Exit poll analysis by 'Febble' on Dailykos

[kansasblue]

DailyKos has some analysis on the 2004 exit polls. ( Calling TIA )

[bleever]

1. Link, please?

[Blue Shark]

2. Link...

...http://www.dailykos.com/story/2005/4/29/161938/921

[On the Road]

3. This Goes to Mystery Pollster's Discussion

http://www.mysterypollster.com/main/2005/04/the_liddle_mode.html

which discusses a paper by Elizabeth Liddle:

http://www.geocities.com/lizzielid/WPEpaper.pdf
(temporarily down, possibly due to high activity)

I'm about halfway through, and it's absolutely fascinating. This Scottish woman has come up with some amazing analyses. Some of them contradict USCV, both other bolster their case.

This thing isn't over.

[tommcintyre]

4. The html version of her paper is available here.

http://66.102.7.104/search?q=cache:gZysBzp4iC0J:www.geocities.com/lizzielid/WPEpaper.pdf+WPEpaper.pdf+&hl=en

This is the html version of the file http://www.geocities.com/lizzielid/WPEpaper.pdf.
G o o g l e automatically generates html versions of documents as we crawl the web.
To link to or bookmark this page, use the following url: http://www.google.com/search?q=cache:gZysBzp4iC0J:www.geocities.com/lizzielid/WPEpaper.pdf+WPEpaper.pdf+&hl=en

Google is not affiliated with the authors of this page nor responsible for its content.
These terms only appear in links pointing to this page: wpepaper pdf
-----------------------

Title: Edison/Mitofsky Exit Polls 2004: sampling bias or vote-count
corruption?

[FreedomAngel82]

13. Glad to hear it's not

I'm glad people are still working on this. Kerry did win I believe. Exit polls aren't just something that's cute and nice like kitties.

[davidgmills]

5. Her analysis his been discussed twice already on

Threads posted by me called new USCV criticism and criticism revised.

Both are now on page two

[tommcintyre]

6. False premises lead to false conclusions

Hypotheses need to be tested, not proved or disproved (see more on this below). Edison/Mitofsky readily admit that the reluctant Bush responder is merely a hypothesis - and one that cannot be measured:

"While we cannot measure the completion rate by Democratic and Republican voters, hypothetical completion rates of 56% among Kerry voters and 50% among Bush voters overall would account for the entire Within Precinct Error that we observed in 2004." <The estimates are conjecture, or even wishful thinking.>
http://66.102.7.104/search?q=cache:gZysBzp4iC0J:www.geocities.com/lizzielid/WPEpaper.pdf+WPEpaper.pdf+&hl=en

"A hypothesis is a testable idea. Scientists do not set out to “prove” hypotheses, but to test them."
http://evolution.berkeley.edu/evosite/footshooting/Iterminology.shtml

Simply put: ALL hypotheses MUST be tested; and you can NOT test what you can not measure. Edison/Mitofsky readily admit they "cannot measure the completion rate <rBr>"; so the rBr hypothesis should be quickly disposed of since it can't be tested.

More detailed info here:
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x363706#363731

And further discussion here:
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x363383#363733

[Febble]

11. Yes, you can test what you can't measure

Most hypotheses are tested by measuring a proxy.

Do you think a pregnancy test measures the baby you might have?

The way science works is you make a hypothesis. You then make a prediction based on your hypothesis. If your prediction is confirmed, then your hypothesi is supported. If it is not, it isn't.

But support isn't proof. So you make and test alternative hypotheses.

E-M made the hypothesis that Bush voters were more reluctant to respond. A prediction flowing from this is that where there was more opportunity to avoid being polled, the error should be greater. This was so, and thus their hypothesis was supported.

Unfortunately the measure they used was a poor one. I have provided a better one. I hope they will re-do their analyses.

And a problem for all of us, including USCV, is that we do not even have their measures. We have a few averages. It is my contention that the claim made by USCV on the basis of these averages is not supported. That does not mean it is wrong.

[TruthIsAll]

12. Febble. There is a contradiction. Please explain.

"E-M made the hypothesis that Bush voters were more reluctant to respond. A prediction flowing from this is that where there was more opportunity to avoid being polled, the error should be greater. This was so, and thus their hypothesis was supported".

