The Canadian government prioritized the right to food in 1998 in response to the 1996 World Food Summit( 1 ) and since 1948 has signed many agreements emphasizing food as a human right( Reference Rideout, Riches and Ostry 2 ). Yet in 2007–2008, 7·7 % of Canadian households (almost one million) were food insecure( 3 ); a concerning public health problem that is echoed in other high-income countries such as the USA( Reference Coleman-Jensen, Nord and Andrews 4 ) and Australia( Reference Rosier 5 ). In the USA, for instance, one of the richest countries in the world, 14·5 % of American households (17·2 million) were food insecure at some point during 2010. On average, these households experienced food insecurity for seven out of the twelve months of the year( Reference Coleman-Jensen, Nord and Andrews 4 ).
Food insecurity exists when there is limited or uncertain access to nutritionally adequate and safe foods or limited or uncertain ability to acquire acceptable foods in socially acceptable ways( Reference Andersen 6 ). There are different stages of severity, starting with not being able to buy and eat what one would like due to income-related resource constraints( Reference Tarasuk 7 ). This incorporates issues of food quality including variety, safety and nutrient content. The next stage involves a decrease in quantity and attempts to make food last until there is money to buy more. Decreases in food quantity may then lead to the physical sensation of hunger( Reference Tarasuk 7 ). Finally, the most severe stage is the complete absence of food intake. In addition to quality and quantity elements, psychological distress (e.g. worry or concern about not having enough to eat) and social–familial perturbations (e.g. resorting to socially unacceptable ways of getting food) are also important dimensions of food insecurity( Reference Radimer 8 – Reference Radimer, Olson and Campbell 10 ).
Studies, conducted for the most part in the USA, have uncovered health correlates of food insecurity that include increased depression( Reference Heflin, Siefert and Williams 11 ) and nutritional inadequacies( Reference Kirkpatrick and Tarasuk 12 , Reference Rose and Oliveira 13 ) among adults, and psychosocial( Reference Slopen, Fitzmaurice and Williams 14 , Reference Alaimo, Olson and Frongillo 15 ) and physical( Reference Kirkpatrick, McIntyre and Potestio 16 , Reference Casey, Szeto and Robbins 17 ) developmental problems among children. Food insecurity has been found to relate to overweight or obesity; however, results are more consistent among women than among men and children( Reference Larson and Story 18 , Reference Eisenmann, Gundersen and Lohman 19 ). There is also some evidence that food insecurity is related to CVD risk factors( Reference Parker, Widome and Nettleton 20 , Reference Seligman, Laraia and Kushel 21 ). And finally, those dealing with food insecurity may also be more likely to postpone needed medications and medical care, in order to make sure that basic needs are addressed first( Reference Kushel, Gupta and Gee 22 ). Since most of these studies are based on the US population, correlates may be different in different countries. This may be especially true for postponing needed medications and medical care, as health-care systems differ among high-income countries.
The major determinant of food insecurity is well understood to be a lack of financial resources( Reference Rose 23 ). Active public policies to decrease poverty and protect the vulnerable non-poor are major ways society can meet the needs of citizens( Reference Barrett 24 ); yet, in countries such as Canada the social safety net is shrinking( Reference Rideout, Riches and Ostry 2 ) and more and more Canadians are relying on food banks. In 2010, food bank use across Canada was the highest on record, and did not decrease in 2011( 25 ). This is concerning for Canadians, especially when taking into consideration the current world economic outlook, high food prices and high food price volatility( 26 ).
Considering multiple levels of potential societal influence is consistent with a social-ecological approach to understanding public health problems. The link between macro socio-political and individual-level economic factors and food insecurity is relatively well established. What is less well known is how attributes of the local environment may be implicated. A focus on developing an understanding of the local environment in relation to food insecurity therefore deepens our overall understanding of this complex public health problem.
