You are viewing an obsolete version of the DU website which is no longer supported by the Administrators. Visit The New DU.
Democratic Underground Latest Greatest Lobby Journals Search Options Help Login

Reply #76: A dead Greek wrote, elsewhere: [View All]

Printer-friendly format Printer-friendly format
Printer-friendly format Email this thread to a friend
Printer-friendly format Bookmark this thread
This topic is archived.
Home » Discuss » Topic Forums » Election Reform Donate to DU
Febble Donating Member (1000+ posts) Send PM | Profile | Ignore Wed Nov-23-05 02:21 PM
Response to Reply #59
76. A dead Greek wrote, elsewhere:
Edited on Wed Nov-23-05 02:34 PM by Febble
"So redshift, and "swing" (Bush's performance relative to 2000) ought to be positively correlated."

Once again, she substitutes a new thesis instead of debating the old one. Yeah? Prove it.

Conveniently, we can't.... We got no data.

Or, you could prove that "redshift, and 'swing'" ought to... shoulda, coulda... be "positively correlated" by proving that correlation independently of this specific fraud "debate". I suggested one way to test Febble's thesis. There are others. Of course, WE can't do it - we still have no data...

Otherwise, all this just smells like another Mitofsky red-herring. OH no... Mitofsky has never introduced one of those into this debate, has he?

I am happy to debate the old one, but I have to correct you: this is not "a new thesis" whatever that means. It is simply a testable hypothesis.

You know this, because we have discussed it before, but for others who might be interested, details are given here:

Further discussion is given here:

Full disclosure: both Mark Lindeman and myself (Elizabeth Liddle) were credited by Mitofsky as contributing to these analyses.

If you read both pieces you will see that there are a number of ways in which the plot could be compatible with fraud, even with widespread fraud; it is, however, very difficult to reconcile the plot with a fraud mechanism that could have contributed substantially to the exit poll discrepancy. It would therefore seem more likely than not that the exit poll discrepancy was caused largely by polling bias rather than fraud.

However BECAUSE the analysis is perfectly compatible with the kinds of electoral injustice that we KNOW happened in 2004, just as they did in 2000, then the analysis does nothing to rule out fraud. The analysis simply says that the magnitude is not indexed by the exit poll discrepancy. It could (probably) be less. It could (conceivably, I suppose) be greater.

And the take-home message is: stop regarding the exit polls as prima facie evidence of a stolen election. There is plenty of other work to be done, and frankly, the exit polls are getting in the way.

And while I'm attuned to the other world, let me channel another of the dead:

"Finally, ESI examined whether the proportions of the vote that Bush
received in each precinct, in 2000 and 2004, were related to the difference
between the reported vote and exit poll results for those same precincts. If
systematic fraud or error in vote counting occurred in 2004 but not in 2000,
Bush would have done significantly better in those precincts in 2004 {than
in 2000} and we would see larger differences between the reported vote and
the exit poll in those precincts."

The premise is that there was ZERO fraud in 2000.
Umm, really?

OK, let's do a little model to test it out.

Assume this hypothetical example:
1) 10 precincts comprised the total electorate.
2) Fraud accounted for 70% of the exit poll discrepancies in 2000 and 2004.
3) Bush's vote declined in 5 of the 10 precincts from 2000 to 2004.

Bush 2000 Bush 2004
Prct Rec Exit Fraud Fraud TRUE Chg Rec Exit Fraud Fraud TRUE
Vote %Dev %Dev Dev Vote 2000 Vote %Dev %Dev Dev Vote
1 55% 2% 70% 1.40% 53.60% -1% 54% 4% 70% 2.80% 51.20%
2 40% 2% 70% 1.40% 38.60% -1% 39% 4% 70% 2.80% 36.20%
3 49% 1% 70% 0.70% 48.30% 3% 52% 3% 70% 2.10% 49.90%
4 44% 1% 70% 0.70% 43.30% 4% 48% 3% 70% 2.10% 45.90%
5 53% 1% 70% 0.70% 52.30% -4% 49% 3% 70% 2.10% 46.90%
6 58% 1% 70% 0.70% 57.30% 5% 63% 3% 70% 2.10% 60.90%
7 40% 2% 70% 1.40% 38.60% 6% 46% 4% 70% 2.80% 43.20%
8 57% 1% 70% 0.70% 56.30% -1% 56% 3% 70% 2.10% 53.90%
9 45% 1% 70% 0.70% 44.30% 3% 48% 3% 70% 2.10% 45.90%
10 48% 1% 70% 0.70% 47.30% -1% 47% 3% 70% 2.10% 44.90%

48.90% 1.30% 70.00% 0.91% 47.99% 1.30% 50.20% 3.30% 70.00% 2.31% 47.89%


In 2000: In 2004:
Bush LOST the election with 48.90% Bush WON the election with 50.20%
Bush LOST the TRUE Vote with 47.99% Bush LOST the TRUE Vote with 47.89%

Bush won 4 precincts, lost 6 Bush won 4 precincts, lost 6

I believe the citation is from this document:

In which the hypothesis I referred to is incorrectly stated.

See the second of the links above for clarification.

However, the channelled spirit correctly spots one of the fraud mechanisms that could account for the exit poll discrepancy and not produce a swing-shift correlation: If fraud in 2004 was precisely calibrated to the fraud that occurred in 2000 (and I mean precisely), then yes, that might do it. If every extra 1 percent in Bush's vote share due to fraud in 2000 was mirrored by X times 1% fraud in 2004, yup, maybe you could pull it off. An alternative would be uniform fraud in all precincts. Another alternative is fraud confined to precincts in which Bush was anticipated to do badly relative to 2000 (although you would have to be careful to cover a substantial majority of precincts and be sure to avoid any where he was doing well).

If anyone can suggest an practical algorithm that could achieve this, bearing in mind that it in all NEP precincts where the vote counts for are collected at the precinct, that the fraud must be perpetrated at precinct level, not at tabulator level (about 60% of precincts) while the remainder has to be done at tabulator level, then I will concede, yes, perhaps the exit poll discrepancy was due to fraud.

But remember - there is very little room for any deviation from the fraud perpetrated in 2000. It has to be directly proportional. Any variance and it will show up in a swing-shift correlation. I challenge my channelled spirit to produce a model of 1250 precincts (not 10) with a realistic distribution of WPEs, vote shares, counting methods and swing values in which the WPE (or any other measure of bias) is not correlated with a measure of swing, and yet fraud is responsible for both (I'm happy with 70% shared variance, or even less). Oh, and you can't end up with a large mean WPE in 2000 because there wasn't one.

Estimates of realistic variance in bias and vote share can be gleaned from the plots linked in this DKos diary:

(edited to replace accidentally deleted word)
Printer Friendly | Permalink |  | Top

Home » Discuss » Topic Forums » Election Reform Donate to DU

Powered by DCForum+ Version 1.1 Copyright 1997-2002
Software has been extensively modified by the DU administrators

Important Notices: By participating on this discussion board, visitors agree to abide by the rules outlined on our Rules page. Messages posted on the Democratic Underground Discussion Forums are the opinions of the individuals who post them, and do not necessarily represent the opinions of Democratic Underground, LLC.

Home  |  Discussion Forums  |  Journals |  Store  |  Donate

About DU  |  Contact Us  |  Privacy Policy

Got a message for Democratic Underground? Click here to send us a message.

© 2001 - 2011 Democratic Underground, LLC