Background
Food intake is considered a complex behaviour of multifactorial origin(Reference Story, Kaphingst and Robinson-O’Brien1). A socio-ecological approach to understanding such behaviour is recognised as being a useful framework for integrating the numerous influences present at both the individual and environmental levels(Reference Story, Kaphingst and Robinson-O’Brien1–Reference Kawachi and Berkman5). There is growing interest in the environmental context as related to food behaviour; this includes both the social and physical environment. In this relatively recent field of research, Glanz et al.(Reference Glanz, Sallis and Saelens6, Reference Glanz7) identified different aspects of the food environment. These include the ‘community nutritional environment’ defined as the number, type, location and accessibility of food outlets, and the ‘consumer nutritional environment’ defined by what consumers encounter in and around food outlets (prices, promotions, nutritional quality). In terms of community nutrition, a number of studies and reviews emphasise the influence of spatial accessibility of food upon the relationship between food environment, individual food choice and, ultimately, risk of chronic diseases such as obesity(Reference Booth, Pinkston and Poston3, Reference Glanz, Sallis and Saelens6, Reference Papas, Alberg and Ewing8–Reference White12). An important research issue lies in identifying and describing the different methodological procedures that can be used to specifically assess the spatial accessibility of food outlets.
Various methods, both objective and subjective, have been used to assess variables related to the presence and type of food outlet. Subjective methods include surveys of individual perception of food outlets available to neighbourhood residents(Reference Rose and Richards13, Reference Inglis, Ball and Crawford14). Among objective methods, geographic measures are most frequently used to assess the food environment(Reference McKinnon, Reedy and Morrissette15). Some of these are provided by spatial analysis methods based on geographic information systems (GIS). GIS are computer-based methods and tools, which, via different information sources, enable spatial and thematic data to be organised, managed and combined, and results to be represented and analysed according to geographic location(Reference Longley, Goodchild and Maguire16). Analyses can then be carried out to localise and model potential spatial interactions between the different types of information at hand.
In public health, examples of the use of GIS methods include the analysis of disparities in access to health-care(Reference McLafferty17) and, more recently, the association between built environment and physical activity(Reference Glanz, Sallis and Saelens6, Reference Wendel-Vos, Droomers and Kremers18). Application of GIS to the food environment is relatively new in public health nutrition. Use of a geographic model of analysis may help to identify spatial inequalities in access to food outlets, and in turn, influence policies and incite urban planners to modify the food environment accordingly. In this context, a major challenge lies in ensuring appropriate and effective use of GIS data and spatial analysis methods to measure the food environment(Reference Matthews, Moudon and Daniel19, Reference McKinnon, Reedy and Handy20). Despite the growing use of GIS, we were unable to find a literature review of GIS methods used to assess the food environment. The aim of the present methodological article was to describe GIS methods already in use in this field and to discuss their relevance for increasing our understanding of food environment attributes.
Methods
We sought to identify all studies that used GIS to measure the proximity and/or density of food outlets so as to characterise attributes of the food environment. A search of the literature was conducted with the OVID interface in the following social and health science databases: Medline/Pubmed, PsycINFO, Francis and GeoBase. The search was conducted using different combinations of keywords (in the title or abstract) such as ‘food environment’, ‘food outlets’, ‘access’, ‘availability’ and ‘geographic information system’. The search was restricted to human populations, and studies on both adults and children were included. In addition, the reference sections of the articles included were reviewed. The search was restricted to English language articles published between January 1999 and June 2008.
The main inclusion criterion in the review was the use of GIS-based techniques of spatial analysis to measure the food environment. We excluded studies that relied only on survey participants to characterise the food environment(Reference Rose and Richards13) and articles that used GIS only as a geocoding tool (process assigning geographic coordinates to a point, e.g. street addresses to food outlet) to map various data(Reference Liese, Weis and Pluto21). Data extracted in addition to GIS methods were: study location, scale (e.g. census tract), food environment outcome (e.g. supermarkets, fast food), covariates and main findings.
