Abstract
We estimate a Quadratic Almost Ideal Demand System (QUAIDS) for twelve food groups plus tobacco and alcohol-included to capture substitution with addictive goods-using microdata from Brazil's 2017-18 Household Budget Survey, correcting sample-selection bias via bivariate copulas selected empirically by canonical maximum likelihood estimation. We propose a method to derive nutritional income elasticities from a nutritional content matrix-a mapping from food-group demand to nutrient demand that extends any estimated demand system to produce nutritional elasticities without requiring a separate nutrient equation system. With this structure, we specify education-income interaction effects that reveal a pattern consistent with the redistribution-rather than a uniform increase-of nutritional sensitivity to income. Protein is the only nutrient with an income elasticity above unity, while protective micronutrients-vitamin C, fiber, calcium-respond significantly less to income. The main result lies in education: the education gradient is asymmetric-tertiary-educated households exhibit lower income elasticities for protein and zinc, but higher elasticities for 1 vitamin C, fiber, and calcium-so that the same income increment is associated with distinct nutritional returns depending on household education. This asymmetry suggests that the nutritional effectiveness of income transfer programs such as Bolsa Família varies with the educational level of the beneficiary household-a pattern that points to a possible complementarity between income and education in such policies.