Recent research attempting to elucidate the effects of local social and physical characteristics of place on physical activity, diet and weight status of community residents may be relevant for identifying factors that can promote or prevent food insecurity (see Leal and Chaix( Reference Leal and Chaix 27 ) for an example). In the USA, there is convincing evidence of the existence of food deserts: area-level disparities in affordable, healthy food access by income and race( Reference Beaulac, Kristjansson and Cummins 28 ), which could translate into increased food insecurity for area residents. Availability of supermarkets has been linked to healthier eating( Reference Moore, Diez Roux and Nettleton 29 ) and supermarkets have also been shown to have lower food prices compared with smaller stores( Reference Drouin, Hamelin and Ouellet 30 ). Food quality may also vary depending on affluence of the area and living location( Reference Cummins, Smith and Taylor 31 ). Areas with high social capital – in particular, trust, reciprocity, shared norms and the willingness to enforce these norms – may allow residents to obtain food from neighbours or other institutions more easily in times of need and to act collectively to address food insecurity issues( Reference Kawachi, Subramanian and Kim 32 ). Disintegrating aesthetics of an area due to anti-social activity may act to dissuade food service establishments and other institutional supports from locating in particular areas, and perceived danger may prevent residents from accessing nearby food resources( Reference Macintyre, Ellaway and Cummins 33 ).
Finally, there is some evidence from the USA to show that community-based initiatives such as community gardens( 34 ) may increase the consumption of fruits and vegetables among disadvantaged participants. Thus, interventions or programmes that are place-based may increase the availability, accessibility and utilization of food to local residents and therefore work to decrease individual/household food insecurity.
The purpose of the present paper, therefore, was to conduct a comprehensive and critical review of the published literature in order to shore up the knowledge base on place and food insecurity. The intent was not to conduct a systematic review, rather to undertake a review which may serve to inform future reviews and identify research gaps for further study. The specific research review question was: among experimental and observational studies, have local physical and social environmental factors been found to significantly relate to individual/household-level food insecurity in the general population?
Methods
Eligibility criteria
Research studies were considered for inclusion if they examined the relationship between features of place and self-reported food insecurity (either at the household/family or individual level). The target population for the review included non-institutionalized individuals in the general population, but did not exclude studies that focused specifically on demographic subgroups. Only those studies that examined populations living in countries: (i) with a democratic political system and (ii) defined as ‘very high human development’ by the UN( 35 ) were included.
‘Place’ was broadly defined as having a spatial or area component beyond an individual's residence; although the scale of place considered in the review ranged from the street to the county level. Studies examining larger areas were excluded. Place predictors could be perceived by individuals or objectively measured. Specific measures of interest included social capital (various definitions), crime, safety, density/distance to food stores, quality and prices at these local food stores, population/residential density, socio-economic status (SES)/deprivation and local infrastructure making access to food easier (e.g. availability of public transportation, well-maintained sidewalks, route connectivity and directness, traffic, etc.). The outcome, food insecurity, had to be self-reported by the participating individual, on behalf of him- or herself, or on behalf of the household, and measured by a survey/questionnaire touching on one or more dimensions of not getting enough to eat due to lack of financial resources (e.g. food quality and variety, food quantity, physical hunger, anxiety/psychological distress, social aspects affected such as stealing or not inviting people over for dinner). Studies that used proxy measures for food insecurity, such as food stamp use, poverty status, food bank or pantry use, were excluded. Those that assessed community food insecurity as the outcome were also excluded.
Individual sociodemographic/socio-economic characteristics may act as confounders in the place–food insecurity relationship; thus, studies that did not adjust for some measure of household or individual SES were excluded. Considering that studies had to conduct multivariate analysis, those that had small sample sizes (n < 100) were excluded. Only primary studies and reviews that used systematic search methods were considered for inclusion.
Study designs could be observational, as long as there was a comparison group or groups that was either not exposed to the place predictor or had varying levels of exposure. Intervention studies of different programmes/initiatives were included if they were newly implemented or modified, local in scope and delivered on a community-wide scale, helped to increase food availability and accessibility, and used (at the very least) a before-and-after design. Qualitative studies were excluded. Finally, due to resource and time constraints articles that were not written in French or English, or not published in peer-reviewed scientific journals, were excluded.
Search strategy and identification of studies
Six electronic databases were searched in December 2011 using a systematic search strategy. MEDLINE In-process and Other Non-indexed Citations, MEDLINE, EMBASE and PsychINFO were searched using the OvidSP interface, while Social Services Abstracts and Sociological Abstracts were searched using the ProQuest interface. No restrictions on time period were imposed. The search strategy was developed first in OvidSP and then refined as appropriate in ProQuest to account for changes in indexing of subheadings. Both free-text terms and indexed subheadings were used. The full search strategy for MEDLINE is available in the Supplementary Materials.
