Vote this up DUers! New Paper By USCV Dopp Baiman, Simon, Mitteldorf

[Melissa G]

http://electionarchive.org/ucvAnalysis/US/Presidential-Election-2004.pdf
Stevepol's DU discusion here..
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x398055

From the conclusion...
Conclusion: Evidence for Vote Miscount in the 2004 Presidential Election

The possibility that the 2004 election exit poll discrepancy was caused by vote miscount has become increasingly credible as successive (E/M and ESI) reports claiming support for exit-poll error have instead provided more evidence for vote miscount.

The nonpartisan U.S. Government Accounting Office (GAO) in its September 2005 report "ELECTIONS --Federal Efforts to Improve Security and Reliability of Electronic Voting Systems Are Under Way,
but Key Activities Need to Be Completed"50 on page 38 said,

"...there is evidence that some of these concerns —including weak controls and inadequate testing—have caused problems with recent elections, resulting in the loss and miscount of
votes."

The Ohio precinct-level exit poll data that was recently released shows highly irregular patterns of exit poll discrepancies that are not explainable by any exit poll error hypothesis, or “hypothetical”,
offered to date. Neither a "constant mean" nor a “pervasive” pro-Kerry exit poll bias could possibly explain the E/M national aggregate exit poll data, or the detailed Ohio precinct-level exit poll data. To date no evidence-supported Exit Poll-based explanation of the Great Discrepancy has been provided.

http://electionarchive.org/ucvAnalysis/US/Presidential-Election-2004.pdf

[Melissa G]

1. More Conclusion...

Edit to put conclusion in top post and a bit more here..

This next bit should make for lively discussion around the election
forum...

The refusal by Edison/Mitofsky to permit independent analysis of their trove of data has deepened public concern. The shoddy and inadequate analysis (claiming, for example, that linear correlation
analysis, or a 56%-to-50% response bias, is sufficient to support the E/M hypothesis) that has been released to the public has deepened the uncertainty about what happened in the 2004 elections.
The Mitofsky/Liddle pervasive mean bias conjecture is unsupported by and inconsistent with the publicly available data.

Spin and obfuscation have spread the myth that the "exit polls are unreliable". The support of the media for the pollsters' exit poll response bias hypothesis as an explanation of the discrepancies
between the exit polls and the election results in the presidential election, without any serious evidence, has been a travesty.


http://electionarchive.org/ucvAnalysis/US/Presidential-Election-2004.pdf

[Febble]

6. Liddle here

Cite, please for "shoddy and inadequate analysis".

NO-WHERE has either Mitofsky or Liddle to my knowledge claimed that "linear correlation...is sufficient to support the E/M hypothesis) and Liddle was not responsible for 56%-50% response bias which WAS the hypothesis, not offered in support of it.

What Liddle and Mitofsky did was to demonstrate that there was no linear correlation between the magnitude of the discrepancy at precinct level and Bush's share of the counted vote. This neither refutes nor supports non-response bias as an explanation for the discrepancy. It does, however, infirm a specific claim made by NEDA/USCV that the discrepancy was greater where Bush's share of the vote was greater ("Bush Strongholds have more Vote-Count Corruption) see USCV March 31st paper.

The lack of a linear correlation between bias and Bush's vote share at precinct level suggest that this is not a significant or generalizable trend.

Also: my understanding from two of the authors of this paper is that they were not aware that it had been "published" and have not read the whole paper. Publishing what purports to be an academic, multi-author work without endorsement of the final version from at least two of the authors strikes me, frankly, as "shoddy".

[Time for change]

10. Thank you for the clarification Febble

Could you comment on what purports to be the smoking gun -- i.e., two precincts where even if all voters who did not respond to the exit poll had voted for Bush, the total Bush vote count would not have equalled the Bush vote count that was reported?

[Febble]

11. Yes, I can respond to that

The two precincts where, even if all voters who did not respond to the exit poll had voted for Bush, the exit poll would not have matched the vote count, are "impossible" only if you assume that only form of polling bias is literal "non-response" bias. If the sampling itself was not random - i.e. if Kerry voters were sampled at a higher rate than Bush voters - and there is credible evidence that this was so - then the discrepancy between poll and count is perfectly plausible.

Of course, so is fraud. But it is certainly not a smoking gun. Sampling bias is a real phenomenon. It is usually subsumed under the heading "non-response bias" because in one sense is due to a form of non-response, but my own view is that non-response bias (one group of voters declining to be interviewed at a different rate from another group) and sampling bias (one group of voters being sampled at a different rate from another) should probably be distinguished. Both are well-documented phenomena.

[Time for change]

14. Could you please clarify

The USCV document refers to "all voters who did not respond to the exit poll". I thought that would apply both to voters who were approached but who refused to participate in the exit poll, and to voters who were not approaced. Is that not true?