Results
Study characteristics
An initial search of online databases and of the reference sections of the articles included identified 1070 papers. After preliminary screening based on the title or abstract, thirty-eight full-text papers were retrieved for further assessment. In the final step, twenty-nine articles with GIS-based measurements of the food environment were included in the review. Table 1 summarises data extracted for each paper. Seventeen reviewed studies were conducted in the United States, four in Australia, three in Canada, three in New Zealand and two in England. The selected studies fell into two categories: (i) studies that explore the relationships between characteristics of the food environment and measurements of individual food behaviours; and (ii) studies that compare accessibility of food outlets in different types of neighbourhoods.
Summary descriptive table of the studies included in the review
GIS, geographic information systems; LGA, local government area; SES, socio-economic status; SEIFA, Socio-Economic Index For Areas; TV, television; MESA, Multi-Ethnic Study of Atherosclerosis; AHEI, Alternate Healthy Eating Index; NZDep, New Zealand deprivation index; IRSD; Index of Relative Socio-economic Disadvantage.
Relationship between food environment and individual food behaviour
Among the twenty-nine articles reviewed, eleven (38 %) analysed associations between food environment and individual food behaviours(Reference Bodor, Rose and Farley22–Reference Moore, Diez Roux and Nettleton26), weight status(Reference Burdette and Whitaker27–Reference Liu, Wilson and Qi31) or perceived availability of healthy food(Reference Moore, Diez Roux and Brines32). In those studies, the addresses of respondents were geocoded and used as references for GIS analyses. Four studies were performed on children or teenagers(Reference Timperio, Ball and Roberts24, Reference Burdette and Whitaker27, Reference Jago, Baranowski and Baranowski28, Reference Liu, Wilson and Qi31), and the others on adult populations (one specifically concerned pregnant women(Reference Laraia, Siega-Riz and Kaufman33)). The outcomes of selected studies were consumption of fruits and vegetables(Reference Bodor, Rose and Farley22–Reference Timperio, Ball and Roberts24, Reference Jago, Baranowski and Baranowski28), perception of availability of healthy food(Reference Moore, Diez Roux and Brines32), dietary patterns(Reference Moore, Diez Roux and Nettleton26, Reference Laraia, Siega-Riz and Kaufman33) and prevalence of overweight or obesity(Reference Burdette and Whitaker27, Reference Jeffery, Baxter and McGuire29–Reference Liu, Wilson and Qi31). In most data sets (seven out of eleven), individual characteristics were collected from the year 2000(Reference Bodor, Rose and Farley22–Reference Timperio, Ball and Roberts24, Reference Moore, Diez Roux and Nettleton26, Reference Jago, Baranowski and Baranowski28, Reference Liu, Wilson and Qi31, Reference Moore, Diez Roux and Brines32). In three studies, the date on which food outlet lists were drawn up was not mentioned(Reference Jago, Baranowski and Baranowski28–Reference Wang, Kim and Gonzalez30). In the other studies, the date given for the food outlet list corresponded to the date of collection of individual data (±2 years).
The covariates most frequently used in the analyses included individual demographic and socio-economic characteristics, and individual behaviour such as smoking and physical activity or sedentary behaviour (Table 1).
Spatial access to food outlets according to the type of neighbourhood
The aim of most articles retrieved (eighteen out of twenty-nine; 62 %) was to assess and compare neighbourhoods according to spatial access to food outlets. All these articles considered the neighbourhood as the area of study. However, the scale of the neighbourhoods varied: it involved census tracts and postal sectors in North American studies, wards and postal codes in the United Kingdom and census meshblocks in Australia and New Zealand (Table 1). Most studies were based on census tracts, since they had been conducted in the United States (seventeen out of twenty-nine studies), while four were performed in Australia(Reference Timperio, Ball and Roberts24, Reference Burns and Inglis34–Reference Winkler, Turrell and Patterson36), two in the United Kingdom(Reference Clarke, Eyre and Guy37, Reference Donkin, Dowler and Stevenson38), three in Canadian cities(Reference Apparicio, Cloutier and Shearmur39–Reference Larsen and Gilliland41) and three in New Zealand(Reference Pearce, Hiscock and Blakely23, Reference Pearce, Blakely and Witten42, Reference Pearce, Witten and Bartie43).