In OvidSP, all four databases were search simultaneously and duplicates removed. Citations were saved and imported directly into Reference Manager version 12. Given ProQuest's difficulty in downloading a large number of citations, the two social databases (Social Services Abstracts and Sociological Abstracts) were searched separately; citations were handled in the same way as OvidSP. All studies downloaded into Reference Manager first underwent a duplicate search. The first screen involved examining the titles and abstracts to determine if studies met eligibility criteria. Full-text articles were then obtained for those studies that appeared to meet the eligibility criteria and for those where it was unclear. Eligibility was assessed again, based on information in the entire article. Hand-searching the reference lists of included studies was also undertaken in order to retrieve studies missed from the original search strategy. Reviews were included solely for this purpose, as data from the review itself were not collected.
Data abstraction
Details on the study characteristics were abstracted to provide a summary of implementation and results and a critical overview of study quality. Information abstracted included study design (e.g. cross-sectional, ecological, case–control, cohort, experimental/intervention), sampling strategy, survey/questionnaire administration mode (e.g. face-to-face interviews, telephone interviews, self-administered questionnaires), level of analysis (e.g. individual level or multilevel), sample characteristics (e.g. age, country, ethnicity, income level), total and effective/analytic sample sizes, cooperation and/or response rate(s), definition of the place characteristic as well as the area described by the characteristic, definition of food insecurity, statistical method used and the confounders included, results, and other unique aspects of the study that warranted special attention, specifically in regard to potential limitations.
Results
Literature search and general overview
A total of 2502 potential articles were retrieved from the six databases. Of these, eighteen primary studies, and one review( Reference Gorton, Bullen and Mhurchu 36 ), met the eligibility criteria (Fig. 1). Four of the included studies resulted from hand-searching the reference lists of other included studies( Reference Garasky, Morton and Greder 37 – Reference Bartfeld and Dunifon 40 ), while two studies were included based on prior knowledge using other search tools and journals (these had not yet been indexed in Ovid's MEDLINE, although they have been published subsequently)( Reference Chung, Gallo and Giunta 41 , Reference Pilgrim, Barker and Jackson 42 ).
Flow diagram of study selection
Almost three-quarters (13/18) of the primary studies included were conducted in the USA, three were conducted in Australia, one in Canada and one in the UK (Table 1). All except one study were conducted in the year 2000 or later. The cross-sectional research design was almost exclusively used. Sample sizes across studies ranged from 330 to 70 942. Only four studies conducted multilevel analyses, while the rest conducted analyses at the individual level. For the most part, the areas defining the place characteristics were either based on administrative boundaries or perceived by the respondent.
Characteristics of studies included in the present review
XS, cross-sectional design; I, individual level; TQ, questionnaire/survey administered over telephone; CS, convenience sample; SAQ, self-administered paper questionnaire; ML, multilevel; RS, simple random sample; nhood, neighbourhood; SRS, stratified random sample; CRS, cluster random sample; F2F, face-to-face interviews; NHANES III, Third US National Health and Nutrition Examination Survey (1998–1994); PC, prospective cohort; n, number of participants; N, total number of sample; RR, response rate (# responding/# eligible); CR, cooperation rate (# responding/# eligible and able to contact); MSA, Metropolitan Statistical Area; Ref, reference category; def, definition; USDA, US Department of Agriculture; #, number; +, a significant positive/proportionate association; Ø, a null association; −, a significant negative/inverse association; NG, not generalizable; RDD, random-digit dialling; SES, socio-economic status.
In terms of characteristics of the target populations studied, seven of the eighteen studies examined adults of varying ages( Reference Garasky, Morton and Greder 37 – Reference Morton, Bitto and Oakland 39 , Reference Dean and Sharkey 43 – Reference Brisson and Altschul 46 ), seven explicitly examined families with children (sometimes with data collected using the child as the sampling unit)( Reference Bartfeld and Dunifon 40 , Reference Pilgrim, Barker and Jackson 42 , Reference Bartfeld, Ryu and Wang 47 – Reference Bartfeld and Ahn 51 ), three focused on seniors( Reference Chung, Gallo and Giunta 41 , Reference Quine and Morrell 52 , Reference Lee and Frongillo 53 ), and one sampled a range of ages (children and adults)( Reference Foley, Ward and Carter 54 ). Seven out of the eighteen studies (39 %) focused exclusively on low-income or ethnic subgroups( Reference Morton, Bitto and Oakland 39 , Reference Martin, Rogers and Cook 44 , Reference Brisson and Altschul 46 , Reference Kirkpatrick and Tarasuk 48 – Reference Bartfeld and Ahn 51 ).