This hypothesis was contradicted by E-M in their FINAL NATIONAL EXIT POLL.

Febble, could you tell us whether or not you agree with any or all of the following?

Fact.
Gore won the popular vote with 50.999 million to Bush's 50.456.

Fact.
The FINAL 13660 National Exit Poll states that Bush 2000 voters comprised 52.57 million (43%) of the total 122.26 million who voted in 2004. That is 2.11 million more than the 50.456 million who actually voted for him.

Fact.
Gore voters comprised just 37% (45.23 million) of the 2004 vote.

Question.
If Bush voters were reluctant to respond, then how come 43% of Bush 2000 voters participated while only 37% of Gore voters did?

Fact.
The National Exit Poll at 13047 respondents had the split at 41%/39%.

Fact.
The Bush 2000 weight could not have been 41% at 13047 respondents, much less 43% at 13660.

Fact.
The maximum possible Bush 2000 voter turnout/2004 vote was 39.82%, after reducing his turnout by the approximate 1.77 million Bush voters who have died since 2000.

Fact.
Even 39.82% is too high, since at least some Bush 2000 voters stayed home this time.

Fact.
The 52.57 million Bush turnout is overstated by 3.88 million, when you add the 1.77 million resurrected Bush 2000 voters to the 2.11 million phantoms.

Quip.
These 3.88 million Bush voters must have been very motivated, indeed.

Fact.
According to the NEP, only 45.23 million of Gore's 2004 voters participated in 2004, since .37*122.26 = 45.23 million.

Fact.
If the 37% Gore weight is accurate, then you must believe that 5.77 million Gore voters (11.31%) were reluctant to vote in 2004 and stayed home.

Fact.
To accept as accurate the 37% Gore weight, you must also accept that the 43% Bush weight is also accurate.

Fact.
But the 43% Bush weight is impossible - at least 3.2% too high. Therefore, the 37% Gore weight must be impossible - and at least 3.2% too low.

Question.
Was there a DGR (Delinquent Gore Responder)?

Fact.
Kerry was the 51-48% winner at 13047 respondents.

Question.
What happened to change the 41/39% to 43%/37% over the final 613 respondents? Bush won the FINAL by 51-48%.

Fact.
The IMPOSSIBLE 43% Exuberant Bush Responder invalidates the FINAL EXIT POLL, which matched to the recorded the vote for a Bush 51-48% win.

Fact.
The IMPOSSIBLE 43% weight indicates an Exuberant Bush 2000 Responder (at least in the Exit Poll) which is at variance with the Reluctant Bush Responder hypothesis.

Fact.
The IMPOSSIBLE 43% weight is a necessary condition for Bush to win.

Fact.
The 41% Bush 2000 voter weight, though EXTREMELY IMPLAUSIBLE, still showed that Kerry won the first 13047 repondents by 51-48%.

Fact.
If the weights were a more realistic 40% Gore/39% Bush, then Kerry wins in a landslide - by at least 7 million votes.

Fact.
The final 613 had to have voted 100% for Bush to overcome Kerry's lead among the first 13047.

Quip.
Bush voters must have been EXTREMELY EXUBERANT to do that.

[FreedomAngel82]

14. What I've seen

with Bush supporters is they were VERY proud to be voting for him just like we were with Kerry. Some people don't like to tell but I think you'll find more people don't mind telling and like to tell on a presidential election.

[kster]

17. Then they fight you,

stage.Keep up the good work.

[Febble]

23. None of these are facts.

they are responses. And may be wrong

Just as the vote counts are not the votes. They are a record of the votes. And may be wrong

The mismatch between the two requires explanation.

But the fact that the two don't match doesn't tell you which one is wrong.

As you have been at some pains to point out to me.

[TruthIsAll]

26. You ASSUME these are responses. Bad assumption.

THIS WAS NOT A RESPONSE.
THIS WAS A PLUG.
A VERY OBVIOUS PLUG.

41% in 13047 (Kerry wins, 51-48)
and
43% in 13660? (Bush wins 51-48)

It was really something close to 40% Kerry/39% Bush.

