Neighbourhood-Socioeconomic Variation in Women's Diet: the Role Of

ORIGINAL ARTICLE Neighbourhood-socioeconomic variation in women’s diet: the role of nutrition environments

LE Thornton, DA Crawford and K Ball

Centre for Physical Activity and Nutrition Research, School of Exercise and Nutrition Sciences, Deakin University, Burwood, Victoria, Australia

Background/Objectives: Living in socioeconomically disadvantaged neighbourhoods is associated with increased risk of a poor diet; however, the mechanisms underlying associations are not well understood. This study investigated whether selected healthy and unhealthy dietary behaviours are patterned by neighbourhood-socioeconomic disadvantage, and if so, whether features of the neighbourhood–nutrition environment explain these associations. Subjects/Methods: A survey was completed by 1399 women from 45 neighbourhoods of varying levels of socioeconomic disadvantage in Melbourne, Australia. Survey data on fruit, vegetable and fast-food consumption were linked with data on food store locations (supermarket, greengrocer and fast-food store density and proximity) and within-store factors (in-store data on price and availability for supermarkets and greengrocers) obtained through objective audits. Multilevel regression analyses were used to examine associations of neighbourhood disadvantage with fruit, vegetable and fast-food consumption, and to test whether nutrition environment factors mediated these associations. Results: After controlling for individual-level demographic and socioeconomic factors, neighbourhood disadvantage was associated with less vegetable consumption and more fast-food consumption, but not with fruit consumption. Some nutrition environmental factors were associated with both neighbourhood disadvantage and with diet. Nutrition environmental features did not mediate neighbourhood-disadvantage variations in vegetable or fast-food consumption. Conclusions: Although we found poorer diets among women living in disadvantaged neighbourhoods in Melbourne, the differences were not attributable to less supportive nutrition environments in these neighbourhoods. European Journal of Clinical Nutrition (2010) 64, 1423–1432; doi:10.1038/ejcn.2010.174; published online 1 September 2010

Keywords: socioeconomic status; neighbourhood; food intake; fast foods; fruits; vegetables

Introduction attention has shifted to socioeconomic differences at the neighbourhood level. Evidence of associations with neigh- Within developed countries, few individuals meet dietary bourhood disadvantage are mixed, with some studies recommendations for fruits and vegetables (Magarey et al., showing inverse associations between neighbourhood dis- 2006; Tamers et al., 2009), whereas consumption of energy- advantage and dietary quality (Turrell et al., 2009), whereas dense foods is increasing (Guthrie et al., 2002). These dietary others report no difference after controlling for individual behaviours increase the risk of obesity (Pereira et al., 2005; characteristics (Giskes et al., 2006). Buijsse et al., 2009), coronary heart disease (Ness and Powles, Where neighbourhood-level differences exist, it has been 1997) and diabetes (Pereira et al., 2005). suggested that these may operate through variations in Determinants of dietary behaviours have traditionally community and/or consumer nutrition environments been linked to individual characteristics such as socio- (Glanz et al., 2005). The ‘community nutrition environment’ economic position (SEP), with lower SEP associated with less relates to the type and location of food stores in an area, healthy diets (Ball et al., 2006; Roos et al., 2008). Recently, whereas ‘consumer nutrition environment’ relates to within- store factors such as product availability, quality, price and Correspondence: Dr LE Thornton, Centre for Physical Activity and Nutrition opening hours. For community nutrition environments, Research, School of Exercise and Nutrition Sciences, Deakin University, studies have shown patterning between neighbourhood 221 Burwood Highway, Burwood, Victoria, 3125, Australia. disadvantage and access to fast-food restaurants (Cummins E-mail: [email protected] Received 8 April 2010; revised 5 July 2010; accepted 14 July 2010; published et al., 2005; Pearce et al., 2007); however, no trend is online 1 September 2010 decipherable for supermarkets (Pearce et al., 2008; Neighbourhood-socioeconomic variations in diet LE Thornton et al 1424 Ball et al., 2009). Although some studies report that these In 2004, 2400 women aged 18–65 years from these differences are unrelated to diet (Ball et al., 2006; Turrell and 45 suburbs were posted a survey assessing dietary behaviours Giskes, 2008), others have shown greater access to healthier and their determinants. In total, 1136 women responded food stores (for example, supermarkets) is positively corre- (50% overall, excluding from the denominator 127 women lated with dietary quality (Moore et al., 2008; Zenk et al., who had moved/were ineligible): 354 from high-, 407 from 2009), whereas greater fast-food restaurant access is linked to mid- and 375 from low-socioeconomic status neighbour- more frequent fast-food purchasing (Thornton et al., 2009). hoods. A second independent sample was drawn in the same Evidence from the US and UK on consumer nutrition manner for a separate physical activity survey. Women who environments suggests that in more deprived neighbour- responded to this were asked if they were also willing to hoods, there are fewer healthy choices available within complete the dietary behaviours survey. This second phase of stores (Andreyeva et al., 2008; Franco et al., 2008), and that the study resulted in an additional 444 diet surveys (42% of prices for the same foods are higher (Crockett et al., 1992; those completing the original physical activity survey). Sooman et al., 1993). However, other studies, primarily from Excluding data from 13 women who had recently moved outside of the US, show the reverse (Cummins and out of the study neighbourhoods and 168 women who had Macintyre, 2002; Ball et al., 2009) or otherwise indicate few missing data on one or more of the individual-level study or no differences (Winkler et al., 2006; Latham and Moffat, variables, the final sample size was 1399 (see Table 1). For the 2007). Evidence of associations between within-store factors analysis of consumer nutrition environments, only a subset and diets is also mixed (Giskes et al., 2007; Powell et al., 2009; of data were used (n ¼ 928 women from 35 neighbourhoods Zenk et al., 2009). for greengrocers; n ¼ 1082 women from 37 neighbourhoods Further evidence is required to more clearly establish links for supermarkets), as not every participant had a store within between dietary behaviours, neighbourhood disadvantage 3 km of their home, or the stores within 3 km fell outside of and both community and consumer nutrition environ- the 45 suburb boundaries from within which the consumer ments. This study investigated whether dietary beha- nutrition environment information was collected. viours are patterned by neighbourhood disadvantage, and if so, whether features within local community and consumer nutrition environments explain these associations. Measures This is the first study to consider both community and Outcomes. Fruit and vegetable intakes were assessed sepa- consumer nutrition environments as mediators of area-level rately by asking ‘How many servings of (fruit/vegetables) associations. do you usually eat each day?’ (with examples of servings provided). Response options were ‘none’, ‘1 serving’, ‘2 servings’, ‘3–4 servings’ (coded 3.5 for analyses) or ‘5 Subjects and methods servings or more’ (coded 5). These questions were adapted from the Australian National Nutrition Survey, in which they Participants were shown to discriminate between groups with different These analyses are based on 1399 women participating in the fruit and vegetable intakes assessed by 24-h recall. They also Socioeconomic Status and Activity in Women (SESAW) show good test–retest reliability of 0.85 each (Ball et al., study. Study methods have been described previously (Ball 2006). Acceptable (termed ‘high’) fruit and vegetable con- et al., 2008, 2009) and approval for Socioeconomic Status sumption was defined as consuming two or more serves and Activity in Women was obtained from the Deakin daily. Although Australian guidelines recommend five ser- University ethics committee. Women were recruited from vings of vegetables per day, only 5% of our sample consumed the Australian electoral roll (voting is compulsory for all this amount. Respondents were also asked how many meals Australian adults) using a stratified random sampling per week they ate from fast-food restaurants (for example, procedure from 45 neighbourhoods (suburbs) of different pizza, McDonalds), with separate questions for within- levels of disadvantage in Melbourne, Australia. On the basis restaurant consumption and takeaway (at home, work or of the 2001 Census data, the Australian Bureau of Statistics study) consumption. Six response categories were listed assigned suburbs a Socioeconomic Index for Areas (SEIFA) ranging from ‘never’ to ‘6–7 meals per week or more’. score based on the index of relative socioeconomic dis- Responses to the within-store and takeaway were combined. advantage (Australian Bureau of Statistics 2003). The SEIFA The outcome variable was recoded to ‘never’, ‘infrequent’ index of relative socioeconomic disadvantage is an area- (less than one meal per week) and ‘frequent’ consumption based measure that takes into account factors from a range of (one or more per week). These categories were guided by socioeconomic measures, including income and education. previous evidence that shows a higher risk of adverse health All suburbs within 30 km of the central business district were outcomes among frequent fast-food consumers compared ranked according to SEIFA score, and 15 suburbs were drawn with infrequent consumers (Pereira et al., 2005). randomly from each of the lowest, middle and highest SEIFA septiles. This ensured that women from a range of socio- Covariates. Women were asked to report on a range of economic status backgrounds were represented. sociodemographic and socioeconomic characteristics. Covariates