Two studies were related to fast food outlets only(Reference Austin, Melly and Sanchez44, Reference Block, Scribner and DeSalvo45), one to fast food and convenience stores(Reference Zenk and Powell46) and one to fast food, full-service restaurants, convenience and grocery stores(Reference Frank, Glanz and McCarron47). The remaining studies focused on a common type of food store: the supermarket. In all of these studies, residential contexts were characterised by socio-economic indicators (including unemployment rates and single-parent rates(Reference Apparicio, Cloutier and Shearmur39, Reference Larsen and Gilliland41), income(Reference Liu, Wilson and Qi31), race/ethnicity(Reference Zenk and Powell46, Reference Baker, Schootman and Barnidge48), households without cars(Reference Clarke, Eyre and Guy37, Reference Block and Kouba49)) and by other information such as degree of commercialisation(Reference Austin, Melly and Sanchez44), urban/rural status(Reference Pearce, Blakely and Witten42, Reference Zenk and Powell46), safety(Reference Burdette and Whitaker27) and neighbourhood walkability(Reference Frank, Glanz and McCarron47) (environmental attributes that encourage walking(Reference Owen, Humpel and Leslie50)). In nine out of eighteen studies, an index of deprivation (constructed from census data) was used to describe the social–residential context(Reference Pearce, Hiscock and Blakely23, Reference Burns and Inglis34, Reference O’Dwyer and Coveney35, Reference Clarke, Eyre and Guy37–Reference Apparicio, Cloutier and Shearmur39, Reference Larsen and Gilliland41, Reference Pearce, Blakely and Witten42, Reference Sharkey and Horel51).
GIS measurements of the food environment
In the articles reviewed, two main notions were used to assess the food environment: density and proximity. (i) Density is usually the number of food outlets (food stores, restaurants) in an administratively defined area (census or postal units) or an area defined by the authors (specific zone). (ii) Proximity is defined between two locations such as respondent address (home, school) and the closest food outlet. It could be measured by a straight-line distance (Euclidean distance) or by travel time (time needed to travel to a food outlet). Table 2 lists the various methods described in the literature concerning the food environment used for assessing density and proximity, along with the number of corresponding studies for each method. Among the twenty-nine studies examined, twelve combined both spatial approaches (Table 2).
Summary measures of food availability used in twenty-nine published articles
GIS, geographic information systems.
Note: Total number of studies was more than twenty-nine because twelve studies combined measurements of density and proximity.
Density
Buffer
The most common GIS approach (eighteen studies out of twenty-nine) was the buffer. This consists of defining a zone around a given location within a specified distance (or shape). The location can be a point (home, school, work or food outlet address), a line (road) or a polygon (neighbourhood).
Most studies defined buffers in order to quantify the availability or accessibility of food outlets. Seven of these studies used a buffer zone around the respondent’s home(Reference Bodor, Rose and Farley22, Reference Timperio, Ball and Roberts24, Reference Laraia, Messer and Kaufman25, Reference Jago, Baranowski and Baranowski28–Reference Liu, Wilson and Qi31), three around the school(Reference Austin, Melly and Sanchez44, Reference Zenk and Powell46, Reference Frank, Glanz and McCarron47), four around the food store(Reference Clarke, Eyre and Guy37, Reference Donkin, Dowler and Stevenson38, Reference Larsen and Gilliland41, Reference Block and Kouba49) and four around the centroid (geometric center) of each neighbourhood(Reference O’Dwyer and Coveney35, Reference Winkler, Turrell and Patterson36, Reference Smoyer-Tomic, Spence and Raine40, Reference Block, Scribner and DeSalvo45). For one of these studies, analyses were performed using buffers around both the home and the work address(Reference Jeffery, Baxter and McGuire29), while only one study combined a buffer around a point (supermarket) or around a line (bus route)(Reference Larsen and Gilliland41). It should be noted that there are two ways to define the shape of a buffer for the GIS user. It can be constructed either by a zone surrounding a location (circular buffer when the given location is a point) or by a zone along the street network (network buffer; e.g. see figures in Frank et al.(Reference Frank, Andresen and Schmid52)).