NO RBR.
NO RGR.

JUST EPR.
EXIT POLL RESPONDENTS.

43% was NOT a response.
It was an artifice to force the exit poll to MATCH a corrupt recorded vote: Bush 51-48.

And I have showed 43% to be IMPOSSIBLE (39% maximum for Bush).

Now apply some science to THAT discrepancy.

And I don't mean faith-based science either.

THERE IS ONLY ONE CONCLUSION.

BUT YOU CANNOT ACCEPT IT.

BECAUSE IT TOTALLY INVALIDATES RBR

[tommcintyre]

15. Your reasoning is flawed

Measuring a proxy is fine. But you need historical validity to trust measuring by proxy; and the rBr hypothesis has none. Your inferred data is only predictive; and must be confirmed by future DIRECT data. (i.e. developing a method to determine what the ACTUAL reason of rr's may be. At this point you can only speculate.)

Your example:

"Do you think a pregnancy test measures the baby you might have?"

Of course not. The test only measures a predictive element (hormone levels in the urine), just as the rBr hypothesis does now for speculating what the reason for not responding of the rr's was. The difference is that the pregnancy test has much historical confirmation that it IS an accurate predictor of pregnancy (a very high percentage of positive testers actually proving to be pregnant), as where the rBr hypothesis has no historical record to support the claim.

You say (above):

"You then make a prediction based on your hypothesis. If your prediction is confirmed, then your hypothesi is supported. If it is not, it isn't."

You have NOT, and can not, confirm your prediction:

http://elections.ssrc.org/research/ExitPollReport031005.pdf
"A Review of Recent Controversies Concerning the 2004 Presidential Election Exit Polls"

(page 8) Furthermore, it is complicated in a way that many post-survey evaluations are by the fact that some information is essentially unknowable. This is especially true when one of the concerns is nonresponse, and there is no information from the nonrespondents to analyze. As a result, there are some sections of the report in which there is an extremely detailed level of disclosure about what the exit poll data show, but in other parts of the report there are only hypotheses about what might have been the cause for a particular observation. These hypotheses can guide future experiments in exit polling methodology or even direct changes in the methods, but they cannot explain in a strict causal sense what happened in the 2004 data collection.

[Febble]

18. Yes you can confirm your prediction

That's why do the stats. But ideally you'd want to replicate, I agree. And we do not know the R squared of the model, or the p value, so it could be a chance finding. They also need to re-run it using my variable.

But if poor polling protocol really was a significant predictor of Kerry over-estimate, then the direction of causality it fairly clear here.

It is unlikely that poor polling protocoal "caused" the votes to be counted wrong. So if there is an association between poor polling protocol and a mismatch between the poll and the count, it seems far more likely that the cause is in the poll not the count.

This is where the proxy comes in - why should poor polling protocol be associated with high signed WPE, not just high unsigned WPE?

In other words why should poor polling protocol be associated with differential non-response? One reasonable hypothesis is that while both groups of voters were "reluctant" (response rates for both groups were low), Bush voters tended, overall, to be more reluctant than the other. Where poor protocol allowed this differential to be expressed, the signed error was more negative.

It's only a theory. But it fits the data. And it fits the data better, so far than fraud. If you can explain why inexperienced polling interviewers outside a precinct could "cause" fraud to occur inside that precinct, fair enough. Maybe inexperienced interviewers were deliberately assigned to fraudulent precincts as a cover for the fraud.... ;) But sometimes association does imply a direction of causality pretty strongly, even without history. (And sometime history doesn't.)

[Mairead]

19. You're doing "creation science".

When you put forward your hypothesis, implicit in it is the assumption that the problem is known to be in the exit polls, not in the official tallies. Your 'rRv' hypothesis assumes that the tallies are good, an error equivalent to hypothesising that the reason so many angels can dance on the head of a pin is because they make themselves very tiny rather than interpenetrate.

You can't get away with assuming angels, and you can't get away with assuming the validity of the official results. That error is bad enough to get you an 'F' on an Intro to Methodology term paper.

