This article was downloaded by: [Michigan State University]On: 12 January 2014, At: 10:47Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Geocarto InternationalPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/tgei20

Visualizing nutritional terrain: ageospatial analysis of pedestrianproduce accessibility in Lansing,Michigan, USAKirk Goldsberry a , Chris S. Duvall b , Philip H. Howard c & JoshuaE. Stevens aa Department of Geography , Michigan State University , Lansing,Michigan, USAb Department of Geography , University of New Mexico ,Albuquerque, New Mexico, USAc Department of Community, Agriculture, Recreation and ResourceStudies , Michigan State University , Lansing, Michigan, USAPublished online: 05 Aug 2010.

To cite this article: Kirk Goldsberry , Chris S. Duvall , Philip H. Howard & Joshua E. Stevens (2010)Visualizing nutritional terrain: a geospatial analysis of pedestrian produce accessibility in Lansing,Michigan, USA, Geocarto International, 25:6, 485-499, DOI: 10.1080/10106049.2010.510583

To link to this article: http://dx.doi.org/10.1080/10106049.2010.510583

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever orhowsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &

Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

Visualizing nutritional terrain: a geospatial analysis of pedestrian

produce accessibility in Lansing, Michigan, USA

Kirk Goldsberrya*, Chris S. Duvallb, Philip H. Howardc and Joshua E. Stevensa

aDepartment of Geography, Michigan State University, Lansing, Michigan, USA; bDepartmentof Geography, University of New Mexico, Albuquerque, New Mexico, USA; cDepartment of

Community, Agriculture, Recreation and Resource Studies, Michigan State University,Lansing, Michigan, USA

(Received 5 April 2010; final version received 12 July 2010)

This article considers how geospatial analyses can influence cartographic outputsin studies of the spatial structure of food environments. We make twocontributions. First, we present a new approach to conceiving and visualizingurban food environments as ‘nutritional terrains’, in which the opportunities andcosts of locating (healthful) food vary continuously across space. While otherresearchers have conceptualized and represented food environments as contin-uous phenomena, we use detailed data to produce maps of food accessibility thathave high resolution both spatially and in terms of food availability. Second, weshow that decisions made about measuring and modelling food accessibility cancreate artifactual patterns independently of actual variation in food-environmentcharacteristics. Although the type of method-driven patterning we identify willnot surprise cartographers, we argue that non-geographers using geographicinformation technologies to visualize food environments must give greaterattention to the unintended consequences of choices made in geospatial analyses.

Keywords: food environments; accessibility; visualization

Introduction

Built environments may constrain dietary choices, and contribute to diet-relatedpublic health problems, including overweight and obesity, diabetes and cardiovas-cular disease (French et al. 2001, Papas et al. 2007, Ford and Dzewaltowski 2008).Fundamentally, many of the public health problems observed in the developedworld arise from over-consumption of calorie-dense, nutrient-poor foods, such asprocessed foods, and concomitant under-consumption of nutrient-dense, calorie-poor foods, such as fresh produce (Halkjær et al. 2009, Liese et al. 2009). Adequateconsumption of fresh fruits and vegetables contributes to better health outcomes(Zenk et al. 2005a,b, Adebawo et al. 2006, Morland and Filomena 2007), yetmillions of people who can afford to purchase these foods continue to under-consume them.

In the United States, researchers are devoting more attention to the role ofenvironmental context in food consumption patterns (Shaw 2006, Beaulac et al. 2009).

*Corresponding author. Email: kg@msu.edu

Geocarto International

Vol. 25, No. 6, October 2010, 485–499

ISSN 1010-6049 print/ISSN 1752-0762 online

� 2010 Taylor & Francis

DOI: 10.1080/10106049.2010.510583

http://www.informaworld.com

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

The accessibility of healthy food choices varies considerably depending on anindividual’s geographic situation. A number of recent studies have quantified andmapped links between demographics, food accessibility and diet (White 2007, BlackandMacinko 2008,McKinnon et al. 2009, Larson et al. 2009). From this literature, twomain points are particularly relevant to our argument in this article. First, within agiven urban area there may be substantial geographic variation in access to healthyfood choices. In general, economically disadvantaged and minority-dominatedneighbourhoods in the US have lower than average access to large retailers, includingsupermarkets, and thus generally experience higher prices for narrower selections offood items (Inagami et al. 2006). Second, these physical and economic constraints affectdietary choices, such that diet-related health problems tend to be higher ineconomically disadvantaged and minority-dominated neighbourhoods (Laraia et al.2004, Andreyeva et al. 2008).

