Statistical approaches used to assess the equity of access to food outlets: a systematic review

Abstract

OBJECTIVE: Poor dietary quality, found to be higher amongst those residing in areas of lower socioeconomic position (SEP), is one of the contributors to the obesity epidemic. In recent years, many studies have considered the distribution of food outlets, such as supermarkets and fast food outlets, by neighbourhood-level SEP to identify whether residents in low SEP neighbourhoods have less access to outlets providing healthy food choices and increased access to those offering energy-dense foods. These studies involve spatial data meaning traditional statistical techniques, such as regression, may not be appropriate due to the presence of spatial autocorrelation. Ignoring spatial autocorrelation can lead to imprecise standard errors, affecting the results obtained. The aim of this review was to examine studies of the equity of access to food outlets to determine the suitability of the methodology employed to deal with the distribution of the data. Our secondary aim was to assess whether spatial autocorrelation was considered. METHODS: Searches were conducted in health science databases, including Medline and PsychINFO, for articles published from January 2000 to March 2014. Eligible studies included an objective outcome measure of the neighbourhood food environment, mapped using geographic information system (GIS) software, and a measure of neighbourhood-level SEP. Neighbourhoods had to have been defined using small area measures such as census blocks or postcode districts. RESULTS: Fifty-four papers were included. Outlet accessibility was typically defined as either the distance to the nearest outlet from the neighbourhood geographic or population-weighted centroid, or as the number of food outlets within a neighbourhood (or pre-specified buffer), some adjusting for population or area size. Statistical analysis techniques adopted included one-way analysis of variance, correlation, and Poisson or negative binomial regression. Although the majority of studies featured geographically contiguous areas, few considered spatial analytical techniques or adjusted for spatial autocorrelation in analysis. CONCLUSIONS: With advances in GIS software, it is possible to consider sophisticated measures of neighbourhood outlet accessibility. However, approaches to statistical analysis often appear less well-considered. Care should be taken to consider the spatial nature of the data and the possibility of correlated residuals which could affect the results obtained.