[Febble]

20. No I'm not

There is no implicit assumption that the exit polls were wrong, but it is statistically as clear as statistics can be that the polls didn't match the count.

The numbers themselves don't tell you which was wrong.

What we need to do is to find out whether the greatest mismatches were in places that suggest fraud (Ohio? touch screen precincts?) or whether they were in places that suggest interviewer error (precincts where the interviewers were a long way from the precinct? crowded precincts? Precincts with inexperienced interviewers?).

The trouble with the fraud hypothesis is that there are so many ways it could have happened, and to track it statistically you have to have specific hypotheses.

The polling incompetence hypothesis is easier to test because we do know (or E-M does) which precincts had the poor conditions. And the results suggest that it was a significant factor - error was greater where factors that would have compromised good sampling protocol were present.

But it doesn't mean that fraud wasn't a factor as well. It's just that we don't know what its fingerprints look like. But we can take a guess, and test for what we think they look like. And it might be easier now that I've figured out how to get rid of some of the junk.

(copy of my reply to your comment on DKos0

Extra: calling my approach Creation Science is a low blow. Please read my paper. It makes absolutely no assumptions regarding the source of the error. The math is completely fraud-neutral as I explain.

[davidgmills]

21. Back to the counter-intuitive argument

"whether they were in places that suggest interviewer error (precincts where the interviewers were a long way from the precinct? crowded precincts? Precincts with inexperienced interviewers?)."

This would be the clasic description of a Democratic precinct; not a Republican one.

Perhaps this is illustrative of the difficulty someone who is unfamiliar with the United States has in grasping the problem.

[Febble]

22. Well blame it on that if it helps.

But you've actually missed my point.

This was evidence from the E-M analysis. A flawed analysis, but an analysis. They do not tell us whether the inexperienced interviewers were more prevalent in Democratic precincts or not, and you make a good point if you think that the variables may have been confounded. We don't know, because it wasn't a good report.

But it has absolutely nothing to do with what I know about the US. It is simply about what I know about the numbers. Kerry's vote was more greatly over-estimated in precincts where the interviewing protocol was poor. Why? Because by sheer coincidence, that was where fraud was targetted also? Because Karl Rove planted inexperienced interviewers in selected precincts to divert attention from his evil plan?

Or because polling protocol has something to do with the extent of the problem? Commonsense tells me the latter. So what was it? Could have been a number of things. If you think hard enough you may be able to think of some more convincing ones, being American. But whether you are an American or a Martian you need to explain why Kerry's vote was more greatly over-estimated in precincts in which good random sampling was more difficult.

[Mairead]

25. "Kerry's vote was more greatly over-estimated"

How do you know that? How do you know what the vote really was?

If you don't know what the vote really was, how can you possibly assert anything whatsoever about it?

[Febble]

27. It may have been over-estimated because

the vote count was fraudulent.

Do READ what I write.

If I have a shop and I anticipate that I will have $1000 dollars at the end of the day, and don't allow for the fact that I have a light-fingered assistant at the till who has been helping himself all day, I will have "over-estimated" my take.

It doesn't mean I did my sums wrong. It means I had a fraudulent assistant.

Sheesh.

[TruthIsAll]

28. Febble, here's what MP said of Reluctant Responders - BEFORE the election

What do you make of it?

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x365550

"When a voter refuses to participate, the interviewer records their gender, race and approximate age. This data allow the exit pollsters to do statistical corrections for any bias in gender, race and age that might result from refusals".

[Mairead]

29. I don't think I'm the one not reading what you write :-)

We not only don't know what caused the 'over-estimation' (whether it was your light-fingered shop assistant or simply the dreich weather), we don't know THAT THERE WAS any over-estimation. But the language you use PRESUMES over-estimation. If you know the shop assistant nicked the difference, you wouldn't say you'd 'over-estimated' the receipts unless you intended irony. You'd use language that doesn't imply your estimate was an error.

Do you not see that?

In fact, the estimate of Kerry's share of the vote might have been completely correct, neither over- nor under-estimation. But nobody would ever know that from the way you frame the situation.

'Over-estimation' is a loaded term that implies that the results were correct and the estimator was at fault. It assumes a particular conclusion, which is bad science.

