Additionally, it would appear that the Cornell data set, coming from the Census Bureau, suffers from the typical problem in most of these data sets -- it is income only, not income net of taxes and transfers. About 1/2 of the US population pays no direct income tax, and of that 1/2 many receive a large share of transfer from government in the forms of Medicaid, food stamps, EITC, etc. After taking into account all of those non-taxes and transfers, this group is much better off than a lot of the publicly available data sets would indicate. This is mostly due to the difficulty of getting all of that information together on each individual in the data set, whereas looking at a reported W2 value is comparatively easy.
