green-space access (Coombes et al, 2010; Hillsdon et al, 2006),among others. The research reported in this paper proposes the KD2SFCA approachto advance the measurement of spatial accessibility. This new method is applied toanalyzing disparities in food accessibility in southwest Mississippi.

Compared with the traditional 2SFCA method, this new method uses a kernelfunction to account for distance decay in supply ^ demand interactions within acatchment. The results generated from the two methods are generally consistentwith each other, and both highlight the rural ^ urban disparities in food-store acces-sibility. In our case study, the KD2SFCA returns accessibility with an equal or higherstandard deviation than the 2SFCA for bandwidths less than 80 minutes. This isattributable to the kernel function differentiating the supply ^ demand interactionaccording to distance between them and thus capturing more spatial variability

0 5 10 20

miles

Figure 9. Cultural barriers (factor 3) in southwest Mississippi.

Geographic disparities in accessibility to food stores in southwest Mississippi 673

Page 16: Geographic disparities in accessibility to food stores in southwest Mississippi

within a small-to-medium bandwidth. When the bandwidth exceeds 80 minutes(considerably large), the KD2SFCA returns access scores with a lower standarddeviation than the 2SFCA.

Inequality in spatial accessibility is examined by its association with location(urban versus rural) and nonspatial (eg socioeconomic) factors. The results indicatethat cities and major towns enjoy better spatial accessibility than rural areas, which isconsistent with findings from previous research (eg Apparicio et al, 2007). The resultsalso suggest that areas with more cultural barriers (linguistically isolated minorityneighborhoods) tend to be in rural areas. This pattern of cultural barriers, however,may reflect the population settlement unique to the study area. The geographicbarriers for rural areas may result in poor nutrition intake and impose a healthdisadvantage on rural residents (Burns et al, 2004), especially for low-income,carless, and linguistically isolated families. The study further reveals that theadvantage in spatial access for urban areas is extended to areas with more `urbansocioeconomic disadvantages', a `competitive advantage' for inner-city and low-income residents as described in Porter (1995). However, this convenience may nottransfer to true advantage for the urban poor, who often have less mobility, becauseof lower car ownership, and poor food affordability. They often settle for unhealthydiets due to cheap food bought nearby because of the burden of carrying heavygrocery bags via public transit (Algert et al, 2006) or to lack of income for purchas-ing high-quality food (Donkin et al, 1999). In other words, the issue lies in `who theyare' instead of `where they are'.

The findings have some implications for public policy. In rural areas, given theirpoor spatial accessibility, interventions should emphasize reducing the travel barriersto reaching grocery stores. Establishing local stores that provide fresh foods maylessen long-distance traveling for groceries and thus improve the spatial accessibility.Literature shows that opening large grocery retail outlets in rural areas may not bethe best solution as they might deflect sales from smaller stores and this may possiblyresult in their closure (Clarke et al, 2002; Smoyer-Tomic et al, 2006). Therefore,policies need to focus on promoting local farmers' markets, community gardens,and backyard planting to improve food quality and freshness. In urban areas,despite better spatial access scores in the socioeconomically deprived neighborhoods,questions arise as to whether these neighborhoods can truly benefit from such anadvantage. These neighborhoods also tend to have high percentages of carless house-holds, populations below the poverty level, and low household income. These arenonspatial barriers that also affect the accessibility to and affordability of qualityfoods for residents in these neighborhoods. Policies should emphasize improvementin mobility and affordability in these neighborhoods. Solutions in these neighbor-hoods may include sidewalk improvement, enhancing vehicle ownership, improvingpublic transit services near residences (Wang, 2003), and improvement of economicopportunities (Porter, 1997).

Future work may advance the research in several directions. First, empiricalstudies of actual travel behavior when food shopping will help us choose the appro-priate kernel functions and parameters in the functions. Second, surveys are neededto collect data on food items, quantities, qualities, and prices in order to improvethe weight assignments of retailers. Third, accessibility measures can be improvedby accounting for multiple transportation modes such as public transit, cycling, andwalking.