Furrfu. :)

[Febble]

30. OK, OK

I don't know what words to use that will convince readers that I am NOT assuming the fault was in the polls. But I'm NOT.

But there WAS a mismatch between the polls and the count. E-M have told us there was. We don't even need those screen shots. The results E-M got from the precincts counts didn't match their estimates from the precinct polls. It's called the "within precinct error". Or WPE.

OK?

Now, you understand that, if you see anyone else who's misunderstood it, tell them too.

[Mairead]

31. No, there's no 'ok' about it.

Your 'ok, ok' is merely dismissive, not convincing.

It's called the "within precinct error". Or WPE.

No, it's only called that by people who assume that the problem is in the exit polling! Why in heaven's name would anyone call it an 'error' if they knew that the exit-poll numbers were correct but the votes were stolen? What would the nature of that alleged 'error' be in such a case?

You continue to deny that you're assuming the exit polls are at fault, but everything else you say reveals that that is exactly what you're doing.

[TruthIsAll]

32. Mairead, I pointed this out to Febble yesterday.

She kept referring to the exit polls as being "wrong".

That is what I call insidious use of the language to frame the debate as the naysayers have tried all along- purely Orwellian.

They never ONCE entertain the presumption that the votes were corrupted and THAT is why we are analyzing the exit polls.

This asinine twist of logic is a telltale sign of intellectual dishonesty. It smacks of a very unscientific agenda.

Thanks for focusing on this.

[Febble]

33. Edited post

This makes me very cross (previous version was crosser).

You can think what you like of me, but you are simply wrong. I would love to see evidence that it was the count, not the polls that were "wrong". (I would even more like to see some legal action over the shenanigans in Ohio for which there is incontrovertible evidence).

And I have produced a mathematical formula that could help the process of finding out what happened, by clearing irrelevant noise from a variable.

We can now use that variable to investigate the factors that were associated with variance in whatever you like to call it - mismatch, bias (in count or poll) what-you-will.

Frankly my computer doesn't give a damn what I name I give a variable. What I want to know is: is it greater in places that would indicate fraud (swing states, precincts with touch screen voting machines) or is it greater in places that would indicated polling foulup (precincts with inexperienced interviewers, precincts where interviewers had to stand a long way from the precinct). So far the evidence suggests the latter. But there is a potential confound that so far has prevented us testing the former.

I have fixed that confound. It will enable us to test the fraud hypothesis.

If you don't like the name I've given the fix, fair enough. It won't affect the math one jot.

Now I am simply not going to post any more, as debates that get bogged down at the start over issues of perfectly innocent terminology are not worth having.

You can both fight over my carcass. I'm outta here.

[TruthIsAll]

37. Why should we believe you if USCV says you misstate the data in your model?

I guess you found your stay at DU more than you bargained for.

Your fog will dissipate, like all the rest.

[Febble]

38. I visited because I thought people were interested.

Some were, some weren't.

[Mairead]

41. Yes, that's exactly what it is: framing

Here we have someone writing a doc dissert in cognition, specifically non-standard symbolic perception. Yet she uses loaded language while disclaiming intentionality, and goes on doing it even after being called on it. That's not an innocent error, that's a game.

[tommcintyre]

39. Excellent point! "WPE" assumes the error is in the polling NOT the count

I have found the "pro-rBr" argument (by Febble and others) to be riddled with such subjectivity.

It is important that we discredit such "faith-based statistical analysis" gambits, since Bushco is using these "faux facts" in an attempt to "create doubt" (discredit) the election fraud argument.

[TruthIsAll]

34. COULD IT JUST HAVE SIMPLY BEEN THE CASE THAT MORE PEOPLE VOTED FOR KERRY?

Can you get your mind to focus on that very distinct probability?
Because based on everything we know, 99.7% of the voting machine anomalies favored Bush. That was the reality.

Why should we look for contrivances to explain the exit poll deviations WHEN THEY ARE TOTALLY CONGRUENT WITH ACTUAL PHYSICAL EVIDENCE?

We know that 86 of 88 documented touchscreen incidents turned Kerry votes to Bush votes.