Studies of food accessibility represent an important advance in public health, yetmany of these studies are cartographically simplistic, and, as a result, may havemisleading results. In this article, we argue that public health-minded researchersmust give greater attention to geospatial analysis as a possible source of artifactualpatterns of food geography, because different, reasonable assumptions made inmodelling food accessibility can create starkly different cartographic outputs. Wesupport this argument with a case study of the food environment in Lansing,Michigan, USA.

Our research is significant because we explicitly build upon basic cartographicconcepts both to provide a new approach to conceptualizing and visualizing foodaccessibility, and to identify an overlooked methodological issue that must beaddressed in future studies. First, we assert that food environments are continuousgeographic phenomena, and thus must be understood and visualized using surface orisarithmic approaches (Slocum et al. 2009). We conceptualize Lansing’s foodenvironment as a ‘nutritional terrain’, where the opportunities and costs of finding(healthful) food vary continuously. While some geographic information system(GIS) analyses of food environments use surface or isarithmic approaches tovisualization, many others represent food environments using point maps of retaillocations, or choropleth maps that categorize pre-defined areas (such as censustracts) by the number and/or type of retailers located within an area. By comparingdifferent methods of creating surface and isarithmic representations of a single foodenvironment, and also by comparing these representations to derived choroplethmaps, we show that decisions about geospatial modelling of accessibility must be amore prominent concern in health geographic research.

Additionally, we present a novel means of characterizing food retailers by usingthe availability of individual types of food as our fundamental elements of spatialanalysis, as opposed to the more common practice of simply mapping retaillocations. While our maps do represent the locations of food retailers, we did notassign labels such as ‘supermarket’, ‘convenience store’ and ‘grocer’. Instead, weinventoried the fresh produce offerings of each store, and mapped the availability ofeach type of produce in our study area. Thus, the ‘nutritional terrain’ we describe isnot just continuous but also has very high resolution in terms of food availability.While our results mostly confirm the types of spatial inequalities in food accessibilityobserved by other researchers, our focus is on the geospatial methods that underpinmany examples of food-environment research. We caution that cartographicdepictions of food environments depend greatly upon measurement approach.

486 K. Goldsberry et al.

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

Although geographic information scientists (GIScientists) are familiar with howdecisions made in modelling affect visualization (Monmonier 1996), the waysanalytical methods can produce spurious patterns may be unfamiliar to many usersof GISs in health geography who lack cartographic training.

Background

In general, accessibility is an important characteristic of urban geography.Accessibility is commonly cited as a goal in urban planning, building design andsocial justice. However, as pointed out by Peter Gould (1969), accessibility is one ofthose common terms that everyone uses until faced with the problem of defining andmeasuring it. Fundamentally, accessibility is determined by three components: thespatial distribution of potential destinations, the ease of reaching each destination,and the character of each destination (Handy and Niemeier 1997). In the context offood environments, nutritional accessibility metrics must account for the spatialarrangement of food retailers, the costs of travelling to any retailer and the foodsavailable at each retailer.

Research on ‘food deserts’ – areas with low accessibility to (healthful) foods – hasfocused on the spatial distribution of retail food locations (Shaw 2006). One notablefinding is that both quality and quantity of retail food locations vary significantlybetween neighbourhoods (Cummins and Macintyre 2006). Minority-dominated andsocioeconomically disadvantaged neighbourhoods generally have fewer sources offresh produce than non-Hispanic white neighbourhoods (Chung and Meyers 1999,Morland et al. 2002, Zenk et al. 2005b, Algert et al. 2006, Baker et al. 2006, Blockand Kouba 2006, Powell et al. 2007a,b,c). Although these findings are important, themajority of food-environment studies to date have some methodological limitations(Howard and Fulfrost 2007). For our purposes in this article, the most significantproblems include:

. Coarse spatial analysis. The majority of food-environment research has beenconducted at coarse aggregate levels, such as census tracts, ZIP codes orcounties (Morton and Blanchard 2007, Powell et al. 2007a,b,c). The use ofaggregate areas can lead to misleading results, because accessibility is acontinuous phenomenon (Handy and Niemeier 1997). For instance, someportions of aggregate political units may have good access to produce whileother portions do not. Given the likelihood of significant intra-unit variation infood accessibility, it is unwise to analyse access at such coarse spatial scaleseven if other relevant data, such as those derived from the US census, areavailable only as aggregates.

