Re: The 'working poor' scam

From: sinister (sinister_at_nospam.invalid)
Date: 07/03/04 

Date: Sat, 03 Jul 2004 21:17:51 GMT

"Jim Blair" <see@sig.com> wrote in message
news:cc1ku3$k0t$1@news.doit.wisc.edu...
> royls@telus.net wrote:
> >On Mon, 28 Jun 2004 21:45:14 +0000 (UTC), Jim Blair <see@sig.com>
> >wrote:
> >
> >>To check on the
> >>accuracy of PK's claims about how Hubbard collected his data, I emailed
> >>Glen Hubbard. Krugman and Hubbard disagree about how Hubbard got the
> >>data: Krugman said it is from people who PAID taxes in EACH YEAR from
1979
> >>to 1988. But Hubbard says it is from IRS data on people who FILED a
> >>return in both of the years 1978 AND 1988.
> >>
> >>This is quite a difference in the data base. But who am I to believe?
> >>The guy who did the study? Or the guy who wants to discredit it?
>
> royls@telus.net wrote:
> >
> >It's only a difference of degree in any case.
>
> Hi,
>
> But quite a big difference! Krugman claimed that by screening out anyone
> who missed PAYING income tax for any one year during the interval, about
> half of the population would be excluded.

That sounds about right.

> It was by considering the extra work that would be needed to sift through
> the data to weed out those who missed paying income tax in any one year,

Extra work? That's a few lines of computer code.

> to make the study less valid, that caused me to question Krugman's claim.

The presumption isn't that they deliberately weeded people out to make the
study biased. Rather, it's that tax returns are a handy way to get info
about people's incomes. The problem is that if you're not careful, you bias
the results.

> So I sent an e-mail to Hubbard to ask him about it.

Right...and has Hubbard made the text of the paper available for perusal?
If he hasn't, then his contradiction of Krugman's claims are worthless.

> Any study done over a decade interval (or even a year interval) must
> exclude those who were not present at each end of the studty, and the
> longer the interval the more are excluded. Note that the Gottschalk study
> also excludes both the young and the old. And (I say) these are the two
> most income mobile age agoups.

If I recall correctly, the Gottschalk study---which is readily available on
the Web, in apparent contrast to the Hubbard study---examines the issue of
bias.

> >....Hubbard himself admits
> >that his study eliminated all the data on people whose incomes were
> >too low for them to file in 1978, and stayed that low until 1988. He
> >simply chopped off the bottom of the distribution.
> >
> >-- Roy L
>
> People who never filed with the IRS didn't get counted. But remember that
> lots of people FILE who don't PAY taxes (a point Krugman confuses). Poor
> people file either to get back money that was withheld from their
> paychecks or to claim the Earned Income Tax Credit.
>
> Any study is likely to exclude some people. I mean even the census does.

The problem is the kind of inference one attempts to draw from the data, and
whether the estimators you use to draw the inference are biased. If you
limit yourself to income tax data, you are systematically (not randomly)
excluding a big chunk of the population from the sample. You can either (a)
try to correct for this by looking at other samples, or (b) admit it and
point out that you're dealing with a restricted subpopulation.

Krugman's point was that by restricting to people who filed returns at
two or more time points, you introduce bias that underestimates the number
of people whose income has dropped. This point should be pretty clear by
now.

Yes, the census excludes some people, but the people who design it are well
aware of the issues and make attempts to correct for it by statistical
techniques. The question is whether Hubbard was aware of the issues. I
doubt you are. ;-)

>
> ,,,,,,,
> _______________ooo___(_O O_)___ooo_______________
> (_)
> jim blair (jeblair@facstaff.wisc.edu) Madison Wisconsin
> USA. This message was brought to you using biodegradable
> binary bits, and 100% recycled bandwidth. For a good time
> call: http://www.geocities.com/capitolhill/4834
>
>
>
>

Loading