European Journal of Clinical Nutrition Neighbourhood-socioeconomic variations in diet LE Thornton et al 1425 Table 1 Demographic characteristics and fruit, vegetable and fast-food consumption by neighbourhood-level disadvantage

Neighbourhood disadvantage P-value

Total, n (%) Low, n (%) Mid, n (%) High, n (%)

Total 1399 464 (33.2) 539 (38.5) 396 (28.3)

Mean (s.d.) Mean (s.d.) Mean (s.d.) Mean (s.d.) Age 40.9 (12.5) 42.2 (12.4) 40.6 (12.0) 39.8 (13.2) 0.014*

Number in household dependent on income 2.8 (1.3) 2.7 (1.3) 2.9 (1.3) 2.8 (1.3) 0.343*

n (%) % % % Country of birth Australia 1074 (76.8) 82.4 82.6 61.9 w Overseas 325 (23.2) 17.2 17.4 38.1 o0.001

Marital status Married/de facto 911 (65.1) 65.3 68.8 59.9 w Separated/divorced/never married/widow 488 (34.9) 34.7 31.2 40.1 0.017

Education Degree or higher degree 526 (37.6) 55.6 34.1 21.2 Year 12, trade or certificate 556 (39.7) 34.3 44.2 40.2 w Less than year 12 317 (22.7) 10.1 21.7 38.6 o0.001

Occupation Professional 599 (42.8) 55.2 45.8 24.2 White collar 335 (23.9) 19.0 26.7 26.0 Blue collar 128 (9.2) 3.7 7.1 18.4 w Not in workforce 337 (24.1) 22.2 20.4 31.3 o0.001

Income Aus $78 000 or more 324 (23.2) 37.7 22.5 7.1 Aus $52 000–77 999 199 (14.2) 10.8 19.3 11.4 Aus $37 000–51 999 164 (11.7) 6.9 12.2 16.7 Aus $36 999 or less 187 (13.4) 5.6 11.1 25.5 w Missing 525 (37.5) 39.0 34.9 39.4 o0.001

Fruit consumption Low servings (less than 2) 560 (40.1) 36.6 38.2 46.7 w High servings (2 or more) 837 (59.9) 63.4 61.8 53.3 0.006

Vegetable consumption Low servings (less than 2) 430 (30.7) 21.3 27.8 45.7 w High servings (2 or more) 969 (69.3) 78.7 72.2 54.3 o0.001