[Febble]

15. Going from the ESI report

the caption to the bar graph (wrongly labelled) says:

"The gray bar represents total possible response rate, from all exit poll non-respondents voting for Bush and all exit poll non-respondents voting for Kerry."

By "non-respondents" I assume they mean non-completers - either refusers or "missed" voters (selected but not approached). It is not an adequate report (as yet, I gather a proper paper is undergoing peer-review) but my interpretation makes sense.

http://www.votewatch.us/Members/stevenhertzberg/report.2005-07-19.2452304843/report_contents_file/

It's probably worth asking them, but it surely must mean the former.

Look, yes, it must:

"ESI looked at the range of non-response rate possibilities
in the sampled precincts to see whether the reported vote in those precincts fell inside that range."

http://www.votewatch.us/Members/stevenhertzberg/report.2005-07-19.7420722886/report_contents_file/

page 2

"non-response rate" is completers divided by completers, refusers and misses.

[Time for change]

16. So then, it seems to me that the reports of the two precincts

which the USCV report refers to as a "smoking gun" must have involved vote fraud (unless the miscount was accidental of course). Whether you talk of sampling bias or non-response bias doesn't matter, because the USCV report apparently refers to all voters who did not complete the exit poll, whether they refused or were never approached. In other words, even if every voter who refused to take the poll or who was not asked to take the poll were counted in the poll as Bush voters, the official count would still over-state Bush's vote. How can that be due to sampling bias?

Am I missing something?

[Febble]

18. Yup, as I understand it, you are.

I haven't read the USCV piece in great detail, I'm just talking about the ESI study. My understanding of what they did (and it is not at all clear, so it is possible that I am wrong - I should email them) is that the limits of the "possible" are that ALL the voters selected but not interviewed (i.e. either "missed" or refusers) voted for Bush or ALL for Kerry. ESI would have data on the completion rates for each precinct which include numbers selected, out of which a proportion would be interviewed, a proportion missed, and a proportion would refuse. If ALL refusers and missees were Bush supporters, that would set lower limit (negative WPEs mean Bush voters undersampled) on the plausible WPE and if ALL refusers and missees were Kerry voters, that would set an upper limit. WPEs outside these limits cannot be explained by literal "non-response" bias.

However, WPEs outside these limits CAN be explained, non-fraudulently, if the WPE arose from Kerry voters being selected at a higher rate.

This is hypothetical, but then so are the ESI scenarios: if an interviewer tended to select voters who looked friendly (and of course Kerry voters are all lovely people) then she might get a very high response rate - maybe 10% refusers. ESI's lower limit on her WPE would be whatever it worked out to be if all those refusers were Bushies. But her actual WPE would be much lower than that limit because the problem in that precinct is not that Bushies are refusing to participate when asked, but they are not being asked to participate.

Now that is an exaggerated scenario. But we do know, anecdotally, that people volunteered to participate. And Kiwi has some Ohio anecdotes, specifically that suggest that simply catching voters was a challenge, and we also know that when interviewing rate was low, bias was greater - Bushies may simply have been less easily caught.

This is not to argue that it is what happened (though I believe it probably did) but to argue that WPEs outside ESI's limits of the "possible" are also perfectly capable of an explanation that lies in polling methodology rather than fraud, and is actually supported by some evidence.

[Febble]

19. I just tried to email ESI

but failed. It might be necessary to register first, but I can't because it won't accept my British Zip code.

Here is what I wrote:

I have a query about your Ohio study. When you give the limits of the possible ranges of WPE that can be due to non-response bias, I have been assuming that this is computed by postulating, in turn, that all non-responders voted for Bush, and all non-responders voted for Kerry. And by "non-responders" I am assuming that you mean those selected for interview and either "missed" (interviewer busy) or who refused. However, it is presumably possible that another form of non-response bias could occur, beyond your limits, if the actual sampling process was biased. An interviewer who tended to select friendly looking faces might end up with a high completion rate, a narrow pair of limits on his/her "possible" WPE, but a WPE beyond those limits because, say, the friendlier faces tended to belong to Kerry voters.

However I may have misunderstood the nature of your limits. It seems to me that "non-response bias" can operate at the level of literal "non-response" but also at the level of voter selection. Bush voters may have had a greater rate of "non-responding" not because they refused or were "missed" (and recorded missed) but because they, for one reason or another, tended to escape selection.

In which case, unless I have misunderstood your definition, your two "impossible" precincts are explicable in terms of slightly broader definition of non-response bias. I would be grateful if someone could clarify this as I am frequently asked about it.

Elizabeth Liddle


If you'd like to try and send it yourself, I'd like to know the answer!

[Febble]

20. Got my email through

I'll let you know what I hear.

[Time for change]

21. Ok, I see what you're saying

The issue is with the non-selected voters. When the USCV report refers to all voters who didn't complete the exit poll, I was thinking that that meant non-selected voters as well. But when I specified missed and refusing voters in my last post I forgot about the non-selected voters.