Circular buffer
In the studies we reviewed, the values used for the radius of a circular buffer were between 100 and 2500 m. Depending on the study, these distances were selected on the basis of estimations of neighbourhood walkability or distances that individuals might be ready to cover to reach food outlets(Reference Timperio, Ball and Roberts24, Reference Jago, Baranowski and Baranowski28, Reference Austin, Melly and Sanchez44, Reference Block, Scribner and DeSalvo45). In a study by Bodor et al.(Reference Bodor, Rose and Farley22), different distances were chosen according to the type of food store: 100 m for small food stores (e.g. the approximate size of a city block) and 1000 m for large supermarkets. Two authors(Reference O’Dwyer and Coveney35, Reference Winkler, Turrell and Patterson36) used a much wider radius of 2500 m around the geometric centre of the neighbourhood to define the area in which residents were likely to shop.
Network buffer
A network buffer can be defined as being based on the accessibility of food outlets via the mode of transportation used and the type of destination. Larsen and Gilliland(Reference Larsen and Gilliland41) used two network buffers in the town of London (Ontario, Canada). The first buffer was based on a distance of 1000 m by foot around each supermarket. The second buffer was created around each bus route to estimate a 500 m network service line area with public transport access to supermarkets.
Kernel density estimation
Kernel density is a spatial smoothing method employed to transform a sample of geographically referenced point data (e.g. address of food outlet) into a smooth continuous surface(Reference Bailey and Gatrell53–Reference Portnov, Dubnov and Barchana56). As described by Kloog et al.(Reference Kloog, Haim and Portnov57), the general principles of this statistical technique are to estimate the ‘intensity of referenced points across a surface, by calculating the overall number of cases situated within a given search radius from a target point’. A distance function is introduced in the calculation so that ‘points lying near the centre of the search area are weighted more heavily than those lying near the edge’(Reference Kloog, Haim and Portnov57). The various steps for generating kernel densities with GIS software have been described by Guagliardo(Reference Guagliardo58).
Only two studies, both by Moore et al.(Reference Moore, Diez Roux and Nettleton26, Reference Moore, Diez Roux and Brines32), used kernel density estimation to assess the spatial distribution of food outlets (Table 2). In that case, the aim was to create a smooth map of food store density per square mile where the home location proximity was emphasised and more weight was put on closer outlets.
Spatial clustering
A spatial scan statistic is used to assess whether events are randomly distributed within the study area, and if not, to identify significant spatial clusters(Reference Ozdenerol, Williams and Kang59, Reference Alves de Souza, Da Silva-Nunes and Dos Santos Malafronte60). This method consists of creating moving windows of various shapes (circles, squares) and sizes (radius, sides of square). These windows are moved systematically across the map, which enables assessment of the likelihood that events are more prevalent inside than outside that given window (see SatScan process(Reference Kulldorff and Nagawalla61)). With this method, Baker et al.(Reference Baker, Schootman and Barnidge48) identified spatial neighbourhood variation in the rate of supermarkets and fast food outlets in St Louis, MO, United States, and observed clusters of food supermarkets and fast food outlets (i.e. areas with higher or lower rates than expected).
Network analysis and proximity measures
Proximity defined as a distance
Several types of distances are typically used to assess proximity with GIS: Euclidean distance (straight line distance), Manhattan (city block distance) and network distance. The Manhattan distance corresponds to the distance between two points measured along axes at right angles(Reference Apparicio, Abdelmajid and Riva62). In other words, Manhattan distance represents an approximate distance close to a street map and is mainly used on squared city maps.