Calculate the probability of THAT.

[EST]

46. You have crashed against bastions composed of

metaphysical horse shit. To feed my metaphor blender-no matter how carefully you choose your pictures and words to illustrate just where the rocks are, in the hands of some, occam's razor resembles a very dull brick.

Your point is well taken.

[Mairead]

24. You are, you know.

The very fact that you're focussing on the exit polls should tell you that you've made assumptions, perhaps unexamined ones, about the locus and nature of the problem.

Why aren't you complaining that the pollers didn't sample frequency of drinking cider, or the percentage of women wearing tights? Obviously, you don't care about cider or tights because the probability that those factors played a significant role in what happened is on the order of 0.0000000001.

Unless you can show that (e.g.) incompetence is a significant explanatory factor for the outcome, then the fact that it's easy to test for incompetence is as irrelevant as the fact that the light is better under the streetlamp when you've lost your doorkeys in the dark.

If the election was nicked, then whether the folk doing the polling were incompetent or superb or even invented their data over a jar too many simply doesn't matter.

(And I'm sorry that you felt the 'creation science' comparison was a low blow, but making unacknowledged assumptions is exactly what they do and why they are religion, not science.)

[FogerRox]

35. I have an issue here>>>>

Febble said:
"What we need to do is to find out whether the greatest mismatches were in places that suggest fraud (Ohio? touch screen precincts?) or whether they were in places that suggest interviewer error (precincts where the interviewers were a long way from the precinct? crowded precincts? Precincts with inexperienced interviewers?)."

IIRC Ohio had
-4 county's using DRE's
-44 county's usoing punchcards
-40 county's using opscans
Correct me if I'm wrong--

But my 1st inference from Feebles comment quoted above was that --

This person is not cognizant of the basic facts. The idea that the 4 DRE county's played this role is a little----well hard to fathom.

[Rex_Goodheart]

42. Sorry, Febble, but your argument is specious...

and circular... You're assuming what you wish to prove.

"The polling incompetence hypothesis is easier to test because we do know (or E-M does) which precincts had the poor conditions. And the results suggest that it was a significant factor - error was greater where factors that would have compromised good sampling protocol were present."

What you and E/M have presented as "explanations" for a pro-Kerry variance are mere post hoc conjectures. You have no basis to contend that distance from a precinct makes a Bush voter more reluctant to respond. You have no basis to contend that crowded conditions make a Bush voter more reluctant to respond. And you have no basis to contend that a Bush voter would be more reluctant to respond to a new interviewer.

I could contend quite the opposite on each of those... for example, a new interviewer is not so jaded as the older one and not as disinterested, so takes better care with his sample.

Just imagine, for a second, a wholly possible 6% bias in Bush's favor in vote tabulations. You'd suddenly have no need to create post hoc explanations for the exit poll at all.

Ever hear of Occam's Razor?

[mgr]

43. Actually it's an A

What you seem to fail to appreciate is that a null hypothesis must be, because of the logic of "if then" construct, in oppostion to the position you seek to affirm. You may only refute the null when you have the consequent present but the antecedant is not.

I would wait for the formal peer review in mid May to see if Febble's term has merit. Like I have stated before, science and logic are not subject to democratic principles, nor are they popularity contests, nor should they affirm an already fixed political agenda as would be the case with Lysenkoism.

Mike

[FogerRox]

36. Another Febble foible

Febble said__ it seems far more likely that the cause is in the poll not the count.

This seems to be a bias against possible fraud, yet this statement is stated without articulating why it must be so. That is the argument of Authority. SO I have to ask: where is this Authority derived from?

[Febble]

40. My statement was not Argument from Authority

It was this (copied from thread above)

"E-M made the hypothesis that Bush voters were more reluctant to respond. A prediction flowing from this is that where there was more opportunity to avoid being polled, the error should be greater. This was so, and thus their hypothesis was supported.

Unfortunately the measure they used was a poor one. I have provided a better one. I hope they will re-do their analyses."

I will elaborate: not only was their measure poor, they do not give statistical details in the report and so I remain skeptical until they re-do the analyses, using my variable, and give statistical details.