Ok, so if the USCV report really is not referring to non-selected voters, then the two precincts in question would indeed fall within the limits of the possible (though I would think those results would still be highly unlikely).

But then, if the interview ratio was one, which means that there would be no non-selected voters, then the 2 precincts would not fall within the limits of the possible.

Thanks for looking into it.

[Febble]

23. You are right.

If the interviewing ratio was one, what you said would be true. It was a fairly rare interviewing interval, only applied in extremely small precincts. i.e. not precincts likely to be major fraud targets!

[Time for change]

25. Thank you for looking into this

In any event, these two precincts look like extreme outliers. It would be very interesting to look into these and see if these were cases where Bush was doing poorly compared to 2000, so that targeting them for substantial fraud would not be detected by the ESI study. Also, it would be very interesting to see what a recount showed in these precincts -- don't you think?

[Febble]

24. I just had an email

from Fritz Scheuren, confirming that my interpretation was correct. "Non-responders" are those selected who did not respond, either because they were refused, or were "missed". It does not include those who were not selected.

Scheuren, interestingly, classifies what I have been calling "sampling bias" as "measurement error" rather than "non-response bias". I am not going to argue with the president of the American Statistical Association about what terminology should be used for what sources of error in survey data, but he confirmed that what I called "sampling bias" would be an additional source of error, beyond the limits of "non-response bias".

[Melissa G]

12. Hi Lizzie, citing from conclusion as I posted..

Had yet to read entire paper.. Just getting the paper up for folks to read and shred or support.. the conclusion is almost completely present between my first two posts on this thread so I do not believe I cherry picked this quote. As it is almost the whole conclusion of the paper, I thought it worth posting. I am not yet endorsing any position..
Have to do kid things and first date anniversary with hubby things today... will read and get back to all this as soon as i can get a break from my other life..
Very Best Regards,
Melissa

[Febble]

13. Sorry, Melissa

Wasn't clear. My beef is with the paper, not your post. If they are going to claim that Mitofsky or myself claimed something, then they need to provide a source.

I am unaware of any source for their claim that either Mitofsky or myself considered that the lack of linear correlation between my proposed measure of the discrepancy and Bush's share of the vote supports either rBr or fraud. All I have done is to say that it does not support the USCV inference of "Bush Strongholds have more Vote-count corruption".

To be scrupulously fair, the longer I thought about it, the more I began to consider that if large numbers of votes were transferred from Kerry to Bush in randomly selected precincts, then that ought to be manifest as a linear correlation - although possibly not if the fraud was concentrated in Kerry strongholds.

But I have certainly never claimed the finding refuted fraud, and nor, to my knowledge, has Mitofsky.

[Melissa G]

22. Hi Lizzie, That was what I was trying to say in my previous posts to you

and you said you were not clear what i was trying to say..


where you say this..
To be scrupulously fair, the longer I thought about it, the more I began to consider that if large numbers of votes were transferred from Kerry to Bush in randomly selected precincts, then that ought to be manifest as a linear correlation - although possibly not if the fraud was concentrated in Kerry strongholds.

I was not sure how to say it mathematically where it would make sense to you...Just that if votes were switched it could be done with out being so correlation focused the way that you were looking at it..
Supportive Hug,:hug:
Melissa

[Febble]

26. Melissa, you are sweet

and I appreciate the hug.

I am searching for ways in which votes could be switched and neither produce a correlation between redshift and Bush's vote share nor between redshift and swing to Bush.

It is really not very easy! Concentrating fraud in Kerry precincts would help, also in precincts where Bush was doing badly relative to 2000. But I haven't squeezed that many votes out yet.

[jarnocan]

9. handy Contact Media (and more) link

https://ssl.capwiz.com/pdamerica/dbq/media/ they also have great contact-write your own froms for Congress etc. http://vvlobbydays.blogspot.com I will try to add it-I have petitions reguarding Russ Holt's Verified Voting HR550 so important!

[Melissa G]

2. This other thread just needs one more vote...

To make it greatest..
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x398055

[Melissa G]

3. Thanks GD POL... we made it to greatest and You ROCK!

[bleever]

4. This should also be up on Greatest. If any news deserves double posts,

this does.

Thanks, Melissa G

[Melissa G]

5. Thanks Bleever! This paper will take some digesting time so it is good to

keep it up in front of people so many folks will get a chance to read, comment and even perhaps educate those of us who are more mathematically challenged like myself...

:toast:

[Nothing Without Hope]

7. Excellent. Recommended with thanks. n/t

[jarnocan]

8. thanks I will blog /link it also

all the more reason, so many reasons it is mind boggling! I am so gald the Progressive Dems and Cindy Shehan have endorsed this event! I think it is very important and the DATE is MUCH too important to ignore! www.worldcantwait.org

[WillyT]

17. Recommended, Nominated, And Kicked !!!

:kick: