A methodological commentary formalizing the statistical assumptions behind “isocaloric substitution” models (e.g. replacing eggs with food X while holding total energy constant): the substituted-food coefficient is not a free-standing causal contrast but depends on which reference category is omitted from the regression and on the composition of the rest of the diet, and can reverse sign under seemingly minor respecification. Argues such analyses should always report the explicit reference food and be treated as hypothesis-generating, not as causal point estimates, when used to draw dietary conclusions.
relevance_note: The primary methodological artifact explaining the mechanism behind why “eggs vs comparator X” substitution results flip sign in the egg literature.