However, the evidence, though flawed, that poor sampling protocol was associated with greater discrepancy between poll and count suggests that at least part of the problem was in the poll.

This is not argument from authority. It is logic. But I remain skeptical until we have more evidence that the association is a real one.

It seems you guys would love to paint me as a crazy foreign Rove plant.

You might like to check out the USCV website and see the work I have done there to investigate fraud in 2004.

http://uscountvotes.org/index.php?option=com_content&task=category§ionid=4&id=87&Itemid=63

And possibly apologise.

[mgr]

44. I remember each one posted here without attribution n/t

[tommcintyre]

45. Paint you as "as a crazy foreign Rove plant?" False premise AGAIN

Or... should I say assumption?

Do I think you're Crazy? - No (just biased).

Is being foreign a problem? No (your nationality is immaterial).

A Rove plant? Inconclusive (I don't have enough information to form an opinion on this - one way or another. I'm just trying to avoid being biased on this matter. ;) )

What I DO know is this: Regardless of your intentions, your paper plays right into the hands of Rove and company. When you make statements like the following from page 21 of your paper:

"The pattern instead is consistent with the E-M hypothesis of “reluctant Bush responders”..."
http://www.geocities.com/lizzielid/WPEpaper.pdf

You are helping them to create illegitimate doubt in the high probability that the election was stolen. Your conclusions "fly in the face" of the evidence and historical facts. MOUNTAINS of evidence point to fraud; and, historically, exit poll discrepancies have proved to be a good indicator of fraud.

It is reports like yours (supporting the rBr hypothesis) that helps to give an excuse to the Carter/Baker Election Reform Commission to NOT have USCV as a panelist. (If you doubt this, listen to the following interview of Pastor, and see what he says about the USCV report at about 14 minutes into the 30 minute show.)
http://c-span.org/search/basic.asp?ResultStart=1&ResultCount=10&BasicQueryText=Commission+on+Federal+Election&image1.x=20&image1.y=2&image1=Submit

So, whether it is your intention or not (and your previous work notwithstanding), you are "aiding and abetting" Rove and his cohorts in their attempt to cover-up the stolen election. In effect, that makes you a collaborator. I know that's not a very nice label - but, as the saying goes, "if the shoe fits, wear it".

[johnaries]

47. The reluctant responder theory has already been disproven.

it was shown that response was actually HIGHER in the counties that polls showed going to Bush.

Forgot which thread it was, but it's in the archives somewhere.

[kansasblue]

7. Can someone post back on Dailykos?

Dailykos very popular. Can someone rebut her analysis over on Dailykos?

[MontageOfFreedom]

8. Truth is All will do it. Its obvious from the word go...

73,000 state exit poll respondents, nationwide, made a mistake in punching their vote for Kerry? Or is it more accurate that thousands of those were transferred to Bush. End of debate.

[sunshinekathy]

9. US Count Votes' Rebuttal & Plea for Support of Election Data Archive

Bruce O'Dell, VP of US Count Votes, has created a simulation model of Lizzie's algorithm. It indicates that her model requires a "participation by partisanship" profile that is -totally- inconsistent with E/M's empirical data.

Lizzie's model requires -greater- exit poll participation in High-Kerry precincts than in High-Bush precincts. E/M's data shows quite the contrary.

If you reproduce Lizzie's results, and delve down into the data, you face a fundamental contradiction with E/M's "participation rate by precinct" empirical data.

Lizzie (and E/M's) "uniform response bias" hypothesis simply does not fit the facts. Period. There is simply no way to reconcile a "uniform response bias" hypothesis with E/M's "participation rates by partisanship" data.

US Count Votes will eventually publish the raw data and my simulator source code, to allow anyone to confirm my assertions.

For Bruce O'Dell's professional qualifications to simulate mathematical models, please refer to www.digitalagility.com/Odell_home_page.htm
35 years after writing his first Basic program, Bruce has reached the peak of his profession.

He is currently engaged as the lead architect at an enterprise security project at one of the Fortune 20 ("20" is not a typo). He is in charge of the technical aspects of security at one of the twenty top public corporations in America, and for the last 15 years, I have focused a major portion of my consulting practice on formal modeling of complex systems.

