Correctly calculating the cost of food

It is important for readers to be aware that graphs shown by Drewnowski in a recent commentary in this journal and previous publications carry a serious statistical artifact.

Abstract

Correctly calculating the cost of foodIt is important for readers to be aware that graphs shown by Drewnowski in a recent commentary in this journal 1 and previous publications 2-4 carry a serious statistical artifact.Since energy is included in the numerator on one axis (kcal/g) and in the denominator on the other axis ($/kcal), a negative auto-correlation is created.To demonstrate this artifact, we took the energy content (measured in kcal/g) and weights for 212 foods from a supermarket in Melbourne, Australia (2008 data) and applied a random number for the price of each food.The foods included a range of products from different categories in the supermarket including breads, cereals, biscuits, frozen meals, milk, and yoghurt.The resulting scatterplot, illustrated in figure 1, exhibits the same negative autocorrelation as the figure presented by Drewnowski. 1 At the supermarket, shoppers are likely to face the price signal of $/g or $/pack rather than $/kcal.Nevertheless, it is certainly a plausible hypothesis that the "cost per calorie" is an important reason that people, especially poorer people, make particular food choices.More studies in this area are needed to support Drewnowski's hypothesis, rather than relying on the negative relationship between kcal/g and $/kcal.Furthermore, there are likely to