Background Little is known about whether relative or absolute measures of the neighbourhood food environment better predict dietary outcomes. Little is also known about whether using spatially explicit regression models such as Geographically Weighted Regression (GWR) may improve our ability to detect neighbourhood effects. Here we test i) whether relative or absolute measures of the food environment better predict fruit and vegetable intake and ii) whether using a GWR approach improve model fit and reveal spatial variations in exposure-effect relationships undetectable in global models.
Methods Individual adult data from the baseline of The ORiEL Study (n = 980) were used. Kernel density estimations were used to calculate relative and absolute density metrics of the food environment using an adaptive bandwith of 5%. First we modelled whether density measures predicted fruit and vegetable consumption using linear regression. Second, we modelled the same relationship using geographically weighted regression. We controlled for socio-demographic covariates.
Results We found little evidence for the effect of absolute neighbourhood food environment measures on fruit and vegetable intake, but relative measures performed strongly for fruit (β=0.086, p = 0.009) and vegetable intake (β=0.088, p = 0.007). A 10 percentage point increase in the proportion of healthy outlets was associated with an increase of 1.74 portions of fruit and vegetable per day. Use of GWR revealed non-stationarity in exposure-effect relationships for fruit and vegetable intake and improved model fit (a reduction in the Akaike’s Information Criterion across all models). Cartographic presentation of exposure-effect relationships revealed large spatial variations in exposure-effect relationships.
Discussion Relative measures perform better than absolute measures of the food environment in predicting fruit and vegetable intake. The use of spatial regression approaches reveal large localised environmental effects on diet that are obscured if global regression models are used.
- fruit and vegetables