US Count Votes needs funding NOW to hire full-time programmers, system administrators, liason with the 33,000 election offices in America, statiticians, etc. if it is to complete its project to audit elections to have a system in place by Nov 06 to ensure that the right people are sworn into office in Jan 07. This cannot wait. The technical systems US Count Votes is building for its National Election Data Archive project are huge and cannot be built in six months prior to the Nov 06 election.

Every other issue that the Democrats care about will be resolved as soon as we audit elections to ensure that the right candidates are sworn into office in Jan 07. US Count Votes' needs 1/2 million dollars minimum for the staff to accomplish its database systems (which require about 1000 database tables - 15 per state and some shared tables). Please help out by donating or helping us to raise funds:

http://electionarchive.org/fairelection/donate.html

This project is the Democrats' best hope for restoring Democratic control of Congress in Jan. 07.

Best,

Kathy Dopp
http://electionarchive.org

[Febble]

10. Response rates

Regarding response rates:

The E-M report states that there was no significant difference,although they do not give details of the statistical test. However, it is true that the "differential response" hypothesis predicts that response rates should actually be lower in Bush strongholds than in Kerry strongholds - because allegedly, Bush voters don't like responding.

This is an important argument. However, statistically, a small difference is only detectable if you have enough statistical power. The predicted difference is small (in theory you should get 56% response from an all-Kerry precinct, and 50% in an all Bush district, so the maximum difference is 6 percentage points). Statistical power depends on two things - plenty of data and good signal-to-to noise ratio.

Here the numbers are small where it matters - in the extreme precincts where the differences should be maximised. There were only 90 "high Kerry" precincts and only 40 "high Bush precincts". So the data would have to be pretty clean. We know it was not, as the E-M report gives as a value called the "absolute" error, which, without going into too much detail, tells us how much noise there was in the data.

When I did my model, I compared the amount of noise in my model data with the amount of noise in the real data. I had less, and even so, statistical significance of the lower responses in my "high Bush" precincts was only borderline. I estimated that if I increased the noise so that it matched the real data, the difference would be lost in the noise. It only takes a couple of stray data points to lift the response rates at the "high Bush" end, or lower them at the "high Kerry end" to destroy the effect.

So I agree that the lack of evidence (and it is no more than that)of lower response rates in "high Bush" precincts than in "high Kerry" precincts is an argument worth considering. However, the way statistics works that what you test, statistically is a "null hypothesis". If you cannot show that an effect is unlikely to have occurred by chance you "retain the null". If you "retain the null", have not proved that the effect was not there. You simply have not proved that it was. It is the statistical equivalent of "presumption of innocence".

Unless E-M provide us with better data, I do not think we can reject the null hypothesis that over-estimates of Kerry's vote were randomly distributed amongst precincts of all degrees of partisanship.

This in itself, however, does not rule out fraud as a cause of the over-estimate. The same effect would result from under-counting of Kerry votes.

Bruce has done a great job with his simulation, and he is clearly at the top of his field. However, the point here is not the skill of the modeller, but the parameters of the model. With the greatest respect to Bruce (and we have been exchanging information very co-operatively over the modelling, for which I am extremely grateful to him), I simply disagree that his model disposes of widespread rBr, or for that matter, of widespread fraud. (Other forms of analysis may do the latter, but not this one).

This may be a matter of different philosophical approaches to data. Bruce is an engineer, my background is in the social sciences (psychology) although I also have done a bit of engineering. When data points are people, noise behaves badly. I think what we are seeing is noise, and I would argue that there is a sound statistical basis for my claim.

Lizzie

[MontageOfFreedom]

16. I can't support this.

It sounds like the rubber-stamp argument offered by "Mystery Pollster", and because all of the raw-data for the questions in each precinct is being with-held there is no way I can see supporting this.

TIA has outlined the truth numerous times, there should be no issue in sampling of the data. If the issue is there, it should be reflective of all the states.

I just don't see any reason to rehash this.

http://www.exitpollz.org/cnn2004epolls/Pres_epolls/US/P/00/epolls.0.html

State by state the results are consistant.