Fast-food consumption Never 283 (20.9) 29.2 16.8 16.6 Infrequent (1–3 times per month) 713 (52.7) 52.1 56.1 48.8 w Frequent (4 or more times per month) 356 (26.3) 18.7 27.1 34.6 o0.001

Abbreviation: ANOVA, analysis of variance. *P-values determined through ANOVA. w P-values determined through w2. adjusted for analyses included: age, country of birth, marital websites in 2004. Greengrocers were defined as stores that status, highest education level attained, occupation, household primarily sold fresh fruit and vegetables. Supermarkets were income and number of people dependent on that income. identified as belonging to one of the five large supermarket chains within Australia. Nine major fast-food restaurant Exposure chains were included: Dominos, Kentucky Fried Chicken Community nutrition environment. Data on locations of (KFC), Hungry Jacks, McDonalds, Nandos, Pizza Haven, Pizza greengrocers, major supermarkets and fast-food restaurants Hut, Red Rooster and Subway. in and immediately surrounding the 45 neighbourhoods Geocoding of participants and food stores were under- were sourced through extensive searches of online telephone taken using a Geographic Information System (ArcView 3.3, directories, local council/government websites and company ESRI, Redlands, 2002) and overlaid with the road network

European Journal of Clinical Nutrition Neighbourhood-socioeconomic variations in diet LE Thornton et al 1426 (VicMap Transport v2004, owned and supplied by the State created to indicate the overall availability and price of fresh of Victoria). Proximity measures were calculated as road fruit and vegetables, independent of store type. network distance between each participant’s household location and the nearest store of each type (greengrocer, supermarket, fast-food restaurant). Density was measured as Statistical analysis a count of each store type within 3 km of road network Descriptive and multilevel analyses were undertaken in 2009 distance from each participant’s household. A variety of fast- using Stata 10.1 (StataCorp, College Station, TX, USA, 2008). food restaurants (number of different chains within 3 km In the multilevel analytical models, we used logistic regres- from households) was created to represent the choice of sion for the fruit and vegetable outcomes and multinomial options available. The use of 3 km was based on previous regression for fast-food consumption. To test for mediation, findings, which suggested that the majority of people do nutrition environment variables were added separately and their food shopping within this distance (Cairns, 1995). changes in the odds ratios for associations between neighbourhood disadvantage and the dietary outcome were Consumer nutrition environment. In 2006, data were gathered observed to assess the impact of adding each mediator to during store audits on the availability and price of the model. In line with previously described approaches to 15 commonly consumed fruits and 23 vegetables in 134 mediation (Baron and Kenny, 1986; Cerin and Mackinnon, stores, identified as being within the boundaries of the 2009), mediating analysis was only conducted for environ- 45 neighbourhood. These stores were assessed in analyses mental features that: (1) varied significantly by neighbour- only if they also fell within the 3-km boundary from the hood disadvantage; (2) varied in the direction such that in participant’s homes. disadvantage neighbourhoods it would make purchasing of Availability of each item was coded as a binary measure fruit and vegetable more difficult (for example, nearest store (‘available’ and ‘not available’), and the final availability further away, lower fruit or vegetable availability, higher measure was calculated by tallying the total number of fruit prices or reduced opening hours), or the purchasing of fast and vegetable items available within each store. food easier; and (3) were significantly associated with the Data were collected based on the price per kilogram, or the specific dietary outcome, which they were hypothesized to price per item for individually priced items (for example, influence (for example, vegetable and fast-food-related mangoes), of the cheapest item available. Prices per item mediators were not tested for fruit consumption). were converted to price per kilogram by dividing an item’s price by its typical weight, obtained from a comprehensive food list software package (FoodWorks Professional, Xyris Results Software, Brisbane, 2007). The mean price for each fruit and vegetable item across all stores was calculated. For each item Descriptive available within a store, the cost difference between that Sample characteristics by neighbourhood disadvantage. Charac- item and the overall mean price for that item across all stores teristics of the sample are presented in Table 1. Women in (termed the ‘mean difference’) was calculated. If a positive highly disadvantaged neighbourhoods had lower individual value was returned, this item was more expensive than the SEP, and reported less frequent fruit and vegetable consump- overall mean price for that item. For example, if the price of tion and more frequent fast-food consumption. an apple within a store was Aus $1.20 and the overall mean price of apples across all stores was Aus $1.00, then this item Nutrition environment by neighbourhood disadvantage. Table 2 was Aus $0.20 more expensive. A negative value indicated shows the differences in nutrition environment measures by this item was cheaper than the mean. The ‘mean difference’ neighbourhood disadvantage. Greengrocers were more for all items in a store was summed and divided by the accessible to those living in neighbourhoods with low number of items available, resulting in a single price figure disadvantage. Within greengrocers in highly disadvantaged for each store. Essentially this measure reflected whether the neighbourhoods fruit and vegetables were cheaper, but items available were generally cheaper or more expensive availability and opening hours more restricted. For super- than in other stores. markets in low disadvantaged, proximity and density Opening hour information was collected by the auditors indicated greater access, but item availability and opening and variables were created to indicate the total opening hours were slightly lower. Prices of fruits and vegetables were hours on weekends, after 1730 hours (and before midnight) cheapest in supermarkets in high-disadvantaged neighbour- on weekdays, and total throughout the week. hoods. The combined measures for greengrocers and super- The final availability, price and opening hour measures markets showed lower availability and price of fruits and were calculated by averaging the consumer nutrition vegetables in high-disadvantaged neighbourhoods. For fast- environment information across all stores within 3 km of food restaurants, those in high-disadvantaged neighbour- each respondent’s home. Proximity, density and opening hoods were less likely to live close to a fast-food restaurant hour measures were examined specific to either greengrocers and were exposed to a lower density and variety of chains or supermarkets; however, combined measures were also than those in low-disadvantaged neighbourhoods.