Almost all included studies used either validated tools to measure food insecurity or individual questions from validated tools. Four studies used the US Department of Agriculture's (USDA) eighteen-item Food Security Scale( Reference Bartfeld and Dunifon 40 , Reference Martin, Rogers and Cook 44 , Reference Kirkpatrick and Tarasuk 48 , Reference Bartfeld and Ahn 51 ), although Kirkpatrick and Tarasuk (2010)( Reference Kirkpatrick and Tarasuk 48 ) applied Health Canada's thresholds to define moderate and severe food insecurity. Five studies used the six-item short form of the USDA Food Security Scale( Reference Garasky, Morton and Greder 37 – Reference Morton, Bitto and Oakland 39 , Reference Pilgrim, Barker and Jackson 42 , Reference Bartfeld, Ryu and Wang 47 ), three studies used items from the Community Childhood Hunger Identification Project in the National Health and Nutrition Examination (NHANES) III surveys( Reference Brisson and Altschul 46 , Reference Mazur, Marquis and Jensen 50 , Reference Lee and Frongillo 53 ), and one study was based on items administered in the Behavioral Risk Factor Surveillance System( Reference Chung, Gallo and Giunta 41 ). The three Australian studies( Reference Radimer, Allsopp and Harvey 45 , Reference Quine and Morrell 52 , Reference Foley, Ward and Carter 54 ) and Dean and Sharkey (2011)( Reference Dean and Sharkey 43 ) adapted and used items from the Radimer/Cornell measure, while Sharkey et al. (2011) used an adapted version of the complete Radimer/Cornell measure( Reference Sharkey, Dean and Johnson 49 ). Eight studies investigated different types or degrees of severity of food insecurity, by using either more than one questionnaire item or different thresholds for the USDA eighteen-item Food Security Scale or the Radimer/Cornell Scale( Reference Chung, Gallo and Giunta 41 , Reference Dean and Sharkey 43 – Reference Radimer, Allsopp and Harvey 45 , Reference Kirkpatrick and Tarasuk 48 – Reference Bartfeld and Ahn 51 ).
Four general types of place factors emerged from the synthesis. For brevity and integration of findings, results of included studies are discussed under the relevant subheadings below. Table 1 details the characteristics of each included study.
Living location
The most common place characteristic examined was living location, as measured on the urban–rural continuum; eleven studies assessed the potential impact of this area-level exposure on food insecurity( Reference Garasky, Morton and Greder 37 – Reference Bartfeld and Dunifon 40 , Reference Dean and Sharkey 43 , Reference Radimer, Allsopp and Harvey 45 , Reference Bartfeld, Ryu and Wang 47 , Reference Mazur, Marquis and Jensen 50 – Reference Lee and Frongillo 53 ). Most often, this place factor was simply defined as urban v. rural with no clear explanation of the administrative boundaries or criteria used.
Seven studies uncovered a potential protective effect of rural living on food insecurity. Garasky et al. (2006) compared two subject-perceived definitions of rural living to urban living. They found that living outside the city on a farm was related to decreased food insecurity compared with living within the city limits, but there was no difference between living outside the city (not on a farm) and living within the city limits( Reference Garasky, Morton and Greder 37 ). In a study of Oregon adults, the percentage of the county considered rural was inversely related to food insecurity( Reference Bernell, Weber and Edwards 38 ). Similarly, the percentage of people living in urban areas as defined by zip code was positively related to food insecurity in a study of Wisconsin families( Reference Bartfeld, Ryu and Wang 47 ). When examining two types of food insecurity (household and individual) based on two questions, Radimer et al. (1997) found that urban living in Australia was associated with increased odds for both types of food insecurity, v. rural living( Reference Radimer, Allsopp and Harvey 45 ). Among a Hispanic population, living in a non-metropolitan area relative to a metropolitan area was inversely related to food insecurity, as defined by cutting of children's or adult's meals, but not by an individual child feeling like he/she does not have enough to eat( Reference Mazur, Marquis and Jensen 50 ). In a study of low-income families, Bartfeld and Ahn (2011) compared various types of towns/cities with respect to population size, defined by zip code( Reference Bartfeld and Ahn 51 ). Compared with rural areas outside MSA (Metropolitan Statistical Areas), low-income families living in small towns, mid-sized suburbs or mid-sized cities were more likely to be food insecure. Those living in large cities, large suburbs, large towns or rural areas within an MSA were no different from those living in rural areas outside an MSA. Finally, in a large, nationally representative sample of US families, living in a central city v. an ‘other’ metropolitan area was associated with increased odds of food insecurity, while living in a rural area was associated with decreased odds( Reference Bartfeld and Dunifon 40 ).