. Simplified definitions of access. Most Americans purchase food at retaillocations and travel via road networks. With each journey to a retail location,an individual consumer overcomes a cost of separation. This cost is frequentlyviewed as a function of time or distance. Unfortunately, many studies fail toaccurately model separation costs. For example, some studies use Euclideandistance, or Manhattan-block distance (e.g. Zenk et al. 2005a,b) to modelaccessibility. These measures do not mimic the lived experience of consumers,who must travel along pre-defined, often indirect routes.

. Simplified definitions of retail food sources. Many studies have utilizedproximity to the nearest retailer as a measure of access (Laraia et al. 2004,

Geocarto International 487

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

Zenk et al. 2005a,b). This is problematic because store inventories differ, evenbetween externally similar stores; not all ‘supermarkets’ are equal, and somesmall stores may have large inventories based on various factors. Moredetailed characterization of food retailers, based on food availability, isneeded.

. Imprecise location data. Few studies of food access have accurately modelledthe actual locations of retail locations (Sharkey and Horel 2008). Previousinvestigations rely heavily on geocoded addresses (Laraia et al. 2004) thatothers have demonstrated to be significantly inaccurate. These errorspropagate when included in network accessibility analyses.

Perhaps the most important of these limitations is insufficient attention to thetypes of foods offered within retailers. Morland et al. (2006), for example,examined how the presence of a few, broad retail types influences obesity.Unfortunately the meanings of the categories ‘supermarkets’, ‘grocers’ and‘convenience stores’ can be ambiguous. By focusing on the availability andaccessibility of individual food items, GIS analyses can quantify and visualize the‘nutritional terrain’ in an urban area: a continuous, fine-scale geographicalcharacterization of food availability and accessibility. Each individual produce itemwithin a food environment possesses a unique pattern of availability and thusaccessibility, which depends upon the spatial arrangement of retailers offering theitem, as well as the geospatial model employed to measure accessibility. Byimproving the cartographic depictions of inequalities within urban food environ-ments our broad goal is to enable health officials to identify areas that have pooraccess to specific types of food, including produce, rather than simply poor accessto a specific type of retailer.

Data collection and analysis

Overview

Our study area is the Lansing, Michigan, USA, metropolitan area (428440N,848330W), which has a population of about 450,000 (Figure 1). We analysed thedistribution of 94 food retailers who offered fresh produce during the period of datacollection, February–April 2008. We assessed accessibility to these locations byindependently analysing the three fundamental components of geographic access: thespatial distribution of food retailers, the cost of reaching each retailer from anaddress and the produce offerings of each retailer (Handy and Niemeier 1997). Themethods used to analyse each aspect of access are described in the following sub-sections.

Spatial distribution of potential destinations

We compiled a list of all food retail locations (excluding restaurants) in the Lansingarea using commercial data purchased from ESRI (Redlands, California)supplemented with Internet searches, phone book listings and on-the-groundsearches of local streets. Next, we determined whether each location in this list(n ¼ 246) was operational and offered any fresh produce, through telephone callsand in-person visits. We defined ‘fresh produce’ as any plant food offered for sale inan uncooked, unfrozen and undried form. We chose to sample food availability

488 K. Goldsberry et al.

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

based on produce availability because consumption of fruits and vegetables isemphasized in public health campaigns (Lasley and Litchfield 2007, Sharkey andHorel 2009). We identified 94 retail locations that offered fresh produce then visitedeach outlet in person. We used hand-held GPS receivers to determine the location ofeach retailer’s entrance. Some outlets had more than one entry; in these cases werecorded the coordinates of the entry point closest to the outlet’s produce section.

Cost of reaching each destination

Successful travel to a destination incurs costs of separation (Handy and Niemeier1997). These costs are often quantified using geographic distance, travel time orrequired energy. When travel costs are overwhelming, they hinder a consumer’sability to visit a destination. For example, a supermarket may be ‘too far away’ or‘take too long to get to’ for a given consumer, who instead shops at a limited-selection convenience store. These costs are critical to dietary choice; therefore whenanalysing a food environment it is imperative that these separation costs beaccurately modelled. We chose to employ network analysis to estimate the costs-of-separation that the consumers in urban environments are forced to overcome.

We measured network distances and pedestrian travel times outward from everyproduce retailer, which enabled the calculation of expected service areas for eachlocation. Using GIS-based network analysis, we generated multiple network buffers

Figure 1. Study area: The Lansing, Michigan metropolitan area contains over 90 produceretailers including supermarkets, grocery stores, and convenience stores. Many of thesupermarkets are located towards the periphery of the urban core.

Geocarto International 489

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

to enclose all network locations varying distances (e.g. 0.1 miles, 0.2 miles) withineach retailer in the study area. We created a unique set of network buffers for everyproduce item available in the study area, which facilitated the estimation ofaccessibility to both individual produce items and cumulative numbers of produceitems.

We focused on pedestrians. Due to enhanced mobility, individuals with access toautomobiles have lower separation costs relative than those who are limited to othermodes of transport, especially walking. We defined separation costs in terms ofestimated walking time along established street networks. The walking cost (inminutes) was formulated by dividing the length of the optimal network route (inmiles) by an average walking speed of 3 miles per hour. Using the ArcGIS NetworkAnalyst tool, and detailed road network data provided by the Michigan Center ofGeographic Information, we employed network buffering to estimate realisticdistance costs that separate consumers from retail locations.

Character of destinations

We visited each retail location and recorded every type of fresh produce offered ineach location. We recorded each item priced separately in every store, with theseexceptions: multiple sizes of single items, such as large and small zucchini squash,which were recorded as a single item; minimally prepared items, such as slicedmelons, unless a given item was offered only in a minimally prepared form andpackaged spices, because of the difficulty determining whether these were dried orotherwise preserved. These data collection efforts produced a matrix thatsummarized the presence–absence of each produce item (n ¼ 447) in all 94-samplesites. As a means of characterizing individual food retailers, inventorying freshproduce availability allows more precise modelling of food accessibility than moresimplified approaches that only differentiate retailers based on external character-istics, such as square footage, chain membership or sales volume.

Food accessibility can be defined and formalized in countless ways, but theintricacies of the definition are critical to the eventual results. In this article, we applythree different accessibility metrics to estimate pedestrian access to retail produce.We used GIS to calculate accessibility scores for every point within our study in threedifferent ways. These methods are each detailed in the following sections.

Container method

The most straightforward approach to measuring geographic accessibility involvessimply counting the number of opportunities (i.e. food retailers) accessible withinsome pre-defined cost constraint (e.g. distance). Such an approach is sometimescalled a container measure because resulting accessibility scores are basically a countof the number of opportunities within a specific geographic range. For example, acontainer measure of hospital accessibility might simply count the number ofhospitals within 25 miles of an address.

Since the atomic units of analysis in our study are individual produce items, wewere able to generate service areas not only to retail locations but also morespecifically to individual items (e.g. bananas, green beans, organic spinach). Wemodelled 10-min (0.5 miles) service areas for pedestrian access for every produceitem available within the study area (Figure 2).

490 K. Goldsberry et al.

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

Figure 2. Individual item accessibility signatures: This set of small multiples shows howunique produce items have different accessibility signatures.

Geocarto International 491

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

Using GIS overlay analyses, we calculated container measures that indicated howmany unique produce items were available within a 10-min walk of every address inthe study area. Our container-measure accessibility equation is:

Ai ¼Xij2R

Pijf

Ai ¼ Total produce accessibility of location i

Pijf ¼0; produce item j is not accessible at location i within distance f1; produce item j is accessible at location i within distance f

�

f ¼ 10 min of pedestrian walkR ¼ Region containing all locations and produce types to be considered.

Weighted method

The container measure is simple and easy to understand, but treats all accessibleopportunities uniformly. An item that is barely accessible (e.g. near the limit of thecost threshold) is considered just as accessible as an item that is literally next-door.Our second accessibility metric awards higher accessibility scores to those produceitems that are closer than those that are further away. This weighted approachgenerates a higher-resolution model of accessibility for each item by identifyingservice areas that are graded spatially based on proximity.

We calculated weighted services areas for all 447 produce items and delineatedmultiple accessibility zones according to network distances. For example, areaswithin a 2-min walk to a retailer offering a particular produce item receive a higheraccessibility score than areas 42 min away. Our equation for determining weightedaccessibility is:

Ai ¼Xij2R

PijfWa

Ai ¼ Total produce accessibility of location i

Pijf ¼0; produce item j is not accessible at location i within distance f1; produce item j is accessible at location i within distance f

�

f ¼ 10 minutes of pedestrian walkWa ¼ Inverse distance weight from set aa ¼ {0.1, 0.2, 0.3, 0.4, 0.5, 1} min of walkingR ¼ Region containing all locations and produce type to be considered.