European Journal of Clinical Nutrition Neighbourhood-socioeconomic variations in diet LE Thornton et al 1427 Table 2 Community and consumer nutrition environment by neighbourhood-level disadvantage

Neighbourhood disadvantage P-value

Total, n(%) Low (%) Mid (%) High (%)

Greengrocers Proximity Less than 1 km 540 (34.5) 57.7 22.6 25.4 1–2 km 658 (42.0) 40.7 45.1 40.7 2–3 km 208 (13.3) 1.4 18.0 20.3 w a,b 3 km or greater 161 (10.3) 0.2 14.3 13.6 o0.001*, ,

Mean (s.d.) Mean (s.d.) Mean (s.d.) Mean (s.d.) y a,b,c Store density 5.4 (4.8) 10.0 (5.0) 3.9 (3.2) 2.6 (1.8) o0.001 ,z, y a,b,c Fruit availability (max. 15) 11.8 (1.5) 12.1 (12.6) 12.6 (1.0) 10.9 (2.1) o0.001 ,z, y a,b,c Vegetable availability (max. 23) 21.9 (1.6) 22.7 (0.3) 22.3 (1.7) 20.6 (1.9) o0.001 ,z, y a,b,c Fruit price difference (Aus $) — 1.02 (0.87) À0.02 (0.65) À1.32 (0.32) o0.001 ,z, y a,b,c Vegetable price difference (Aus $) — 1.03 (0.98) 0.20 (0.55) À0.99 (0.23) o0.001 ,z, y a,b Total opening hours 75.3 (26.2) 82.2 (35.6) 66.9 (3.8) 70.6 (9.4) o0.001 ,z, y a,c Weekend opening hours 16.9 (10.4) 18.3 (13.1) 13.8 (2.9) 16.9 (8.4) o0.001 ,z, y a,b Hours open after 1730 h 5.8 (7.4) 7.4 (10.1) 4.7 (3.3) 4.1 (1.6) o0.001 ,z,

Supermarkets n (%) % % % Proximity Less than 1 km 500 (31.9) 46.7 24.7 26.0 1–2 km 795 (50.7) 47.9 55.4 49.3 2–3 km 180 (11.5) 5.2 13.0 16.6 w a,b 3 km or greater 92 (5.9) 0.2 6.9 8.2 o0.001*, ,

Mean (s.d.) Mean (s.d.) Mean (s.d.) Mean (s.d.) y a,b Store density 4.2 (2.5) 5.8 (2.2) 3.5 (2.4) 3.4 (2.1) o0.001 ,z, y a,c Fruit availability (max. 15) 10.6 (2.2) 10.2 (2.1) 11.1 (2.5) 10.4 (1.7) o0.001 ,z, y a,b,c Vegetable availability (max. 23) 20.4 (3.9) 19.6 (2.9) 21.3 (5.5) 20.5 (2.9) o0.001 ,z, y a,b,c Fruit price difference (Aus $) — 0.58 (0.49) 0.48 (0.45) À0.21 (0.58) o0.001 ,z, y a,b,c Vegetable price difference (Aus $) — 0.46 (0.49) 0.01 (0.33) À0.22 (0.47) o0.001 ,z, y a,b,c Total opening hours 108.0 (17.7) 106.4 (17.4) 113.2 (19.9) 104.8 (14.4) o0.001 ,z, y a,b,c Weekend opening hours 30.4 (5.1) 29.4 (4.9) 32.3 (5.7) 29.6 (4.2) o0.001 ,z, y a,b,c Hours open after 1730 h 23.5 (7.1) 21.0 (5.6) 26.0 (7.8) 24.0 (7.2) o0.001 ,z,

Both greengrocers and supermarkets y a,b,c Fruit availability (max. 15) 11.3 (1.3) 11.4 (0.9) 11.8 (1.3) 10.6 (1.4) o0.001 ,z, y a,b,c Vegetable availability (max. 23) 21.3 (2.1) 21.4 (0.9) 22.1 (2.5) 20.4 (2.3) o0.001 ,z, y a,b,c Fruit price difference (Aus $) — 0.84 (0.59) 0.35 (0.53) À0.64 (0.53) o0.001 ,z, y a,b,c Vegetable price difference (Aus $) — 0.76 (0.60) 0.05 (0.35) À0.49 (0.46) o0.001 ,z,

Fast food n (%) % % % Proximity Less than 1 km 493 (31.5) 32.9 35.9 25.4 1–2 km 783 (50.0) 54.5 43.7 54.1 2–3 km 198 (12.6) 9.4 11.7 17.4 w a,b,c 3 km or greater 93 (5.9) 3.2 8.6 3.1 o0.001*, ,

Mean (s.d.) Mean (s.d.) Mean (s.d.) Mean (s.d.) y a,b Density 7.7 (6.9) 10.9 (10.1) 6.4 (4.3) 6.2 (3.1) o0.001 ,z, y a,b,c Variety (max. 9) 4.9 (2.2) 5.6 (1.8) 4.3 (2.3) 5.0 (2.1) o0.001 ,z,

Abbreviation: ANOVA, analysis of variance. *P-values determined through w2. w w2 test used to determine significance for pairwise comparisons of proportions. zBonferroni multiple-comparison test used for differences between individual groups. y P-values determined through ANOVA. aLow disadvantage significantly different (Po0.05) from mid disadvantage. bLow disadvantage significantly different (Po0.05) from high disadvantage. cMid disadvantage significantly different (Po0.05) from high disadvantage.