Three of the eleven studies reported null results, whereas one reported a positive association between rural living and food insecurity. Two of the three null studies were conducted on older adults( Reference Quine and Morrell 52 , Reference Lee and Frongillo 53 ). One of these used dichotomous indicators of living location (metropolitan v. non metropolitan living)( Reference Lee and Frongillo 53 ), while the other examined four different types of areas( Reference Quine and Morrell 52 ). The third null study did not find that living in town differed from living in the countryside in two high-poverty Iowa counties with sample mean age of 56 years( Reference Morton, Bitto and Oakland 39 ). In contrast to the seven studies described above, Dean and Sharkey (2011) uncovered a positive association between rural living and food insecurity( Reference Dean and Sharkey 43 ). They conducted multinomial analysis with the response categories serving as an indication of the severity of food insecurity. A significant association was seen only for those in the most severe response category.
Social environment
The nature of social interactions within residential areas was the second most studied place factor in the current review (eight out of eighteen studies), and included various measures of social capital, such as social cohesion, informal social control, collective efficacy, civic structure, and related measures such as religious affiliation, residential mobility and neighbourhood safety( Reference Bernell, Weber and Edwards 38 , Reference Morton, Bitto and Oakland 39 , Reference Chung, Gallo and Giunta 41 , Reference Dean and Sharkey 43 , Reference Martin, Rogers and Cook 44 , Reference Brisson and Altschul 46 , Reference Kirkpatrick and Tarasuk 48 , Reference Foley, Ward and Carter 54 ). Two studies examined characteristics of the social environment at both the individual and neighbourhood levels( Reference Martin, Rogers and Cook 44 , Reference Brisson and Altschul 46 ), while the remainder focused solely on the individual level. Both of the former studies aggregated individual measures up to the neighbourhood level based on specified boundaries and focused on low-income populations. Brisson and Altschul (2011) examined collective efficacy as measured by indices for social cohesion and informal social control. Social cohesion, but not informal social control, was inversely related to food insecurity at the neighbourhood level( Reference Brisson and Altschul 46 ). In the second study, scoring high on a social capital index was found to inversely relate to severe food insecurity (hunger present) but not food insecurity without hunger (of which the authors did not report the results)( Reference Martin, Rogers and Cook 44 ).
At the individual level, four out of the eight studies estimated a potential protective effect of high social capital( Reference Morton, Bitto and Oakland 39 , Reference Dean and Sharkey 43 , Reference Martin, Rogers and Cook 44 , Reference Brisson and Altschul 46 ). Increasing score on a civic structure index was related to decreased odds for food insecurity in a study of two rural, high-poverty counties( Reference Morton, Bitto and Oakland 39 ). And in another study, scoring low or medium on a social capital index was related to increased odds for ‘sometimes’ being food insecure v. food secure, as compared with high social capital, in a mostly rural population of adults( Reference Dean and Sharkey 43 ). Interestingly, low social capital, but not medium, was related to being ‘often’ food insecure( Reference Dean and Sharkey 43 ). Martin et al. (2004) and Brisson and Altschul (2011) estimated similar associations at the individual level as at the neighbourhood level( Reference Martin, Rogers and Cook 44 , Reference Brisson and Altschul 46 ).
The four null studies contained a variety of measures. For example, Foley et al. (2010) examined single items measuring trust in neighbours and safety of the neighbourhood( Reference Foley, Ward and Carter 54 ); Bernell et al. (2006) examined the proportion of the county having a religious affiliation, as well as the proportion that moved within the last five years( Reference Bernell, Weber and Edwards 38 ); Chung et al. (2011) included indices of neighbourhood safety and cohesion for three different food insecurity definitions( Reference Chung, Gallo and Giunta 41 ); and finally, Kirkpatrick and Tarasuk (2010) did not find an association between a social capital index and two types of food insecurity severity( Reference Kirkpatrick and Tarasuk 48 ). Population characteristics of these studies were similarly mixed.