In our review, six studies measured the distance between home/school and food outlets via the Euclidean distance(Reference Bodor, Rose and Farley22, Reference Jago, Baranowski and Baranowski28, Reference Wang, Kim and Gonzalez30, Reference Laraia, Siega-Riz and Kaufman33, Reference Apparicio, Cloutier and Shearmur39, Reference Austin, Melly and Sanchez44) (Table 2). In Eastside Detroit areas with no supermarket, Zenk et al.(Reference Zenk, Schulz and Israel63) used the Manhattan distance to evaluate the shortest distance between home addresses and food outlets in a population of African-American women. Two studies used network distance by road(Reference Timperio, Ball and Roberts24, Reference Zenk, Schulz and Israel63). In other studies, the network distance by street travel was used to evaluate the minimum distance residents must walk from their home/school to the closest food outlet(Reference Burdette and Whitaker27, Reference Liu, Wilson and Qi31, Reference Donkin, Dowler and Stevenson38, Reference Smoyer-Tomic, Spence and Raine40, Reference Larsen and Gilliland41, Reference Frank, Glanz and McCarron47).
Proximity measured by travel time
The travel time between a given place (e.g. school or home address) and the address of a food outlet can be calculated by GIS according to the means of transport and the specificities of the network. Four out of twenty-nine studies used travel time as a proximity measurement (Table 1)(Reference Pearce, Hiscock and Blakely23, Reference Burns and Inglis34, Reference Pearce, Blakely and Witten42, Reference Pearce, Witten and Bartie43). Burns and Inglis(Reference Burns and Inglis34) developed a travel time model between home, fast food outlet and supermarket according to a number of variables including means of transport (car, bus, on foot), type of road (speed limit), topography (barriers as rivers or railway lines) and other characteristics of the public transport network (i.e. frequency of buses). Travel time for each type of transport was compared between underprivileged and privileged neighbourhoods, with the latter having better access to supermarkets.
Discussion
In this review, we investigated which GIS methods have been used to define the food environment and the types of spatial measurements they generate. We found twenty-nine articles that reported GIS methods for measuring spatial accessibility of food outlets as a key feature of the local food environment. We identified two main types of spatial measures to quantify the food environment: density and proximity. The density approach quantifies the availability of food outlets using the buffer method, kernel density estimation or spatial clustering. The proximity approach assesses the distance to food outlets by measurements of distance or travel times. Numerous studies combined both approaches.
How do GIS methods contribute to research on the food environment?
It is clear from the present work that the number of studies that include geographic measurements of density and/or proximity of food outlets as operational variables in the food environment have increased rapidly in recent years. Twenty-two of the twenty-nine articles examined here were published between January 2006 and June 2008. It is likely that the continuous refinement of GIS software and the increased availability of precisely geocoded databases have contributed substantially, and will continue to contribute, to this increase(Reference Matthews, Moudon and Daniel19).
In the studies included in this review, two approaches based on GIS methods were used to characterise the local food environment. One involved assessing the number of food outlets in an area (density) and the other assessed proximity to facilities. Interestingly, a large number of studies combined both approaches. Indeed, as argued by Apparicio et al.(Reference Apparicio, Cloutier and Shearmur39), a single measure of access cannot fully describe accessibility of food outlets. Focusing on the issue of ‘food desert’ (areas characterised by relatively poor access to healthy and affordable food(Reference Beaulac, Kristjansson and Cummins64)), Apparicio et al.(Reference Apparicio, Cloutier and Shearmur39) proposed a methodology based on three measurements of access using the shortest network distance: diversity, proximity and variety (average distance to the three closest different chain-name supermarkets).