European Journal of Clinical Nutrition Neighbourhood-socioeconomic variations in diet LE Thornton et al 1428 Nutrition environment by dietary behaviours. More frequent Europe, suggesting that certain aspects of diet are patterned fruit consumers had access to a greater number of fruit by area-level disadvantage (Subar et al., 1995; Ecob and products within greengrocers, as well as longer total and Macintyre, 2000; Shohaimi et al., 2004; Ball et al., 2006). weekend opening hours (Table 3). Frequent vegetable We hypothesized that any associations between neigh- consumers lived closer to and had a higher density of bourhood disadvantage and diet may be explained by greengrocers and supermarkets in their neighbourhood, and variations in nutrition environments. However, not all more vegetable items available within greengrocers. Fruit environmental features indicated unhealthy diets would be and vegetable price was positively associated with intake, more likely in high-disadvantaged neighbourhoods, particu- with more frequent consumers exposed to higher prices in larly in relation to the price of fruits and vegetables and fast- both greengrocers and supermarkets. Those who reported food restaurant access. Similarly, while bivariate associations never consuming fast food were exposed to a higher density revealed aspects of the nutrition environments that were and variety of fast-food restaurants. associated with diet, others were not, and not all were in the expected direction. Most notably, women exposed to higher Multilevel mediation analyses. In models adjusted for indivi- prices ate more fruits and vegetables, and women with dual SEP, fruit consumption was not associated with reduced neighbourhood exposure to fast-food stores ate fast neighbourhood-level disadvantage. Respondents in high- food more often. These findings could reflect the higher disadvantaged neighbourhoods remained significantly less individual SEP of women living in neighbourhoods where likely to consume two or more servings of vegetables per day prices were higher and fast-food store access greater, as high after adjustment for individual SEP (Table 4). individual SEP is associated with a range of factors that Environmental factors that showed significant bivariate predict healthier diets such as greater nutrition knowledge associations in the expected direction with both neighbour- and health considerations applied to purchasing (Ball et al., hood disadvantage and vegetable consumption were added 2006). Further, higher fruit and vegetable price may reflect separately as mediators in multilevel regression models. better quality, which may relate to consumption. The inclusion of these variables had little impact on the Nutrition environment variables considered appropriate to magnitude or significance of the association between test for mediation were those that differed significantly by neighbourhood disadvantage and vegetable consumption. neighbourhood disadvantage and would have made heal- On comparing with those who never consumed fast food, thier dietary decisions more difficult in high-disadvantaged the odds of consuming fast food infrequently (less than neighbourhoods. Of the environmental variables tested, once per week) were higher in both the mid- and high- none mediated associations between neighbourhood disad- disadvantaged neighbourhoods relative to low-disadvan- vantage and vegetable consumption. The lack of significant taged neighbourhoods after adjustment for individual SEP. mediation may be owing to factors not examined. For Frequent fast-food consumption was more than twice example, taste preferences, family preferences, social norms, as likely in mid-disadvantage neighbourhoods and three cultural factors or support from family or friends all impact times more likely in high-disadvantaged neighbourhoods food purchasing choices (Glanz et al., 1998; Ball et al., 2006), compared with those in low-disadvantaged neighbourhoods. and may be more important mediators than those assessed The three measures of fast-food restaurant access did not here. Alternatively, it may be that the nutrition environment indicate that exposure to fast-food restaurants was greater in was not examined with appropriate precision to detect high-disadvantaged neighbourhoods, thus these were not mediation effects in this sample. For example, inclusion of tested as mediators. other environmental characteristics (for example, presence of farmers’ markets, metres square and quality of fresh produce within supermarkets and greengrocers) may have Discussion provided a more comprehensive picture of the actual environmental influences on dietary behaviours. Further, We observed less frequent fruit and vegetable consumption there may be problems associated with classifying super- and more frequent fast-food consumption among residents markets as healthy food stores owing to the large provision of neighbourhoods with greater disadvantage. However, of unhealthy food within these (for example, energy-dense only associations for vegetable and fast-food consumption snack food and soft drinks); however, supermarkets do stock remained significant after adjustment for individual SEP. a large range of fresh produce at cheap prices. Therefore, we Acknowledging the cross-sectional design does not permit have considered them important in potentially encouraging causal inference, this is suggestive that fruit intake may be fruit and vegetable consumption. In many cases, we did not more influenced by individual- than by neighbourhood-level observe a large exposure gradient among the nutrition factors, which is inconsistent with previous Australian environment variables, thus reducing our ability to detect evidence (Turrell et al., 2009). The significant associations mediating effects. Finally, the objective environment may of neighbourhood disadvantage with vegetable and fast-food not be as predictive of dietary behaviours as environmental consumption, on the other hand, are similar to those perceptions (Giskes et al., 2007), which could also explain previously reported from Australia, the UK, the US and the failure to identify mediation by objectively assessed

European Journal of Clinical Nutrition Neighbourhood-socioeconomic variations in diet LE Thornton et al 1429 Table 3 Community and consumer (within-store) nutrition environment by fruit, vegetable and fast-food consumption

Fruit consumption P-value Vegetable consumption P-value

Low (%) High (%) Low (%) High (%)

Greengrocers Proximity Less than 1 km 36.1 33.5 30.6 36.1 1–2 km 38.3 44.5 37.8 44.0 2–3 km 13.9 12.6 19.6 10.3 w w 3 km or greater 11.7 9.4 0.088 12.0 9.6 o0.001

