Introduction

Worldwide, the proportion of adults with a body-mass index (BMI) of 25 kg/m2 and above increased between 1980 and 2013 from 29 to 37% in men, and from 30 to 38% in women (Ng et al. 2014). The increase in cheap, energy-dense, processed food products has contributed to increasing obesity rates over the past four decades (Vandevijvere et al. 2015). Obesity has increased rapidly over the last 20 years in New Zealand in conjunction with profound changes in the food environment (Ministry of Health 2015). The food environment is defined as the collective economic, policy and social surroundings, opportunities and conditions that influence people’s food and beverage choices and nutritional status (Swinburn et al. 2013). The current food environment promotes obesity and over-consumption of energy-dense nutrient-poor foods (Swinburn et al. 1999, 2011).

Community retail food environments, including the number, density, type, and location of different types of food outlets, may be one factor contributing to high obesity rates in New Zealand. Current evidence is suggestive of an association between community retail food environments and dietary outcomes; however, substantial heterogeneity in study designs, methods and measurement tools makes it difficult to draw firm conclusions (Ni Mhurchu et al. 2013). A wide range of studies investigated the relationships between community retail food environments and area socioeconomic deprivation (Apparicio et al. 2007; Gordon et al. 2011; Pearce et al. 2007a, b, 2008a; Short et al. 2007; Walker et al. 2010). Geographic areas with disproportionate access to outlets with an overabundance of high-energy nutrient-poor foods are described as ‘Food Swamps’ (Luan et al. 2015; Osorio et al. 2013; Rose et al. 2009).

These geographic clusters have usually been assessed through mapping areas according to income or ethnicity and overlaying spatial data on distribution of different types of food retailers (Beaulac et al. 2009). Internationally, studies (Burns and Inglis 2007; Pearce et al. 2007a; Reitzel et al. 2014) have found higher clustering of unhealthy food outlets in more deprived areas than in less deprived areas, yet few studies have characterised these areas as food swamps.

North American studies provide the most support for the existence of food swamps (Hager et al. 2016). Less support for the existence of food swamps has been found in the United Kingdom (Cetateanu and Jones 2014; Fraser et al. 2012). A study conducted in a small deprived urban area in Scotland found that the residents were required to travel more than 4-miles outside the town to purchase healthy food choices from large 24-h supermarkets, while a cluster of takeaway outlets were more readily accessible in the town (Sauveplane-Stirling et al. 2014).

Previously Pearce (Pearce et al. 2007a) found a significant negative association between proximity to the closest fast food outlet and neighbourhood socioeconomic deprivation in New Zealand. The travel distances to unhealthy outlets were at least twice as far in the least deprived neighbourhoods compared to the most deprived neighbourhoods. Higher accessibility was also observed for both convenience stores and supermarkets in the most deprived communities compared to the least deprived communities (Pearce et al. 2008b). A recent national study explored proximity of unhealthy food retailers to schools and found approximately 68% urban schools had a convenience store within 800 m. Access to unhealthy foods through Fast Food/Takeaway or Convenience (FFTC) outlets was greater for the most deprived schools versus the least deprived schools (Vandevijvere et al. 2016). Pearce et al. (Day et al. 2013) found that between 1966 and 2006, the median number of supermarkets/grocery stores within 800 m of schools decreased from 5 to 1, convenience stores decreased from 2 to 1, and fast food outlets increased from 1 to 4.

The objective of this study was to comprehensively assess community retail food environments at a national level in New Zealand, particularly the density of and proximity to different types of food outlets in and from more and less socioeconomically deprived areas.

Methods

In this cross-sectional survey, covering the entire New Zealand territory, we excluded the off shore islands for which the population and the number of food outlets is extremely low.

Data sources

The NZ Meshblocks (MBs) are the smallest spatial unit, holding census information about, on average, 100 individuals. There are a total of 46,637 MBs in the 2013 dataset from Statistics NZ (Statistics New Zealand 2017b). For the purposes of this study, proximity measures were calculated from MB centres. All MBs aggregate to Census Area Units (CAUs). There are a total of 2020 CAUs in NZ, containing a population of about 1700–3000 individuals each. For the purposes of this study (relative) density measures were calculated within CAUs. Rural CAUs were defined as areas with a population of ≤999 people, and urban CAUs as areas with a population of ≥1000 people, consistent with definitions from Statistics NZ (Statistics New Zealand 2017a). CAUs aggregate to define regional Councils.

Information on area socioeconomic deprivation for MBs and CAUs was retrieved from Statistics New Zealand in June 2014.

