That has been done in a few studies. But delivering food doesn't mean that study subjects actually ate it, or didn't eat other food. So it doesn't solve the data quality problem.