Mean (s.d.) Mean (s.d.) Mean (s.d.) Mean (s.d.) Store density 5.2 (4.7) 5.5 (4.9) 0.199* 4.5 (4.3) 5.8 (5.0) o0.001* Fruit availability (max. 15) 11.7 (1.7) 11.9 (1.4) 0.038* 11.6 (1.8) 12.0 (1.4) o0.001* Fruit price difference (Aus$) À0.06 (1.21) 0.11 (1.21) 0.026* À0.32 (1.17) 0.20 (1.20) o0.001* Vegetable availability (max. 23) 21.8 (1.8) 22.1 (1.5) 0.002* 21.5 (1.8) 22.1 (1.5) o0.001* Vegetable price difference (Aus$) 0.06 (1.07) 0.28 (1.15) 0.002* À0.12 (1.07) 0.33 (1.13) o0.001* Total opening hours 73.1 (22.3) 76.9 (28.4) 0.034* 73.6 (22.8) 76.2 (27.6) 0.166* Weekend opening hours 16.1 (9.4) 17.4 (11.0) 0.058* 16.4 (9.5) 17.1 (10.8) 0.379* Hours open after 1730 h 5.4 (6.3) 6.1 (8.1) 0.185* 5.4 (6.4) 6.0 (7.8) 0.221*

Supermarkets % % % % Proximity Less than 1 km 32.3 31.4 24.5 34.9 1–2 km 49.4 52.0 52.7 50.2 2–3 km 12.2 10.8 15.0 9.8 w w 3 km or greater 6.0 5.9 0.730 7.8 5.1 o0.001

Mean (s.d.) Mean (s.d.) Mean (s.d.) Mean (s.d.) Store density 4.1 (2.5) 4.2 (2.5) 0.183* 3.7 (2.4) 4.3 (2.5) o0.001* Fruit availability (max. 15) 10.6 (2.1) 10.5 (2.2) 0.349* 10.7 (2.1) 10.5 (2.2) 0.190* Fruit price difference (Aus$) 0.24 (0.65) 0.34 (0.59) 0.009* 0.17 (0.64) 0.36 (0.60) o0.001* Vegetable availability (max. 23) 20.5 (3.8) 20.3 (4.0) 0.521* 20.8 (3.6) 20.2 (4.1) 0.032* Vegetable price difference (Aus$) 0.05 (0.53) 0.14 (0.52) 0.004* 0.03 (0.55) 0.14 (0.52) 0.002* Total opening hours 108.0 (17.0) 107.8 (18.2) 0.858* 108.8 (17.5) 107.6 (17.9) 0.280* Weekend opening hours 30.5 (4.9) 30.3 (5.3) 0.529* 30.6 (5.1) 30.3 (5.2) 0.271* Hours open after 1730 h 23.8 (6.9) 23.2 (7.3) 0.207* 24.2 (7.1) 23.1 (7.1) 0.015*

Both greengrocers and supermarkets Fruit availability (max. 15) 11.3 (1.3) 11.3 (1.4) 0.980* 11.2 (1.4) 11.3 (1.3) 0.208* Fruit price difference (Aus$) 0.15 (0.84) 0.25 (0.81) 0.022* À0.01 (0.82) 0.31 (0.80) o0.001* Vegetable availability (max. 23) 21.3 (2.1) 21.3 (2.2) 0.844* 21.3 (2.1) 21.3 (2.2) 0.990* Vegetable price difference (Aus$) 0.05 (0.68) 0.17 (0.71) 0.002* À0.04 (0.68) 0.20 (0.70) o0.001*

Fast food Never (%) Infrequent (%) Frequent (%) P-value Proximity Less than 1 km 35.4 32.0 29.6 1–2 km 50.8 49.1 49.0 2–3 km 9.2 12.3 14.9 w y 3 km or greater 4.6 6.7 6.6 0.212 , ,b

Mean (s.d.) Mean (s.d.) Mean (s.d.) Density 9.1 (8.8) 7.5 (6.6) 7.2 (5.8) o0.001*,z,a,b Variety (max. 9) 5.2 (2.1) 4.8 (2.2) 4.7 (2.2) 0.007*,z,a,b

Abbreviation: ANOVA, analysis of variance. *P-values determined through ANOVA or t-test. w P-values determined through w2. zBonferroni multiple-comparison test used for differences between individual groups. y w2 test used to determine significance for pairwise comparisons of proportions. aNever significantly different (Po0.05) from infrequent. bNever significantly different (Po0.05) from frequent.. environmental characteristics. In relation to past studies, spatial area of aggregation, indicators of disadvantage, similarities and differences are not easily explained, as conceptualization of access, availability and price, sample studies vary considerably in terms of location/context, characteristics and confounders.

European Journal of Clinical Nutrition Neighbourhood-socioeconomic variations in diet LE Thornton et al 1430 Table 4 Multilevel regression model of neighbourhood-level disadvantage and the mediating effects of the nutrition environment

Neighbourhood disadvantage

Mid disadvantage High disadvantage

OR (95% CI) OR (95% CI)

Fruit consumption (two or more servings per day) Model 1 (demographics) 0.94 (0.71–1.23) 0.63 (0.47–0.85)* Model 2 (demographics and SEP) 1.10 (0.82–1.49) 0.87 (0.62–1.22)

Vegetable consumption (two or more servings per day) w Model 1 (demographics) 0.69 (0.50–0.94) 0.33 (0.23–0.45)z Model 2 (demographics and SEP) 0.79 (0.57–1.11) 0.43 (0.30–0.62)z

Model 2 þ mediating variables Greengrocers Proximity 0.90 (0.63–1.28) 0.50 (0.34–0.72)z Store density 0.91 (0.61–1.37) 0.51 (0.32–0.81)*

Model 2 (within-store (greengrocer) Subsample (n ¼ 928)) 1.03 (0.65–1.65) 0.43 (0.27–0.67)z Vegetable availability (max. 23) 1.06 (0.68–1.69) 0.50 (0.30–0.82)* Total opening hours 1.07 (0.63–1.79) 0.42 (0.26–0.70)z Hours open after 1730 h 1.03 (0.62–1.72) 0.42 (0.25–0.68)z