674 D Dai, F Wang

Page 17: Geographic disparities in accessibility to food stores in southwest Mississippi

Acknowledgements. The first author is grateful to the start-up support and research initiation grantfrom Georgia State University. He also thanks Laura Joseph and Amy Moore for editorial support,both of whom received graduate research assistantships from the Partnership for Urban HealthResearch at Georgia State University. The second author would like to acknowledge support fromthe National Natural Science Foundation of China (No. 40928001) and an adjunct professorship atYunnan University of Finance and Economics in the summer of 2010. Comments from twoanonymous reviewers helped us prepare the final version of the paper.

ReferencesAlgert S J, Agrawal A, Lewis D S, 2006, ` Disparities in access to fresh produce in low-income

neighborhoods in Los Angeles''American Journal of Preventive Medicine 30 365 ^ 370Anselin L, 1998 Spatial Econometries: Methods and Models (Kluwer, Dordrecht)Apparicio P, Cloutier M S, Shearmur R, 2007, ` The case of Montreal's missing food deserts:

evaluation of accessibility to food supermarkets'' International Journal of HealthGeographics 6 4

Ashman L,Vega J D, Dohan M, Fisher A, Hippler R, Romain B, 1993, ` Seeds of change: strategiesfor food security for the inner city'', UCLA Department of Urban Planning, Los Angeles, CA

Austin B S, Melly S J, Sanchez B N, Patel A, Buka S, Gortmaker S L, 2005, ` Clustering of fast-food restaurants around schools: a novel application of spatial statistics to the study of foodenvironments''American Journal of Public Health 95 1575 ^ 1581

Bingham R D, Zhang Z, 1997, ` Poverty and economic morphology of Ohio central-cityneighborhoods'' Urban Affairs Review 32 766 ^ 796

Brunsdon C, 1995, ` Estimating probability surfaces for geographical point data: an adaptive kernelalgorithm'' Computers and Geosciences 21 877 ^ 894

Burns CM,Gibbon P, BoakR,Baudinette S,Dunbar JA, 2004,` Food cost and availability in a ruralsetting in Australia''Rural and Remote Health 4 311

Cao R, Cuevas A, Gonzalez-ManteigaW, 1994, `A comparative study of several smoothingmethods in density estimation'' Computational Statistics and Data Analysis 17 153 ^ 176

Centers for Disease Control and Prevention, 2006, ` State-specific prevalence of obesity amongadultsöUnited States, 2005''Morbidity and MortalityWeekly Report 55 985 ^ 988

Cervigni F, Suzuki Y, Ishii T, Hata A, 2008, ` Spatial accessibility to pediatric services'' Journalof Community Health 33 444 ^ 448

Clarke G, Eyre H, Guy C, 2002, ` Deriving indicators of access to food retail provision in Britishcities: studies of Cardiff, Leeds and Bradford''Urban Studies 39 2041 ^ 2060

Coombes E, Jones A P, Hillsdon M, 2010, ` The relationship of physical activity and overweightto objectively measured green space accessibility and use'' Social Science and Medicine 70816 ^ 822

Dai D, 2010, ` Black residential segregation, disparities in spatial access to health care facilities, andlate-stage breast cancer diagnosis in metropolitan Detroit'' Health and Place 16 1038 ^ 1052

Darmon N, Ferguson E, Briend A, 2002, `A cost constraint alone has adverse effects on foodselection and nutrition density: an analysis of human diets by linear programming'' Journalof Nutrition 132 3764 ^ 3771

DonkinA JM,Dowler EA, Stevenson S J,Turner SA,1999,` Mapping access to food at a local level''British Food Journal 10 554 ^ 564

Edelstein S L, KnowlerW C, Bain R P, Andres R, Barrett-Connor E L, Dowse G K, Haffner S M,Pettitt D J, Sorkin J D, Muller D C, CollinsV R, Hamman R F,1997, ` Predictors of progressionfrom impaired glucose tolerance to NIDDM: an analysis of six prospective studies''Diabetes 46701 ^ 710