Cumulative distance method

The container method and weighted method of accessibility modelling capturedistance only in terms of pre-specified thresholds, and do not distinguish amongopportunities outside of those ranges. For our third metric, we included a measurewhere accessibility at a given location is defined as the sum of the networkseparations from all available produce items to that location. In other words, thescore reflects the total network distance a consumer would have to travel in order to

492 K. Goldsberry et al.

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

purchase every individual produce item (one item per journey) available in the studyarea. In general, the larger the result, the less accessible produce items are to thatlocation. Furthermore, a centrally located origin is more likely to have a loweraccumulated sum and therefore be more accessible than peripheral locations.

Ai ¼Xij2R

PijDit

Ai ¼ Total produce accessibility of location i

Pij ¼0; produce item j is not accessible at location i1; produce item j is accessible at location i

�

Dit ¼ Distances from set t for which accessibility at i is measuredt ¼ {0.5, 1, 1.5, 2, 2.5, 3,. . ., 12, 12.5, 13, 13.5, 14}R ¼ Region containing all locations and produce types to be considered.

Results

The three accessibility metrics each produced a different depiction of nutritionalterrain for the study area. The resulting accessibility maps exhibit different thematicsignatures; ‘food deserts’ – areas with low food accessibility – appear differentlydepending on the measurement approach.

Container approach

The output from the container method presents a patchy nutritional landscape inwhich accessibility to fresh produce is poor for most of the study area (Figure 3). Inparticular, few addresses in the densely populated centre of the study area have anyproduce items within a 10-min walkshed; more addresses in the suburban peripheryof the study area have better access to fresh produce. In other words, locations withthe greatest pedestrian access within are in relatively sparsely populated, automobile-oriented areas.

One of the benefits of the container measure is that the results directly translateto an intuitive description of fresh produce accessibility. For example, the mostproduce-abundant 10-min walkshed has access to 272 produce items. Unlike theother two accessibility metrics applied in this study, the output from the containermeasure provides a simple count of the number of produce items within a 10-minwalk. Consequently, the container approach can facilitate certain types of map tasksbetter than the others. For example, a map-reader could easily identify zones thathave pedestrian access to less than 10 different produce items. However, a drawbackof this approach is that zones outside of the 10-min constraint appear uniformlydisadvantaged; retailers that are 11 walking minutes from an address are lumped inwith others that are much more distant.

Weighted approach

The depiction of nutritional terrain using the weighted method (Figure 4) is differentthan the output based on the container metric, because the weighted method usesmultiple distance measures to characterize retailers. Furthermore, since the weighted

Geocarto International 493

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

method includes buffers of up to 20 walking minutes (one mile), its output containsfewer zones of no access. Thus, the nutritional terrain produced through theweighted method is less patchy than that produced through the container method.Most of the study area appears to have at least some pedestrian access to freshproduce.

However, the quantitative output from the weighted method is less intuitive thanthe container approach. It does not correspond to a simple count, but instead offersa more abstract scoring of accessibility. This is the result of a measurement approachthat places an emphasis on proximity to a retailer. Due to the scoring gradient in thisweighted approach, map-readers cannot simply translate accessibility scores into aclear set of nutritional consequences (e.g. zones that have pedestrian access to lessthan 10 different produce items).

Cumulative distance approach

The cumulative distance metric produces a very different depiction of nutritionalterrain (Figure 5). Using this method to model accessibility, the central parts of thecity appear to have the best accessibility to fresh produce, because these areas arelocated relatively near to all suburban locations, whereas different suburbanlocations may be located on opposite sides of the city. However, this map revealsthat some suburban areas have much greater produce accessibility than others. For

Figure 3. Results from the container method produce a patchy depiction of nutritionalterrain. A majority of the study area has no pedestrian produce access and the zones withhigher amounts of access are sparsely populated.

494 K. Goldsberry et al.

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

example, the eastern suburban periphery of the study area has better access to freshproduce than Lansing’s western suburbs – a phenomenon that is much less obvioususing the other two methods of modelling access (Figures 3 and 4). The output fromthe cumulative distance approach suggests a much more concentrated, and lesspatchy nutritional landscape.