Supermarkets Proximity 0.89 (0.64–1.23) 0.50 (0.35–0.71)z Store density 0.92 (0.65–1.30) 0.51 (0.35–0.74)z

Both greengrocers and supermarkets Model 2 (within-store (both greengrocers and supermarkets) subsample (n ¼ 1199)) 0.88 (0.62–1.26) 0.43 (0.29–0.64)z Vegetable availability (max. 23) 0.91 (0.64–1.29) 0.41 (0.28–0.61)z

Infrequent fast-food consumption Model 1 (demographics) 1.78 (1.30–2.45)z 1.68 (1.18–2.39)* w Model 2 (demographics and SEP) 1.77 (1.26–2.47)z 1.63 (1.08–2.45)

Frequent fast-food consumption Model 1 (demographics) 2.41 (1.64–3.55)z 3.14 (2.07–4.77)z Model 2 (demographics and SEP) 2.42 (1.61–3.63)z 3.04 (1.89–4.89)z

Abbreviations: CI, confidence interval; OR, odds ratio; SEP, socioeconomic position. *P-value o0.01. w P-value p0.05. zP-value o0.001. Ref. group: Low-disadvantage neighbourhoods (OR 1.00). Models for fruit and vegetable outcomes run as multilevel logistic regression (ref. category, low consumption). Models for fast-food outcome run as multilevel multinomial logistic regression (ref. category, never consumed fast food). Model 1: Adjusted for age, country of birth, marital status and number of dependents. Model 2: Model 1 þ further adjustment for education, occupation and income.

Study strengths include the objective and comprehensive over this period. We recognize that the multilevel analysis assessment of both the community and consumer nutrition may have been underpowered at the area level, with a large environments, and the simultaneous examination of both study potentially providing a greater chance of detecting healthy and unhealthy dietary behaviours. In addition to the area-level differences. Finally, we acknowledge the chal- cross-sectional design, this study was limited by the time lenges related to the assessment of dietary intake. Although difference between survey data and in-store data collection alternate approaches exist, such as Food Frequency Ques- (just under 2 years). However, given that we examined tionnaires, these can potentially result in the over-reporting established urban areas, we do not expect major changes in of fruit and vegetable consumption (Cade et al., 2002). the food environment over this period of time and there is In conclusion, this study does not support the hypothesis no reason to expect that the relative differences of in-store that poorer diets among women living in disadvantaged features by neighbourhood disadvantage had varied greatly neighbourhoods in Melbourne are attributable to less