Food environment
Five studies investigated the potential influence of characteristics in the local food environment that could impact on food insecurity( Reference Garasky, Morton and Greder 37 , Reference Chung, Gallo and Giunta 41 , Reference Bartfeld, Ryu and Wang 47 – Reference Sharkey, Dean and Johnson 49 ); although only three uncovered significant associations. In their study of Wisconsin families with children, Bartfeld et al. (2010) estimated a positive association between living 15–22 miles from the nearest supermarket or grocery store v. less than 2 miles( Reference Bartfeld, Ryu and Wang 47 ). However, distances longer than 2 miles, but shorter than 15 miles, were not related to being food insecure. Similarly, distance to the main store for purchasing groceries increased the odds for adult food insecurity, but not household or child food insecurity, in a study of Spanish-speaking women of Mexican origin living in a poor area of Texas( Reference Sharkey, Dean and Johnson 49 ). This same study also found that those who perceived little variety in the types of foods that could be purchased in local stores were more likely to report child food insecurity, but not adult or household food insecurity (although the odds ratios were high and in the same direction for both types). In terms of features of the built environment that could improve access to food, Chung et al. (2011) estimated an inverse association between ‘walkability’ of older adults’ neighbourhoods and two measures of food insecurity (concern about not having enough to eat; hungry because could not get out to buy food) and marginally with a third (eating less because of lack of money, P = 0·056)( Reference Chung, Gallo and Giunta 41 ). In a similar vein, Bartfeld et al. (2010) found that household-perceived access to public transit decreased the odds for food insecurity( Reference Bartfeld, Ryu and Wang 47 ).
Two of the five studies were unable to detect any significant effects of the food environment( Reference Garasky, Morton and Greder 37 , Reference Kirkpatrick and Tarasuk 48 ). In a low-income population, Kirkpatrick and Tarasuk (2010) investigated the association between food insecurity and living within 2 km of a number of food resources, including discount supermarkets, food banks, community kitchens and community gardens( Reference Kirkpatrick and Tarasuk 48 ). No significant associations emerged, even after considering severe food insecurity, or when using continuous distances or a shorter cut-off (1 km or less). Other null results included perceived adequacy/number of local food stores in three studies( Reference Garasky, Morton and Greder 37 , Reference Kirkpatrick and Tarasuk 48 , Reference Sharkey, Dean and Johnson 49 ) and perceptions/estimations of high food prices in one study( Reference Sharkey, Dean and Johnson 49 ).
Socio-economic environment
Three out of three studies that investigated rent prices at the area level estimated a positive association with food insecurity( Reference Bernell, Weber and Edwards 38 , Reference Bartfeld, Ryu and Wang 47 , Reference Bartfeld and Ahn 51 ); although two of these were conducted by the same lead author. One study uncovered a positive association between an index of area deprivation and food insecurity( Reference Pilgrim, Barker and Jackson 42 ), while one study that used a similar index( Reference Foley, Ward and Carter 54 ) and another that focused on poverty level( Reference Bartfeld, Ryu and Wang 47 ) did not find any significant associations. Bernell et al. (2006) also examined percentage of the county unemployed and average wage, with null results( Reference Bernell, Weber and Edwards 38 ).
Discussion
Among the studies included in the present review, a range of place factors were examined. Summarizing the results by type of place factor revealed a potential protective effect of rural living on food insecurity that may or may not be applicable among older adults. Studies on the quality of the social environment, namely social capital, also suggested a possible protective role; however, half found no significant associations, while those with positive findings tended to focus only on low-income populations. Among studies investigating the food and socio-economic environments, relationships were less clear.
Potential limitations of included studies
As a result of conducting this critical review and from the perspective of furthering knowledge about place and food insecurity, consideration of some of the limitations of the existing research is warranted. The most common limitations are summarized below, but see also the last column in Table 1 for unique study concerns.
The most obvious limitation of this body of research is the exclusive reliance on the cross-sectional study design. Although some physical environmental characteristics do take some time to change, social environmental characteristics may not take as long; longitudinal studies can account for these changes. In addition to taking into account changes in place factors, longitudinal studies can also account for and provide insight into how food insecurity changes over time. This increases our understanding of the problem and increases power to detect significant differences between variables. In longitudinal studies, subjects act as their own controls; thus, confounders that are unobserved and do not change over time can be controlled for, which increases the study's robustness.