Discussion

The relative newness of food-environment studies means that relevant methodologiesare still in stages of development; the methods used in many food-environmentstudies have been mixed in terms of quality and conceptual appropriateness (Boothet al. 2005, Papas et al. 2007, Beaulac et al. 2009). There have been few criticalassessments of geographic methods used in food-environment research (Sharkey andHorel 2008), although inappropriate spatial analyses have led to questionableconclusions about food environments (Spielman 2006). Spatial analysis is rooted ingeographic information science (GIScience), the body of theory necessary to developand implement GISs. Many researchers use computer-based GISs as research toolsfor spatial analyses; GIScientists test the theoretical, conceptual and spatialappropriateness of different uses of geographic data. For instance, GIScientistshave shown that decisions about the acquisition of geographic data and its insertioninto GISs can strongly affect public health research findings (Elliott and Wartenberg

Figure 4. Results from the weighted method produce a less patchy, and more gradeddepiction of pedestrian produce access.

Geocarto International 495

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

2004, Zandbergen 2007, Boone et al. 2008). Similar assessment of geographicmethods in the specific context of food-environment research is necessary to validatedata sources and analytical approaches widely used to develop public-health policyin the US (Sharkey and Horel 2008).

GISs can be used effectively to identify disparities in nutritional terrain acrossan urban landscape. However, the quality of output visualizations and analysesdepend upon both data quality as well as decisions made in data analysis. Manyresearchers recognize that map quality depends upon data quality – the ‘garbage-in-garbage-out’ reality of GIS analyses is widely known – but the dependence ofmap output upon analytical decisions that occur several steps prior to mapproduction is less widely recognized amongst GIS users who are not trained inGIScience. While many map-readers tend to see maps as ‘true’, all maps include‘white lies’ that cartographers use to simplify geographic reality (Monmonier1996). Furthermore, previous food-environment investigations often rely on so-called one-map solutions (Monmonier 1991) that ‘foster highly selective, authoredviews perhaps reflecting consciously manipulative or ill-conceived designdecisions’, which in turn can misrepresent or skew the geographic complexitiesof urban food environments.

Our results show that visual depictions of food environments are as vulnerablecartographic distortion as maps depicting other geographic phenomena. Despite

Figure 5. Results from the cumulative distance method produce a more concentrateddepiction of produce accessibility. Zones towards the eastern portions of the study area appearto have easier access to produce than the areas closer to the western periphery of the studyarea.

496 K. Goldsberry et al.

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

similar production methods using identical inputs, cartographic outputs can beconsiderably different and these differences can influence geovisual narratives. Forexample, if someone wanted to show that the western periphery of Lansing hasparticularly poor access to produce, a map using our data and the cumulativedistance metric in Figure 5 would be more effective than a map using either thecontainer method or the weighted method (Figures 3 and 4).

The topics of sensitivity analysis and uncertainty are not new to GIScientists, butfew food-environment studies have considered how these phenomena may affecttheir analyses. Furthermore, and perhaps more troubling, most previous investiga-tions use only one approach to model nutritional accessibility, and rely on its outputto correlate food-environment characteristics with other geographic variables. Yetour results indicate that applying different accessibility metrics to a single dataset canproduce very different, but equally valid cartographic outputs; these differenceswould influence all subsequent statistical operations, such as analyses designed totest possible correlations between food accessibility and diet-related healthindicators. The implications of decisions made in the beginning stages of anutritional accessibility study propagate throughout its entire methodology.

There is a need to improve the methods through which urban food environmentsare visually depicted. While it is clear that differences in food accessibility withinurban areas contribute to positive or negative health outcomes, our results show thatit is unclear how at-risk areas should best be identified through accessibilitymodelling. We have used three different approaches to model food accessibility, andhave shown that each of these approaches produces a significantly different map ofLansing’s nutritional terrain. Although each of our approaches is rooted in previousaccessibility research (Handy and Niemeier 1997), we have found uses of only thecontainer method in food-environment studies, and no assessment of whether this isthe best method of describing or analysing food accessibility. Public officials anddecision-makers rely upon maps of food accessibility in developing and implement-ing public-health policies, yet the cartographic implications of the geospatialanalyses used to produce such maps have received very little consideration.Geographic accessibility is a critical component to many types of food-environmentresearch, yet most studies have employed simplistic approaches to accessibilitymeasurement. Accessibility is determined by the spatial distribution of destinations,the costs of reaching these destinations, and the character of these destinations.Modelling accessibility requires many decisions about how to simplify geographicreality, meaning that food accessibility studies inevitably include a number ofsimplifying assumptions that influence cartographic outputs. Research is needed toevaluate which metrics – of all three components of accessibility – provide the mostmeaningful description of the environmental context of diet-related public healthproblems. Furthermore, as suggested by Monmonier (1991), future food-environ-ment investigations should consider adopting a multiple-output approach, based onmultiple access models and cartographic design decisions, that present a morediverse, if also more ambiguous depiction of nutritional accessibility. The dangersassociated with the ‘one-map’ solutions that currently dominate food-environmentinvestigations are particularly potent when cartographic outputs are so sensitive toinput parameters, which is exactly what our results demonstrate.