European Journal of Clinical Nutrition Neighbourhood-socioeconomic variations in diet LE Thornton et al 1431 supportive nutrition environments. Future research is neces- Cummins S, Macintyre S (2002). A systematic study of an urban sary to identify those environmental, social and personal foodscape: the price and availability of food in Greater Glasgow. factors that mediate socioeconomic inequalities in diet. Urban Stud 39, 2115–2130. Cummins SC, McKay L, Macintyre S (2005). McDonald’s restaurants and neighborhood deprivation in Scotland and England. Am J Prev Med 29, 308–310. Ecob R, Macintyre S (2000). Small area variations in health related Conflict of interest behaviours; do these depend on the behaviour itself, its measure- ment, or on personal characteristics? Health Place 6, 261–274. Franco M, Diez Roux AV, Glass TA, Caballero B, Brancati FL (2008). The authors declare no conflict of interest. Neighborhood characteristics and availability of healthy foods in Baltimore. Am J Prev Med 35, 561–567. Giskes K, Turrell G, van Lenthe FJ, Brug J, Mackenbach JP (2006). A multilevel study of socio-economic inequalities in food choice Acknowledgements behaviour and dietary intake among the Dutch population: the GLOBE study. Public Health Nutr 9, 75–83. Giskes K, Van Lenthe FJ, Brug J, Mackenbach JP, Turrell G (2007). We are grateful to Rebecca Roberts for assistance with Socioeconomic inequalities in food purchasing: the contribution geospatial analyses and Nick Andrianopoulos for advice on of respondent-perceived and actual (objectively measured) price the statistical approach. This study was funded by the and availability of foods. Prev Med 45, 41–48. Glanz K, Basil M, Maibach E, Goldberg J, Snyder D (1998). Why Australian Research Council (DP0665242) and the National Americans eat what they do: taste, nutrition, cost, convenience, Heart Foundation of Australia (G02M 0658). LT is supported and weight control concerns as influences on food consumption. by a National Health and Medical Research Council Capacity J Am Diet Assoc 98, 1118–1126. Building Grant, ID 425845; KB is supported by a National Glanz K, Sallis JF, Saelens BE, Frank LD (2005). Healthy nutrition Health and Medical Research Council Senior Research environments: concepts and measures. Am J Health Promot 19,330–333. Guthrie JF, Lin BH, Frazao E (2002). Role of food prepared away from Fellowship, ID 479513; and D is supported by a VicHealth home in the American diet, 1977–78 versus 1994–96: changes and Research Fellowship. consequences. J Nutr Educ Behav 34, 140–150. Latham J, Moffat T (2007). Determinants of variation in food cost and availability in two socioeconomically contrasting neighbourhoods of Hamilton, Ontario, Canada. Health Place 13, 273–287. References Magarey A, McKean S, Daniels L (2006). Evaluation of fruit and vegetable intakes of Australian adults: the National Nutrition Andreyeva T, Blumenthal DM, Schwartz MB, Long MW, Brownell KD Survey 1995. Aust NZ J Public Health 30, 32–37. (2008). Availability and prices of foods across stores and Moore LV, Diez Roux AV, Nettleton JA, Jacobs Jr DR (2008). neighborhoods: the case of New Haven, Connecticut. Health Associations of the local food environment with diet quality—a Affairs 27, 1381–1388. comparison of assessments based on surveys and geographic Australian Bureau of Statistics (2003). Census of Population and information systems: the multi-ethnic study of atherosclerosis. Housing: Socio-Economic Indexes for Areas (SEIFA). Australian Bureau Am J Epidemiol 167, 917–924. of Statistics: Canberra. Ness AR, Powles JW (1997). Fruit and vegetables, and cardiovascular Ball K, Crawford D, Mishra G (2006). Socio-economic inequalities in disease: a review. Int J Epidemiol 26, 1–13. women’s fruit and vegetable intakes: a multilevel study of Pearce J, Blakely T, Witten K, Bartie P (2007). Neighborhood individual, social and environmental mediators. Public Health deprivation and access to fast-food retailing: a national study. Nutr 9, 623–630. Am J Prev Med 32, 375–382. Ball K, Jeffery RW, Crawford DA, Roberts RJ, Salmon J, Timperio AF Pearce J, Day P, Witten K (2008). Neighbourhood provision of food (2008). Mismatch between perceived and objective measures of and alcohol retailing and social deprivation in urban New physical activity environments. Prev Med 47, 294–298. Zealand. Urban Pol Res 26, 213–227. Ball K, Timperio A, Crawford D (2009). Neighbourhood socio- Pereira MA, Kartashov AI, Ebbeling CB, Van Horn L, Slattery ML, economic inequalities in food access and affordability. Health Jacobs Jr DR et al (2005). Fast-food habits, weight gain, and insulin Place 15, 578–585. resistance (the CARDIA study): 15-year prospective analysis. Lancet Baron RM, Kenny DA (1986). The moderator–mediator variable 365, 36–42. distinction in social psychological research: conceptual, strategic, Powell LM, Zhao Z, Wang Y (2009). Food prices and fruit and and statistical considerations. J Pers Soc Psychol 51, 1173–1182. vegetable consumption among young American adults. Health Buijsse B, Feskens EJ, Schulze MB, Forouhi NG, Wareham NJ, Sharp S Place 15, 1064–1070. et al (2009). Fruit and vegetable intakes and subsequent changes in Roos E, Talala K, Laaksonen M, Helakorpi S, Rahkonen O, Uutela A body weight in European populations: results from the project on et al (2008). Trends of socioeconomic differences in daily vegetable Diet, Obesity, and Genes (DiOGenes). Am J Clin Nutr 90, 202–209. consumption, 1979–2002. Eur J Clin Nutr 62, 823–833. Cade J, Thompson R, Burley V, Warm D (2002). Development, Shohaimi S, Welch A, Bingham S, Luben R, Day N, Wareham N et al validation and utilisation of food-frequency questionnaires—a (2004). Residential area deprivation predicts fruit and vegetable Public Health Nutr review. 5, 567–587. consumption independently of individual educational level and Cairns S (1995). Travel for food shopping: The Fourth Solution. occupational social class: a cross sectional population study in the Traffic Eng Control 36, 411–418. Norfolk cohort of the European Prospective Investigation into Cerin E, Mackinnon DP (2009). A commentary on current practice in Cancer (EPIC-Norfolk). J Epidemiol Commun Health 58, 686–691. mediating variable analyses in behavioural nutrition and physical Sooman A, Macintyre S, Anderson A (1993). Scotland’s health—a activity. Public Health Nutr 12, 1182–1188. more difficult challenge for some? The price and availability of Crockett EG, Clancy KL, Bowering J (1992). Comparing the cost healthy foods in socially contrasting localities in the west of of a Thrifty Food Plan market basket in three areas of New York Health Bull (Edinb) State. J Nutr Educ 24, 71–78. Scotland. 51, 276–284.

European Journal of Clinical Nutrition Neighbourhood-socioeconomic variations in diet LE Thornton et al 1432 Subar AF, Heimendinger J, Patterson BH, Krebs-Smith SM, Pivonka E, Turrell G, Bentley R, Thomas LR, Jolley D, Subramanian S, Kavanagh Kessler R (1995). Fruit and vegetable intake in the United States: AM (2009). A multilevel study of area socio-economic status and the baseline survey of the Five A Day for Better Health Program. food purchasing behaviour. Public Health Nutr 12, 2074–2083. Am J Health Promot 9, 352–360. Winkler E, Turrell G, Patterson C (2006). Does living in a Tamers SL, Agurs-Collins T, Dodd KW, Nebeling L (2009). US and disadvantaged area entail limited opportunities to purchase fresh France adult fruit and vegetable consumption patterns: an fruit and vegetables in terms of price, availability, and variety? international comparison. Eur J Clin Nutr 63, 11–17. Findings from the Brisbane Food Study. Health Place 12, 741–748. Thornton LE, Bentley RJ, Kavanagh AM (2009). Fast food purchasing Xyris Software FoodWorks Professional. Xyris Software: Brisbane, 2007. and access to fast food restaurants: a multilevel analysis of Zenk SN, Lachance LL, Schulz AJ, Mentz G, Kannan S, Ridella W VicLANES. Int J Behav Nutr Phys Activ 6, 28. (2009). Neighborhood retail food environment and fruit and Turrell G, Giskes K (2008). Socioeconomic disadvantage and the vegetable intake in a multiethnic urban population. Am J Health purchase of takeaway food: a multilevel analysis. Appetite 51, 69–81. Promot 23, 255–264.

European Journal of Clinical Nutrition