Almost half of included studies used the eighteen-item or six-item USDA Food Security Scale and recommended cut-offs to define food insecurity. Benefits of using this measure include rigorous development and validity/reliability testing that has spanned a number of years( Reference Radimer 8 ), as well as increased comparability across studies. Many studies, on the other hand, relied on single question measures, often for secondary data analyses of large population-based surveys. While these questions were largely based on previously validated work and/or derived from items in the USDA Food Security Scale or the Radimer/Cornell measure of food insecurity, they are not likely to measure food insecurity comprehensively and make it difficult or impossible to compare across studies( Reference Radimer 8 ). Given that measures with many items are burdensome to respondents, especially so when administered as part of a large-scale population-based survey, the USDA Food Security Scale six-item short form may be an ideal candidate for more widespread use in future intervention and observational studies, instead of simply relying on single item measures. The shorter length of the questionnaire does not appear to affect its discriminatory power( Reference Blumberg, Bialostosky and Hamilton 55 ). However, this measure does not directly ask about child food insecurity and cannot measure the most severe form of adult food insecurity, where children's intake is likely to be reduced( 56 ). In addition to single item measures, some studies used multiple questions as more than one outcome. This can increase the likelihood of estimating a significant association by chance and makes overall interpretation of results more difficult. Analysing types or severity of food insecurity (e.g. household, adult and child) as individual outcomes or in multinomial logistic regression analysis, without a priori hypotheses and context, also makes results difficult to interpret.
Many studies did not control for sex of the respondent. Given societal gender roles, men and women often perceive situations differently. For example, females are generally responsible for food management in the household and therefore would likely be more attuned to problems with food security. Gender differences may also affect how the surrounding environment is perceived. Some studies that did not adjust for sex selected respondents who were primarily responsible for buying and cooking food in the household, which could partially adjust for this difference (e.g. Kirkpatrick and Tarasuk (2010)( Reference Kirkpatrick and Tarasuk 48 )). Additionally, it would be less concerning if there was very little variability in the sex of the respondent; however, no studies appeared to discuss this. In addition to not adjusting for sex, several studies did not adjust for area-level income, SES or some other measure incorporating relative area disadvantage. Thus, these studies cannot provide information on the potential for area disadvantage to confound associations between the place factors and food insecurity, or the possibility that the place factors mediate the effects of area disadvantage.
There was much heterogeneity among measures of the food environment and clearly there is a lack of a critical mass of studies examining the same or similar features in order to make any type of conclusions about the potential effect of the food environment on food insecurity. Additionally, there needs to be more discussion about how some of these measures were derived, particularly with respect to respondent perceptions; questions seemed vague, not always tested, and thus open to bias as an explanation for findings.
Interestingly, area SES was underexplored, perhaps because a large proportion of included studies (39 %) limited populations to low-income or certain ethnic groups, or because of the well-known link between individual income/SES level and food insecurity. Nevertheless, there still may be some variability in area SES to explain differences in individual food security status, even among low-income populations. Certainly, among population-based samples, area SES should continue to be explored. And as discussed previously, in any study on place and food insecurity, area SES should be considered as a potential confounder, or even an effect modifier, when analysing other features of place.
Sample selection, along with generalizability, should be kept in mind when interpreting the results of the present review. Limiting sample populations to a particular demographic subgroup reduces generalizability of the results. Generalizability is especially problematic when attempting to synthesize information from a relatively new area of research, with few studies, and even more so when most of those studies are conducted in one country; here the reader is cautioned that most studies were conducted in the USA. Once a larger evidence base is established, future reviews should conduct sensitivity analyses to determine potential differences among population subgroups. Limiting to particular subgroups also makes it difficult to detect important significant differences, due to less variability within the sub-population sample than in a population-based sample.
Of the individual-level studies that relied on some form of cluster sampling method, not all adjusted for potential correlations between individuals within a cluster, which could bias the results from statistical tests of association. Additionally, sampling frames were sometimes based on telephone lists. This likely resulted in an underestimation of food insecurity, as having a telephone land-line is related to income, which is related to being food insecure. Lower study power and decreased ability to generalize are likely outcomes of this sampling method. Selection bias may also result if under-coverage is related to the place characteristic(s) under study. Selection bias is also a concern when certain participants are excluded because of missing data. A discussion of the impact of missingness was lacking overall in this body of research.
While the results of the review suggest a potential protective effect of rural living, the measures of living location were generally crude, encompassed large areas (counties), were heterogeneous and often were not the main interest of the study. Given this, one can only speculate as to what it is about rural environments that may protect against food insecurity. More precise definitions and comparisons may yield different results.