We used The 2013 New Zealand Index of Deprivation (NZDep2013) which combines eight dimensions of deprivation: communication, income, employment, qualifications, owned home, support, living space, and transport. NZDep2013 apportions each MB and CAU into a decile of deprivation, with Decile 1 representing the 10% of areas with the lowest levels of deprivation, while Decile 10 depicts the most deprived 10% of areas across NZ. In this study we aggregated the Deciles into Quintiles, in which, quintile 1 (Decile 1 and 2) represents the 20% least deprived areas and the fifth quintile (Decile 9 and 10) represents the 20% most deprived areas in New Zealand.

Information on population density was sourced from Statistics New Zealand on 1st September 2014. The dataset includes all people who usually live in and were present in NZ on census night, while visitors from overseas, visitors from elsewhere in NZ, and residents temporarily overseas on census night are excluded. The population data by MB were randomly rounded to the nearest 3 by Statistics New Zealand to protect citizen confidentiality in accordance with the Statistics Act (1975).

The NZ road network, a line shape file containing a total of 73,070 line features, was retrieved from Land Information New Zealand (LINZ) and last updated in August 2014. The compilation of the New Zealand list of food outlet addresses was performed using two main data sources: The 2014 food outlet lists from the 66 City and District Councils—the primary data source—and Zenbu.com, which is a supplementary online source with location data on food premises in New Zealand.

Food outlet categorization

A food outlet classification tool (Supplementary material 1) was developed to automatically categorise the 23,939 food outlets based on their retail name in the compiled list into 14 food retail classes: ‘bakery’, ‘café’, ‘confectionary’, ‘dairy’, ‘farmer’s market’, ‘fast food’, ‘fish and meat’, ‘fruit and vegetables’, ‘takeaway’, ‘juice bars’, ‘restaurant’, ‘supermarket’ ‘service station’, and ‘specialist store’.

To correct for misclassification any ‘excluded’ (e.g., places with licence but food service not being their primary form of business, such as hotels or sport clubs) and ‘not found’ outlets (those that did not automatically match to one of the categories through using the tool) were manually reviewed (38.5%). After classification by type and exclusion of irrelevant food outlets such as hotels and sport clubs, 18,140 outlets were included in the final list (Fig. 1).

Fig. 1
figure 1

Flow chart of food outlet categorisation and geocoding process (FFT fast food and takeaway outlets, SFV supermarkets and fruit and vegetable stores) in New Zealand (National study on community retail food environments, New Zealand, 2014–2015)

Full size image

Geocoding and spatial validation

The Google Geocoding application programming interface was used to retrieve the coordinates of each food outlet. A 95% matching confidence was aimed to ensure that the coordinates generated were as accurate as possible. Spatial validation—the process of confirming the accuracy of geocodes—was undertaken to ensure if the coordinates of the food outlets mapped were reasonably accurate (i.e., the food outlet locations fall within their Council area). Any outlets that were identified as located outside of the Council boundaries were removed from the analysis (<2%).

Validation of the compiled food outlet list

The validation process involved randomly selecting 1% of all classified and geocoded food outlets, stratified by Council and outlet type. Retailers were called by phone to check whether their address and type of food outlet were correct.

Exclusion of food outlets

For the purposes of the study, ‘café’, ‘fish and meat’, ‘juice bar’, ‘restaurant’ and ‘specialist store’ outlets were excluded since they were not considered as either ‘healthy’ or ‘unhealthy’ outlets. Farmer’s markets were also excluded since they are usually only available once a week and completeness and accuracy of the farmer’s markets list could not be verified. The retail classes ‘bakery’, ‘confectionary’ ‘dairy shops’ and ‘service stations’ were aggregated into one outlet category ‘convenience stores’, for the purposes of this study. The total number of food outlets included in this study (comprising supermarkets, fruit and vegetable stores, convenience stores and fast food/takeaway outlets) was 9674 (Fig. 1).

Spatial and statistical analysis

ArcMap version 10.2 and SAS 9.3 software were used to perform the spatial and statistical analysis, respectively. However, NZ MB units are too small in scale to examine food retail density, so this study used two different spatial units (MBs and CAUS) to examine proximity (from meshblock centres) and density (within CAUs).

To measure proximity, the Network Analysis closest facility tool was used on ArcMap 10.2 and road proximity was calculated from each MB centroid to the closest retail facility, separately for each food retail outlet type. To measure density, a spatial join was used to count the availability of the different types of food outlets within each CAU. The results were adjusted for population density and expressed as the number of outlets per 1000 population. Kruskal–Wallis tests were conducted to compare density and proximity measures between quintiles of area socioeconomic deprivation.

A Retail Food Environment Index (RFEI) was calculated for each CAU by dividing the number of unhealthy outlets (convenience stores, fast food/takeaway outlets) by the number of healthy outlets (supermarkets, fruit and vegetable stores) for each CAU. A higher RFEI suggests a greater relative density of unhealthy food outlets.