An important advantage of the GIS approach is that it enables assessment of spatial variations in prevalence independently of administrative boundaries(Reference Chaix, Merlo and Chauvin65). Many phenomena are continuously distributed over space and are independent of arbitrarily defined boundaries(Reference Chaix, Merlo and Subramanian66, Reference Matthews, Detwiler and Burton67). Estimating the density of food outlets within buffers, or by means of kernel density estimation rather than administrative area, enables one to take into account the fact that individuals often cross the boundaries of their residential area to go shopping. However, it should be emphasised that the appropriate size of the area around the place of residence to be defined as the neighbourhood remains subject to debate(Reference Spielman and Yoo68–Reference Chaix, Merlo and Evans70). The choice of this area size is based on assumptions concerning the geographic zone that includes food environment elements influencing food behaviour. In the studies reviewed here, the distance used to define the residential area varied depending on different criteria such as the age of the respondent, type of food outlet and type of transportation. It is also important to underline that few studies exist which question individuals as to the distance they would be prepared to cover for food needs. Thus, because of the complexity of the relationship between environment and behaviour, defining the size of the neighbourhood in which this relationship operates remains a challenging methodological issue(Reference Spielman and Yoo68, Reference Brownson, Hoehner and Day71).
GIS methods enable the modelling of proximity to food outlets using metric distance and travel time to food outlets. In general, modelling of travel time using the GIS leads to more realistic measurements (taking into account speed limit, topography and network complexity) than the usual mathematical distances, particularly at the local level in sub-metropolitan areas(Reference Apparicio, Abdelmajid and Riva62) or in rural areas(Reference Lovett, Haynes and Sunnenberg72). However, the use of this travel time model, which requires spatial information, is more complex than calculating the mathematical distances between two points.
In the articles that we reviewed, which used travel time to food outlets, the car was the type of transportation evaluated in four papers, with public transport evaluated in only one(Reference Burns and Inglis34). None dealt with travel time by foot or ‘mixed’ travel. This is an important point because families with low income may not own a car or even have access to public transportation. In future studies, a methodological challenge therefore lies in measuring travel time from the respondent’s address to food outlets according to the different types of transport available (car, public transport, or on foot). In addition, modelling travel time according to public transport or on foot requires more sophisticated GIS modelling than private car transport(Reference Martin, Wrigley and Barnett73).
On the other hand, Larsen et al.(Reference Larsen and Gilliland41) showed that, with the GIS, the geographic distribution of supermarkets has changed over time, thus influencing the relationship between people and places in a spatial access approach. Through GIS use, it is possible to capture the temporal changes in localisation of food outlets and land use, which will improve our understanding of the relationship between food environment and food behaviour over time(Reference Matthews, Moudon and Daniel19, Reference Burgoine, Lake and Stamp74).
One of the major challenges when using GIS for studying the food environment concerns the quality of the data available. The validity of GIS-based measures of environmental features of the food environment has recently been discussed(Reference Matthews, Moudon and Daniel19). Since street addresses of facilities were often obtained from commercial databases or had been collected for other purposes, data accuracy and comprehensiveness must be viewed with caution(Reference Brownson, Hoehner and Day71, Reference Boone, Gordon-Larsen and Stewart75). In addition, there may exist a mismatch between the geocoded location of a facility and its true location, e.g. via the GPS (global positioning system) technique(Reference Boone, Gordon-Larsen and Stewart75, Reference Porter, Kirtland and Neet76).
A major challenge: which concepts should be used to characterise access?
The articles reviewed here focused on spatial access as estimated by GIS methods. Nevertheless, it should be noted that few authors specifically use the term ‘spatial’ or ‘geographic’ when dealing with the broad concept of access(Reference Apparicio, Cloutier and Shearmur39, Reference Sharkey and Horel51). Access that includes material and social dimensions is a complex notion, and geographic proximity does not systematically imply accessibility. Gould(Reference Gould77) describes accessibility as ‘a notion difficult to grasp… one of these common terms everybody uses until the problem arises of defining and measuring the concept’. Penchansky and Thomas(Reference Penchansky and William Thomas78) defined five dimensions for access, including availability, accessibility, affordability, acceptability and accommodation. Only the first two dimensions, corresponding to spatial measures, reflect the geographic distribution (e.g. of facilities around the home address) and can be estimated by GIS methods. This may be viewed as a possible weakness of these methods. However, by definition, the other dimensions reflecting the cultural, social and economic factors are not taken into account.
