Adjusting for spatial autocorrelation in assessing access to fast food outlets by neighbourhood deprivation
Higher rates of obesity have been observed in more socially disadvantaged neighbourhoods, possibly due to increased exposure to outlets serving energy dense food. Studies worldwide have examined access to fast food outlets by neighbourhood deprivation, with inconsistent findings. However, there exist differences in the analytical approaches used, with correlation between observations often ignored. The need to take spatial autocorrelation into account is increasingly emphasised among neighbourhood and health researchers but little is known about the impact of adjusting for spatial autocorrelation or the implications of failing to do so in studies of equitable access to neighbourhood resources.
Our study examined access to fast food outlets by neighbourhood deprivation in Victoria, Australia, investigating the importance of adjustment for spatial autocorrelation.
METHODS:
Major chain fast food outlets (e.g. McDonald’s, KFC) were identified using company websites and mapped to obtain a count within each neighbourhood in Victoria (N=422). Data were linked to Australian Bureau of Statistics data on neighbourhood deprivation and population size.
We examined associations between fast food outlet availability and deprivation using poisson regression, adjusting for population size. A neighbourhood adjacency matrix was created, defining areas sharing a common boundary as neighbours. Spatial autocorrelation was assessed using Moran’s I.
RESULTS:
There was little spatial autocorrelation between fast food outlets (ρ=0.06, p=0.02). We found evidence of differences in fast food outlets by deprivation; fewer in least deprived neighbourhoods compared to most deprived (incidence rate ratio (IRR)=0.27, 95% confidence interval (CI) (0.12, 0.61)) and second most deprived (IRR=0.27, 95% CI (0.13, 0.60)). There was no residual spatial autocorrelation (p=0.39).
CONCLUSIONS:
Results suggest that spatial regression models are not necessarily required in examining neighbourhood deprivation and fast food access. Further research should examine spatial autocorrelation in the analysis of neighbourhood effects on individual health outcomes in order to in order to ensure correct inferences are made.