EpanechnikovVA, 1969, ` Non-parametric estimation of a multivariate probability density'' Theoryof Probability and its Applications 14 153 ^ 158

Epstein L H, Roemmich J N, Paluch R A, Raynor H A, 2001, ` Behavioral therapy in the treatmentof pediatric obesity'' Pediatric Clinics of North America 48 981 ^ 993

Fotheringham A S, Brunsdon C, Charlton M, 2000 Quantitative Geography: Perspective on SpatialData Analysis (Sage, Thousand Oaks, CA)

FotheringhamAS, BrunsdonC, CharltonM, 2002GeographicallyWeightedRegression:TheAnalysisof Spatially Varying Relationships (JohnWiley, NewYork)

Guagliardo M F, 2004, ` Spatial accessibility of primary care: concept, methods and challenges''International Journal of Health Geographics 3 3

Guy C M, 1977, `A method of examining and evaluating the impact of major retail developmentsupon existing shops and their users'' Environment and Planning A 9 491 ^ 504

Geographic disparities in accessibility to food stores in southwest Mississippi 675

Page 18: Geographic disparities in accessibility to food stores in southwest Mississippi

Guy C M, 1983, ` The assessment of access to local shopping opportunities: a comparison ofaccessibility measures'' Environment and Planning B: Planning and Design 10 219 ^ 238

Guy C M, David G, 2004, ` Measuring physical access to `healthy foods' in areas of socialdeprivation: a case study in Cardiff'' International Journal of Consumer Studies 28 222 ^ 234

Helling A, Sawicki D S, 2003, ` Race and residential accessibility to shopping and services''Housing Policy Debate 14 69 ^ 101

Hillsdon M, Panter J, Foster C, Jones A, 2006, ` The relationship between access and quality ofurban green space with population physical activity''Public Health 120 1127 ^ 1132

Hwang H-L, Rollow J, 2000, ` Data processing procedures and methodology for estimating tripdistances for the 1995 American Travel Survey (ATS)'', http://www.osti.gov/bridge/servlets/purl/763239-wnN01B/native/763239.pdf

Immergluck D, 1999, ` Neighborhoods, race, and capital: the effects of residential change oncommercial investment patterns''Urban Affairs Review 34 397 ^ 411

Johnson K, Percy S,Wagner E, 1996, ` Comparative study of food pricing and availability inMilwaukee'', a report prepared for The Food System Assessment Project of The HungerTask Force. Urban Research Center, University of Wisconsin-Milwaukee, Milwaukee,WI

Kyle R, Blair A, 2007, ` Planning for health: generation, regeneration and food in Sandwell''International Journal of Retail and Distribution Management 35 457 ^ 473

Langford M, Higgs G, 2006, ` Measuring potential access to primary healthcare services: theinfluence of alternative spatial representations of population'' The Professional Geographer58 294 ^ 306

LarsenK,Gilliand J, 2008,``Mapping the evolution of `food deserts' in a Canadian city: supermarketaccessibility in London, Ontario, 1961 ^ 2005'' International Journal of Health Geographics 7 16

Larson N I, Story M T, Nelson M C, 2009, ` Neighborhood environments: disparities in accessto healthy foods in the US''American Journal of Preventive Medicine 36 74 ^ 81

Levine N, 2009, ` CrimeStat III: a spatial statistics program for the analysis of crime incidencelocations'', http://www.icpsr.umich.edu/CRIMESTAT/

LuoW,QiY, 2009,`An enhanced two-step floating catchment area (E2SFCA)method for measuringspatial accessibility to primary care physicians''Health and Place 15 1100 ^ 1107

LuoW,Wang F, 2003, ` Measures of spatial accessibility to health care in a GIS environment:synthesis and a case study in the Chicago region'' Environment and Planning B: Planning andDesign 30 865 ^ 884

Moore LV, Roux AV D, Brines S, 2008a, ` Comparing perception-based and geographicinformation system (GIS)-based characterizations of the local food environment'' Journalof Urban Health 85 206 ^ 216