Finally, exposures may not be the same for individuals living in the same area. For example, living in a disadvantaged area may not actually be an important exposure for a particular individual, depending on his/her own compositional factors such as income and car ownership, as well as interacting contextual factors such as social capital, high-income neighbourhoods located close by or availability of subsidized school meals, to name a few. Given the complexities of exploring these types of interactions, it is not surprising that none of the included studies conducted this analysis. A handful of studies did discuss car ownership and/or other transportation methods( Reference Garasky, Morton and Greder 37 , Reference Kirkpatrick and Tarasuk 48 , Reference Sharkey, Dean and Johnson 49 ) and some did adjust for other place factors in multivariable models. A relational understanding of place that takes into account spatial and temporal mobility has been recommended in the literature( Reference Cummins, Curtis and Diez-Roux 57 ) and is likely applicable in this area of research, at least with respect to the development of theory and study design.
Limitations of the present review
In addition to the limitations of included studies, the review itself has a number of limitations that may have affected which studies were and were not included. The grey literature was not searched; another recent review on environmental characteristics and food insecurity uncovered several papers that were not published in scientific journals( Reference Gorton, Bullen and Mhurchu 36 ). These studies, for the most part, did not examine place as defined in the present review and so would not have been eligible for inclusion. Conference proceedings and abstracts as well as dissertations were not included; it is therefore possible that studies with null or non-intuitive findings were not included. Hand-searching the reference lists of included articles was conducted, but not for entire journals in the field, and only one person (M.A.C.) selected articles based on a priori eligibility criteria and abstracted the data. Thus, pertinent studies may have been missed. Studies may have also been missed if they analysed a place factor which was not part of the main objective(s) (e.g. it would not have been evident from the title or abstract and then potentially screened out). The review did not focus on household environments or the broader socio-economic environment, both of which may play a role in food insecurity( Reference Nord and Parker 58 ).
Recommendations and conclusions
This critical appraisal and synthesis of published research allowed for the formulation of recommendations for future research studies, which should help to drive the field forward. These are detailed below.
Sampling methods should avoid using telephone lists to recruit participants whenever possible. Some variant of cluster random sampling may be most appropriate, where surveys are administered in person. Population-based samples that do not focus exclusively on low-income and/or largely rural populations may be most informative, especially from a policy-making perspective. In order to reduce respondent burden, make use of a solid evidence base and avoid basing measurement of food insecurity on one question, authors of future studies may consider using, at the very least, the USDA Food Security Scale six-item short form. This will help to increase comparability across studies. Finer grained definitions and a specific focus on place are also needed; especially with respect to living location. Including housing/residential density, land-use mix, farming, social capital, as well as exploring how car ownership and other transportation methods can influence food insecurity in the context of rural and urban living, is important. Well-conducted longitudinal observational studies are preferred to cross-sectional studies, and testing for interaction with the place variables of interest could help to make results more robust and informative. Furthemore, adjustment for sex as well as other confounders, such as individual and area SES, is necessary to reduce bias in the resulting associations between the place factors and food insecurity.
Finally, community-based initiatives, such as community gardens, were not evaluated in the present review because none included measures of individual/household food insecurity as outcomes. Studies that were screened out consisted mostly of process evaluations of single programmes. This area of research could benefit immensely from applying more rigorous experimental and quasi-experimental methods and evaluating changes in individual/household-level food security status of area residents. Randomized community-wide interventions that are not necessarily programmatic in nature (e.g. changes to the physical environment) are also important areas for further research. Place and food insecurity is a fairly new and evolving area of research. Given that everyone should have access to healthy, acceptable food, regardless of income, and that developing redistributive income and other equitable policies is socio-politically complex, time consuming and contentious in some high-income democratic countries, focusing on how the immediate local environments may improve or inhibit food security could be a potentially fruitful area of research, especially in today's economic climate. The literature synthesized in the present review points to rural living as a potential protective factor, although a number of methodological limitations prevent any decisive conclusions from being made at this time. Recommendations have been formulated and presented in an attempt to improve the quality of research in this field.
Acknowledgements
Sources of funding: M.A.C. received the Banting & Best Canada Graduate Scholarship from the Canadian Institutes of Health Research in support of this work. Ethics: This review did not require ethics approval. Conflicts of interest: The authors declare that they have no conflicts of interest. Author contributions: M.A.C. and L.D. came up with the topic for review; M.A.C. conducted the review and drafted the paper; M.S.T. and L.D. provided direction and advice during the drafting of the paper, as well as feedback and edits on various versions of the paper. All authors read and approved the final submission of the paper.
