For want of a CURF

Some things are just hard to know without decent public use microdata.

There's been a lot of furore about ACT's flat tax proposal, with many on the left outraged that a libertarian party would support lower taxes.

Many have also pointed out that a 17.5% flat tax would represent a tax hike - not just for those currently on the 10.5% lowest tax rate, but also for those who are at a low enough point in the 30% tax band that the inframarginal increase in taxes on the first $14,000 of earnings would outweigh the reduction in taxes on earnings above $48,000.

No quibbles on that part - it's just maths.

But I am a bit more curious about household distributions. How many of those who would see that inframarginal increase in taxes are in households where the other earner would see a real tax cut accompanied by a drop in their tax rate? I remember Pacheco and Maloney's work showing that many minimum wage earners are in higher earning households, so the minimum wage is poorly targeted. I expect there could be similar effects here.

To know that we'd need a two-way earnings table. The columns would show the earnings of the primary household earner, split into different bands. The rows would show the earnings of the secondary household earner, split also into different bands. Each cell would provide the count of households where the primary earner earned the column amount, and the secondary earner earned the row amount.

For households with two earners on lower wages, the flat tax would be a tax increase. For some of those, it would only be an inframarginal tax increase. For others, it could represent a tax increase at the margin (anyone earning less than $14,000 per year).

For households with one higher earner and one lower earner, I'd expect most would see a tax reduction with effects at the margin, but I don't know the numbers.

And households with two higher-earners would definitely see a tax cut.

I guess the substantive point is that a lot of the folks being currently touted as seeing a tax hike under ACT's proposal would be in households where the net effect would be the opposite, and where the tax hike part would be inframarginal while the tax reduction part would have effects at the margin. We don't expect big labour supply effects on primary earners from tax cuts, and we do expect more supply responsiveness from second earners - but if the effect on second earners is more likely to be inframarginal, then that's a bit different.

And the more substantive point is that it's just dumb that it's hard to know what the numbers actually are. In the US, I'd just download the ACS PUMS and tally it. Here, because Stats' back-end systems are archaic, the only tables we have are the ones somebody thought to code in ages ago. You can't dynamically generate them.

Stats does have Confidentialised Unit Record Files, but you have to go through an application procedure to get them. And, once you have them, you can only use them for that one specific thing that you asked to use them for, then delete them. If you want to check something out just out of idle curiosity - like this tax question - well, you can't. Even if you have the CURF from a prior request that you're still working on, you have to get permission to use it to ask this specific question.

And the most recent CURF that might answer the question would be the one from the 2013 Census; they don't CURF the HES.

I have a query in with Stats in case this table already exists somewhere on a disused server in the bottom of a locked filing cabinet stuck in a disused lavatory with a sign on the door saying 'Beware of the Leopard', as that's always a possibility. But this kind of stuff shouldn't take hassling Stats staff (they are exceptionally helpful, and it wouldn't surprise me at all if somebody there is already building the table) or doing a whip-round of other economists to see whether they know whether this kind of data exists. You should just be able to download the darned Census CURF and check it - or have back-end systems that can support the kind of data access that IPUMS provides.

UPDATE: Stats' helpful advisor informs that that cross-tab does not exist and would take either a custom data request, or a research project through the IDI.

UPDATE2: Customised data requests charge out at $155/hour. If Stats' back-end systems weren't rubbish, or if they decided to free the Census CURF, any of us could just do it on our own.