Moore LV, Roux AV D, Nettleton J A, Jacobs D R J, 2008b, `Associations of the local foodenvironment with diet qualityöa comparison of assessments based on surveys and geographicinformation systems''American Journal of Epidemiology 167 917 ^ 924

Morland K, Diez Roux AV,Wing S, 2006, ` Supermarkets, other food stores, and obesity: theatherosclerosis risk in communities study''American Journal of PreventiveMedicine 30 333 ^ 339

Must A, Spadano J, Coakley E H, Field A E, Colditz G, Dietz W H, 1991, ` The disease burdenassociated with overweight and obesity'' JAMA 282 1523 ^ 1529

Pearce J,Witten K, Bartie P, 2006, ` Neighbourhoods and health: a GIS approach to measuringcommunity resource accessibility''Journal of Epidemiology and Community Health 60 389 ^ 395

Porter M E, 1995, ``The competitive advantage of the inner city''Harvard Business ReviewMay ^ June, 55 ^ 71

Porter M E, 1997, ` New strategies for inner-city economic development'' Economic DevelopmentQuarterly 11 11 ^ 17

Raja S, Ma C,Yadav P, 2008, ` Beyond food deserts: measuring and mapping racial disparities inneighborhood food environment'' Journal of Planning Education and Research 27 469 ^ 482

Sharkey J R, Horel S, 2008, ` Neighborhood socioeconomic deprivation and minority compositionare associated with better potential spatial access to the ground-truthed food environment in alarge rural area'' Journal of Nutrition 138 620 ^ 627

Shen Q, 1998, ` Location characteristics of inner-city neighborhoods and employment accessibilityof low-income workers'' Environment and Planning B: Planning and Design 25 345 ^ 365

Smoyer-Tomic K E, Spence J C, Amrhein C, 2006, ``Food deserts in the Prairies? Supermarketaccessibility and neighborhood need in Edmonton, Canada'' The Professional Geographer 58307 ^ 326

Wang F, 2003, ``Job proximity and accessibility for workers of various wage groups''UrbanGeography 24 253 ^ 271

676 D Dai, F Wang

Page 19: Geographic disparities in accessibility to food stores in southwest Mississippi

Wang F, 2006 Quantitative Methods and Applications in GIS (CRC Press, Boca Raton, Florida)Wang F, 2009, ``Methods: factor analysis and principal-components analysis'', in International

Encyclopedia of Human Geography,Volume 4 Eds R Kitchin, N Thrift (Elsevier, Oxford)pp 1 ^ 7

Wang F, LuoW, 2005, `Assessing spatial and nonspatial factors for healthcare access: towardsan integrated approach to defining health professional shortage areas''Health and Place 11131 ^ 146

Wang M C, Kim S, Gonzalez A A, MacLeod K E,Winkleby M A, 2007, ` Socioeconomic andfood-related physical characteristics of the neighbourhood environment are associated withbody mass index'' Journal of Epidemiology and Community Health 61 491 ^ 498

Williamson D, McLafferty S, Goldsmith V, Mollenkopf J, McGuire P, 1998, ` Smoothing crimeincident data: new methods for determining the bandwidth in Kernel estimation'', in 18thESRI International User Conference, San Diego, http://proceedings.esri.com/library/userconf/proc98/PROCEED/TO850/PAP829/P829.HTM

Yang D H, Goerge R, Mullner R, 2006, ` Comparing GIS-based methods of measuring spatialaccessibility to health services'' Journal of Medical Systems 30 23 ^ 32

Zenk S, Schultz A, Israel B, James S, Bao S,Wilson M, 2005, ` Neighbourhood racial composition,neighborhood poverty, and the spatial accessibility of supermarkets in metropolitan Detroit''American Journal of Public Health 95 660 ^ 667

ß 2011 Pion Ltd and its Licensors

Geographic disparities in accessibility to food stores in southwest Mississippi 677

Page 20: Geographic disparities in accessibility to food stores in southwest Mississippi

Conditions of use. This article may be downloaded from the E&P website for personal researchby members of subscribing organisations. This PDF may not be placed on any website (or otheronline distribution system) without permission of the publisher.