The GeoDa platform was used to assess spatial autocorrelation, or the extent to which spatial data from nearby locations are more likely to be similar than the data from distant locations. It is based on the Local Moran’s I statistic (Anselin et al. 2006). The relative density of unhealthy food outlets (as per the RFEI) and area socioeconomic deprivation level of each census area unit were compared to those from neighbouring CAUs. Neighbours were defined as CAUs sharing a common boundary. If the majority of CAUs are similar to one another than this is considered positive autocorrelation (Moran I value of 0.3 infers strong correlation). The spatial scan statistic (Kulldorff 1997) was used to identify areas of significant geographical clustering of unhealthy food outlets (fast food outlets, takeaways and convenience stores) taking into account area deprivation level and population density. A particular strength of the scan statistic is its ability to recognise statistically the spatial dependence of adjacent areas while enabling analyses to be adjusted for different covariates, in this case area socioeconomic deprivation. The automated procedure involves moving numerous windows of different shapes and sizes over a grid reference map over New Zealand to which food outlets (incidents) and the denominator populations were attached. The window radius continuously changes from zero to a user defined maximum, based on the denominator population, which was defined in this analysis as 50% of the population at risk as recommended by Kulldorff (1997). For each window, at every location the log likelihood ratio was calculated to determine whether the relative density of unhealthy food outlets (as per the RFEI) within a circle differs significantly from the rate outside the circle. Next, a Monte Carlo simulation was performed by the software to test the statistical significance of the likely clusters of unhealthy food outlets in New Zealand. Significant clusters were defined as those with a probability p value of <0.05.

Results

Validation of the compiled food outlet dataset

Out of the 18,140 retail addresses, 16,166 (90%) were geocoded successfully (Fig. 1). Overall, a spatial validation confirmed that 98% of the coordinates were correctly mapped. Those 2% of incorrectly mapped outlets were examined further using Google Street View and were found to be residential buildings.

These addresses were assumed to be mailing addresses of shop owners and were, therefore, excluded from further spatial analysis, which left a total of 15,802 outlets after spatial validation. A random sample of outlets was further verified by phone calls. Of the 182 phone calls made, 157 were answered by shop attendants. About 2% had incorrect addresses and 3% had a completely different primary service. Overall, 90% of the total answered calls by retailers confirmed correct address and classification. In the final food outlet dataset (N = 9674), after excluding the outlets not relevant for this study, about 42.2% of food outlets were convenience stores, 44.6% fast food outlets and takeaways, and 13.1% fruit and vegetable shops and supermarkets (Fig. 1).

Proximity

The median road network distance from meshblock centres to the closest outlet for all types of food retailers was significantly greater in rural areas than in urban areas. The median travel distance from meshblock centres to the closest outlet for all types of food retailers was about twice as far in the least deprived areas compared with the most deprived areas (Tables 1, 2).

Table 1 Median travel distance (metres) to the closest food outlet in urban New Zealand neighbourhoods (from meshblock centres) stratified by level of area socioeconomic deprivation as per the New Zealand Deprivation Index 2013 (NZDep2013) (National study on community retail food environments, New Zealand, 2014–2015)
Full size table
Table 2 Median travel distance (metres) to the closest food outlet in rural New Zealand neighbourhoods (from meshblock centres) stratified by level of area socioeconomic deprivation as per the New Zealand Deprivation Index 2013 (NZDep2013) (National study on community retail food environments, New Zealand, 2014–2015)
Full size table

The gradient of increasing median distance from meshblock centres to all types of food outlets with decreasing level of area socioeconomic deprivation was found to be statistically significant for both urban and rural areas (Tables 1, 2).

Density and relative density

Overall, the results show that higher socioeconomic area deprivation was associated with higher food retail outlet availability for all outlet types (p < 0.001) (Table 3). Areas in the most deprived quintile had 73% higher availability of fast food and takeaway outlets, 64% higher availability for convenience stores and 66% higher availability of supermarket and fruit and vegetable stores compared to areas in the least deprived quintile. The spatial regression confirmed a statistically significant relationship between area socioeconomic deprivation and the density of unhealthy outlets per 1000 Census Area population (Table 4). There was also a statistically significant relationship between area socioeconomic deprivation and the Retail Food Environment Index (RFEI) (Table 4).

Table 3 Mean Availability [standard error (SE)] of different types of food outlets (expressed per 10000 people) in New Zealand Census area Neighbourhoods Stratified by New Zealand Deprivation Index 2013 (NZDep2013) Quintile; (National study on community retail food environments, New Zealand 2014–2015)
Full size table
Table 4 Spatial regression results on the relationship between area socioeconomic deprivation and the density of food outlets in New Zealand (National study on community retail food environments, New Zealand, 2014–2015)
Full size table

Food swamps in New Zealand

We calculated the Moran’s I measure of spatial autocorrelation to test for food swamp areas in New Zealand. The Moran’s I value of 0.01 indicated a statistically significant, but weak positive spatial autocorrelation between neighbouring CAUs. The spatial scan analysis found 114 potential food swamps (clusters), namely areas of differing size and shape, with significantly higher relative density of unhealthy outlets than other areas around NZ (Supplementary Material 2). Each cluster contained a different number of CAUs. In total, 722 CAUs were found within the 144 potential food swamps.