The ‘ideal’ study of access to food outlets would appear to be one that associates all dimensions related to accessibility: proximity, diversity, availability, affordability (cost) and perception, with the term ‘diversity’ referring to the types of food outlets and ‘availability’ referring to the food supply at the food outlets. Only four of the articles(Reference Bodor, Rose and Farley22, Reference Frank, Glanz and McCarron47–Reference Block and Kouba49) combined assessment of spatial access to food outlets with an evaluation of the actual food content of the outlet. Among those articles, only two took into account cost and quality(Reference Frank, Glanz and McCarron47, Reference Block and Kouba49) in addition to the availability of foods, especially healthy foods. Access to food outlets may also be limited by the subject’s perception of the environment in his/her neigbourhood(Reference Moore, Diez Roux and Brines32, Reference Kamphuis, van Lenthe and Giskes79). Moore et al.(Reference Moore, Diez Roux and Brines32) suggested that the availability of healthy foods as reported by residents (perception) and their availability as measured by GIS application (density) provide complementary information for characterising the local food environment. In other words, methodology for conducting an ‘ideal’ research study would have to combine GIS potential and survey approaches to describe both spatial and social accessibility of healthy foods.
Conclusions
Accessibility to services and facilities and, in particular, to healthy food, is an important social equity issue(Reference Apparicio, Cloutier and Shearmur39). Geographic analysis models may provide local authorities and policy makers with new views and possibilities for making decisions as to the location of services in order to offer a fair choice to the entire population. For example, Banos et al.(Reference Banos and Huguenin-Richard54, Reference Banos and Banos80) have designed a GIS application that identifies hot spots by spatial regression(Reference Anselin, Syabri and Kho81). These results enabled the targeting of parts of the road network that needed modifications(Reference Banos and Banos80). Gatrell and Naumann(Reference Gatrell and Naumann82) adapted this tool to the field of health-care and suggested potential sites for building new hospitals, with various scenarios being examined according to traffic density.
It should also be noted that spatial accessibility of healthy food is only one of the multiple determinants of a healthy lifestyle, as emphasised by socio-ecological models of behaviours(Reference Sallis and Glanz2–Reference Townshend and Lake4). Further development of spatial analysis methods should help to better define its importance in various settings(Reference Cummins83). On the basis of the articles reviewed here, we suggest two avenues for future methodological research when analysing accessibility of facilities relevant to food behaviour. First, there is a need to test and compare more sophisticated spatial GIS modelling such as travel time or potential model principles and gravity models(Reference Guagliardo58, Reference Weber and Hirsch84). The latter combine diversity (type of facilities) and accessibility by using distribution of facilities throughout the area, together with a distance function to calculate the attractiveness of a food outlet (catchment area). Second, future research should benefit from a combination of GIS methods and survey approaches to describe both spatial and social food outlet accessibility, and to better understand how the food environment influences food behaviour and health.
Acknowledgements
This work is part of the ELIANE (Environmental LInks to physical Activity, Nutrition and hEalth) study. ELIANE is a project supported by the French National Research Agency (Agence Nationale de la Recherche, ANR-07-PNRA-004). The authors declare that they have no competing interests. H.C. designed the study, performed the literature search and data extraction, and drafted the manuscript. J.-M.O. supervised the study design and data collection, and contributed to the finalisation of the paper. R.C., P.S., C.S., B.C., A.B., D.B. and C.W. assisted with the literature search and the writing of the manuscript. J.-M.O. is the coordinator of the ELIANE study; C.S., B.C. and C.W. are the principal investigators in the ELIANE study. All the authors read and approved the final version of the manuscript.
