Changes in Household Diet: Determinants and Predictability

American diets are, on average, unhealthy relative to nutritional guidelines. We combine household food purchase data with health information and the timing of diet-related news releases to estimate what types of information or education improves diet and for whom. Our primary finding is that the average household does not improve diet quality in response to serious disease diagnosis (diabetes, hypertension, obesity), changes in government diet recommendations, or major research findings. Households with more economic incentives to be healthy higher income, higher education, younger also do not respond significantly. This suggests that dietary choices are difficult to alter. We identify households in the panel who do show large improvements in their diet quality; this is 5 to 6% of the sample. We use a machine learning approach to predicting these households, and identify that the concentration of baseline diet the extent to which it loads on a small number of categories is a strong predictor of diet quality improvement. We comment on policy.