From feenberg@nber.org Thu Oct 19 15:40:18 2017 Date: Thu, 19 Oct 2017 15:40:18 -0400 (EDT) From: Daniel Feenberg To: Bryan Justin P Cc: Pierce Kevin K Bcc: feenberg@nber.org Subject: RE: data by state (fwd) On Thu, 19 Oct 2017, Bryan Justin P wrote: > > The AGI categories do not add up to the total, so it looks like the data > was deleted, rather than collapsed but kept in the totals. So there > should be at least two sets of ** in any category to avoid disclosure by > subtraction. I worried that the sum provided might differ from the rowsum of the data because of rounding of the data, and I have not thought to impute it to the censored cells. Now I am not sure this is right. > > I'm not sure what year we started collapsing instead of deleting, but if > the parts add up to the totals, then it is collapsed rather than > deleted. If it is collapsed, our rule for collapsing cells for > disclosure is to move them to the next available AGI category that is > lower than the collapsed cell. So just to take your example below, they I think I see exceptions. Look at Wyoming 2015 where Alternative Minimum Tax is collapsed as you describe it should be, but lowest 2 cells for Nonrefundable Education Credits and NIIT are consolidated in the higher income cell. I wrote a lot of code to worry about this. Mostly it shouldn't make a difference, but there may be ambiguities if the rule isn't always followed. Also, the footnote says "Adjacent cell" which I took to mean that in the pattern: "** 40", "**", "**", "** 10","**" The first ** was consolidated with the 40 and the other **'s with the 10. I take it that you don't agree, the 40 consolidates the first 2 **s and the 10 only the last. This pattern occurs in Wyoming 2015 for Additional Medicare Tax, and in other places too. thanks for any clarification dan