We have shown that given a set of precise inputs, GISs can effectivelyproduce high-resolution depictions of nutritional terrain. Indeed, more food-environment research should be based upon direct observations of food-environment

Geocarto International 497

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

characteristics (Papas et al. 2007, Ford and Dzewaltowski 2008), such as theinventories of fresh produce availability we used in this research. However, our mainpoint is that the appearance of nutritional disparities depends upon the intricacies ofthe accessibility modelling approach. This suggests that there is inherent uncertaintywithin any quantitative approach to nutritional accessibility modelling. To date thisparticular issue has been neglected by most studies, and future investigations shouldstart to consider its impact on their own results.

References

Adebawo, O., et al., 2006. Fruits and vegetables moderate lipid cardiovascular risk factor inhypertensive patients. Lipids Health, 5, 14.

Algert, S.J., Agrawal, A., and Lewis, D.S., 2006. Disparities in access to fresh produce in low-income neighborhoods in Los Angeles. American Journal of Preventive Medicine, 30 (5),365–370.

Andreyeva, T., et al., 2008. Availability and prices of foods across stores and neighborhoods:the case of New Haven, Connecticut. Health Affairs, 27 (5), 1381–1388.

Baker, E.A., et al., 2006. The role of race and poverty in access to foods that enableindividuals to adhere to dietary guidelines. Preventing Chronic Disease, 3 (3), 1–11.

Beaulac, J., Kristjansson, E., and Cummins, S., 2009. A systematic review of food deserts.Preventing Chronic Disease, 6 (3), 1–10.

Black, J.L. and Macinko, J., 2008. Neighborhoods and obesity. Nutrition Reviews, 66 (1), 2–20.

Block, D. and Kouba, J., 2006. A comparison of the availability and affordability of a marketbasket in two communities in the Chicago area. Public Health Nutrition, 9 (7), 837–845.

Boone, J.E., et al., 2008. Validation of GIS facilities database: quantification and implicationsof error. Annals of Epidemiology, 18 (5), 371–377.

Booth, K.M., Pinkston, M.M., and Poston, W.S., 2005. Obesity and the built environment.Journal of the American Dietetic Association, 105 (5), S110–S117.

Chung, C. and Myers, S.L. Jr, 1999. Do the poor pay more for food? An analysis of grocerystore availability and food price disparities. The Journal of Consumer Affairs, 33 (2), 276–296.

Cummins, S. and Macintyre, S., 2006. Food environments and obesity – neighborhood ornation? International Journal of Epidemiology, 35, 100–104.

Elliot, P. and Wartenberg, D., 2004. Spatial epidemiology: current approaches and futurechallenges. Environmental Health Perspectives, 112 (9), 998–1006.

Ford, P.B. and Dzewaltowski, D.A., 2008. Disparities in obesity prevalence due to variation inthe retail food environment: three testable hypotheses. Nutrition Reviews, 66 (4), 216–228.

French, S.A., et al., 2001. Environmental influences on eating and physical activity. AnnualReview of Public Health, 22, 309–335.

Gould, P., 1969. Spatial diffusion. Resource Paper 17.Washington, DC: American Associationof Geographers (AAG).

Halkjær, J., et al., 2009. Dietary predictors of 5-year changes in waist circumference. Journal ofthe American Dietetic Association, 109 (8), 1356–1366.

Handy, S.L. and Niemeier, D.A., 1997. Measuring accessibility, an exploration of issues andalternatives. Environment and Planning A, 29, 1175–1194.

Howard, P.H. and Fulfrost, B., 2007. The density of retail food outlets in the Central CoastRegion of California: associations with income and Latino ethnic composition. Journal ofHunger & Environmental Nutrition, 2 (4), 16.

Inagami, S., et al., 2006. You are where you shop: grocery stores locations, weight, andneighborhoods. American Journal of Preventive Medicine, 31 (1), 10–17.

