Do differences in reported expenditures between commercial household-based scanner data and government surveys matter in a structural model of food demand?