There were about 14% more potential food swamp CAUs in the most deprived CAUs (n = 147) compared to the least (n = 126) deprived CAUs (Fig. 2).

Fig. 2
figure 2

Number of food swamps by New Zealand Deprivation Index Quintiles [higher RFEI (Retail food environment index) indicates a higher relative availability of unhealthy food outlets] (National study on community retail food environments, New Zealand, 2014–2015). NZDep New Zealand Deprivation Index

Full size image

Discussion

Summary of results

A significantly positive association was observed between the level of area socioeconomic deprivation and density of all types of food retailers in NZ, corrected for population size. A significantly negative association was observed between the level of area socioeconomic deprivation and proximity to all types of food retailers from MB centre points. Across New Zealand, 147 CAUs were identified as potential food swamps in the most deprived areas of NZ. The observed differences in distance to each outlet type from MB centres between more and less deprived areas were statistically significant and are consistent with findings of other NZ studies (Pearce et al. 2007a, 2008b) that found similar relationships between distance to fresh produce stores and fast food outlets and area deprivation.

Such findings also coincide with findings in the United Kingdom (Smith et al. 2010). These findings suggest that people who live and work in more deprived neighbourhoods in NZ are likely to experience a higher relative exposure to unhealthy outlets, which increases the likelihood of unhealthy food consumption.

The results of this study suggest that there are mechanisms other than population density which might be encouraging more unhealthy food outlets in poorer areas.

It is known from other studies that companies target more deprived communities through their marketing practises. (Grier and Kumanyika 2008; Isgor et al. 2016; Powell et al. 2014; Settle et al. 2014). Therefore, it is equally possible that companies strategically target the development of new unhealthy outlets in lower socioeconomic neighbourhoods.

Policy implications

In NZ, the zoning laws are designed to permit activity reflective of their anticipated land use (Swinburn et al. 2014). Commercial activities of all kind are permitted and can take place without Council consent. Councils do not regulate commercial activity unless it is impacted by other regulations such as the Hazardous Substances Legislation that controls the number and certain type of goods which can be stored. At present, the NZ Government has no specific rules for regulating the number and location of food outlets. However, the Health Act 1956 imposes on Councils a general duty to improve, promote and protect public health.

Councils have the ability to produce bylaws under the Local Government Act 2002. A bylaw allows plan changes to development and subsequently zoning laws.

Based on the evidence from this study, the following policy recommendations are made following the results from this study:

  • Introducing legislation that allows considerations like health and wellbeing (not just safety) to be taken into account in zoning and planning permissions.

  • Developing guidelines for urban planners and Councils on the promotion and favourable zoning treatment for fruit and vegetable store outlets.

  • Engagement of communities and health promoters with convenience store owners to increase the healthiness of the in-store food availability.

  • Identification of high risk communities and restriction of future introduction of fast food outlets through zoning rules. In such cases, a proposal to develop an outlet in a high risk area would then require notified resource consent. This would give an opportunity for community members to express their views on the potential of a fast food outlet opening in their neighbourhood.

Strengths and Limitations

This study includes a number of limitations. This study only evaluates one dimension of accessibility which relates to spatiality (e.g., proximity and density). It is well understood that accessibility is more than just a component of distance over space (Penchansky and Thomas 1981). The phone call validation recognised that the categorisation of outlets does not always reflect the true nature of the retailer and the reality of their service. Only 1% of outlets were validated in this way in this study. However, a large portion of the dataset, approximately 4000 outlets, was manually checked through Google Street View. In addition, a large number of food outlets (such as restaurants and cafes) were excluded because there is no agreement among experts about whether these should be classified as ‘healthy’ or ‘unhealthy’ in New Zealand. Another major limitation was using predefined spatial units to define neighbourhoods. This is related to the well documented Modifiable Areal Unit Problem. In New Zealand, the most readily available spatial unit for research are MBs and CAUs. However, MBs contain very few people, while CAUs might be too large to be described as a community.

While there are a number of noted limitations, there are also significant strengths of this study. This study had the largest assembled food outlet list amongst non U.S studies. It was the first New Zealand study to use spatial validation techniques to assess the accuracy of the geocoding process successfully. This is the first survey of community food environments to account information from all 66 Councils in New Zealand. The study developed an effective, automatic self-classification tool to categories outlets.

In conclusion, this study has found evidence to support the potential existence of food swamps in New Zealand. Policy implications include restricting the introduction of new unhealthy food outlets in high risk areas with a significantly higher relative density of those outlets than in other areas in New Zealand