Laraia, B., et al., 2004. Proximity of supermarkets is positively associated with diet qualityindex for pregnancy. Preventive Medicine, 39 (5), 869–875.

Larson, N.I., Story, M.T., and Nelson, M.C., 2009. Neighborhood environments: disparitiesin access to healthy foods in the U.S. American Journal of Preventive Medicine, 36 (1),74.e10–81.e10.

498 K. Goldsberry et al.

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014

Lasley, E.C. and Litchfield, R.E., 2007. Fresh produce in rural Iowa: availability andaccessibility. Journal of Hunger & Environmental Nutrition, 2 (2/3), 5–13.

Liese, A.D., et al., 2009. Food intake patterns associated with incident type 2 diabetes.Diabetes Care, 32 (2), 263–268.

McKinnon, R.A., et al., 2009. Measures of the food environment: a compilation of theliterature, 1990–2007. American Journal of Preventive Medicine, 36 (4, Supplement 1),S124–S133.

Monmonier, M., 1991. Ethics in map design. Six strategies for confronting the traditional one-map solution. Cartographic Perspectives, 10, 3–9.

Monmonier, M., 1996. How to lie with maps. Chicago, IL: University of Chicago Press.Morland, K., Diez Roux, A.V., and Wing, S., 2006. Supermarkets, other food stores, and

obesity: the atherosclerosis risk in communities study. American Journal of PreventiveMedicine, 30 (4), 333–339.

Morland, K. and Filomena, S., 2007. Disparities in the availability of fruits and vegetablesbetween racially segregated urban neighbourhoods. Public Health Nutrition, 112, 1481–1489.

Morland, K., et al., 2002. Neighborhood characteristics associated with the location of foodstores and food service places. American Journal of Preventive Medicine, 22 (1), 23–29.

Morton, L.W. and Blanchard, T.C., 2007. Starved for access: life in rural America’s fooddeserts. Rural Realities, 1 (4), 1–10.

Papas, M.A., et al., 2007. The built environment and obesity. Epidemiologic Reviews, 29, 129–143.

Powell, L.M., et al., 2007a. Associations between access to food stores and adolescent bodymass index. American Journal of Preventive Medicine, 33 (4 Suppl), S301–S307.

Powell, L.M., Chaloupka, F.J., and Bao, Y., 2007b. The availability of fast-food and full-service restaurants in the United States: associations with neighborhood characteristics.American Journal of Preventive Medicine, 33 (4 Suppl), S240–S245.

Powell, L.M., et al., 2007c. Food store availability and neighborhood characteristics in theUnited States. Preventive Medicine, 44 (3), 189–195.

Sharkey, J.R. and Horel, S., 2008. Neighborhood socioeconomic deprivation and minoritycomposition are associated with better potential spatial access to the ground-truthed foodenvironment in a large rural area 1,2. The Journal of Nutrition, 138 (3), 620.

Sharkey, J.R. and Horel, S., 2009. Characteristics of potential spatial access to a variety offruits and vegetables in a large rural area. USDA Economic Research Service, and GeraldR. Ford School of Public Health. N. P. Center. Washington D.C., University of Michigan.Available from: http://www.npc.umich.edu/news/events/food-access/sharkey.pdf [Ac-cessed 8 September 2010].

Shaw, H.J., 2006. Food deserts: towards the development of a classification. GeografiskaAnnaler Series B – Human Geography, 88B (2), 231–247.

Slocum, T.A., et al., 2009. Thematic Cartography and Geovisualization (3rd ed.). Upper SaddleRiver, NJ: Pearson/Prentice Hall.

Spielman, S., 2006. Appropriate use of the K function in urban environments. AmericanJournal of Public Health, 96 (2), 205.

White, M., 2007. Food access and obesity. Obesity Reviews, 8 (Suppl 1), 99–107.Zandbergen, P.A., 2007. Influence of geocoding on environmental exposure assessment of

children living near high traffic roads. BMC Public Health, 7 (37), 1–13.Zenk, S.N., et al., 2005a. Fruit and vegetable intake in African Americans income and store

characteristics. American Journal of Preventive Medicine, 29 (1), 1–9.Zenk, S.N., et al., 2005b. Neighborhood racial composition, neighborhood poverty, and the

spatial accessibility of supermarkets in metropolitan Detroit. American Journal of PublicHealth, 95 (4), 660–667.

Geocarto International 499

Dow

nloa

ded

by [

Mic

higa

n St

ate

Uni

vers

ity]

at 1

0:47

12

Janu

ary

2014