The Gun is Smoking: Ohio Exit Poll - Evidence of Vote Miscount

[sunshinekathy]

Release: November 2, 2005

The Gun is Smoking: Ohio Exit Poll Data Provides Virtually Irrefutable Evidence of Vote Miscount

Organization: The National Election Data Archive (NEDA)

Summary: New analysis of the precinct-level Ohio exit poll data provides virtually irrefutable evidence of large scale vote miscounts in Ohio during the 2004 presidential election. 6% of Ohio's exit-polled precincts had impossible vote counts and 57% had significant discrepancies (with less than 5% chance of occurring in any one precinct). The pattern of Ohio's exit poll results is not consistent with any exit poll error hypothesis. However, it is consistent with pro-Bush vote miscounts.

NEDA's full paper "The Gun is Smoking: Ohio Precinct-level Exit Poll Data Show Virtually Irrefutable Evidence of Vote Miscount" is available at
http://electionarchive.org/ucvAnalysis/OH/Ohio-Exit-Polls-2004.pdf

In two Ohio precincts, even if all voters who did not complete exit polls had voted for Bush, the total Bush vote count would have been less than the official count. In a third precinct, all voters who did not complete exit polls would have had to vote for Bush to equal the official count. Unless Bush voters lied much more than Kerry voters on exit polls, or massive exit poll error occurred that was not detected by the pollsters, the results are mathematically impossible.

The Ohio exit poll data are a smoking gun for vote miscounts in Ohio. Ohio exit poll results are consistent with earlier findings of similar unexplained and implausible exit poll discrepancies in the national exit poll sample as described in the January 21, 2005 Edison/Mitofsky report.

For these reasons, NEDA urges

1) public release of all detailed exit poll data and methodologies by Edison/Mitofsky so that independent analysts may determine whether possible vote fraud occurred or not;

2) routine public posting on the Internet of accessible detailed election results (at the precinct level broken out by absentee, early, provisional, and Election Day counts) by local election officials as soon as polls close, to enable independent analysts to identify precincts with possible vote count errors in time to contest elections;

3) routine independent audits of vote count accuracy in all elections to detect and correct errors that might be deliberately or innocently introduced by insiders; and

4) widespread media coverage of this vitally important issue.

The data that NEDA analyzed was provided in the Election Sciences Institute (ESI) June 6, 2005 report, “Ohio Exit Polls: Explaining the Discrepancy”


NEDA's full paper "The Gun is Smoking: Ohio Precinct-level Exit Poll Data Show Virtually Irrefutable Evidence of Vote Miscount" is available at
http://electionarchive.org/ucvAnalysis/OH/Ohio-Exit-Polls-2004.pdf

This paper follows a recent mathematical proof by NEDA that Mitofsky and ESI's earlier analysis of the national and Ohio exit poll data is logically invalid
http://uscountvotes.org/ucvAnalysis/US/exit-polls/ESI/ESI-hypothesis-illogical.pdf

The National Election Data Archive (NEDA) is a nonprofit organization of statisticians and mathematicians devoted to the accuracy of U.S. vote counts. Please see http://ElectionArchive.org for more information.

Contacts: Kathy Dopp [email protected] 435-658-4657 or cell 917-656-0066 and Ron Baiman [email protected]

NEDA is seeking funds to implement a new national election data archive in order to collect and publicly distribute detailed election data to pinpoint possible vote miscounts immediately following elections. http://electionarchive.org/donate.html

For common-sense ways to ensure accurate vote counts in America read this short 2 page recommendation: http://uscountvotes.org/ucvAnalysis/US/election_officials/Audits_Monitoring.pdf

To receive NEDA's announcements email [email protected]

[whatever4]

1. Good!

Now maybe that will convince some of my conservative, libertarian friends that the election was, REALLY was stolen. Not just the whining of a loser. Not just a "theory". Not just a few missed votes. Not just "on accident" due to new machines. Nothing accidental at all. Elections ARE stolen.

Maybe then they can get their arms around the fact that 2000 was stolen too. And the Iraq war is bad, and that there IS a conspiracy, that we are not and were NEVER planning on winning anything in Iraq but the profits the war would bring.

When I realized, that morning, that the election had been corrupted, and later realized that though it was plain to see to me (because it wasn't just one COUNT that was wildly off, more than one measure didn't meet expectations and it all added up to something which is numerically far less likely and historically unprecedented) that is when I realized how fractured we are. What's the point in arguing with anybody else when those I used to agree with disagree?

Only a handful of people took it seriously then. I hope now they do. I wish they'd been as suspicious then as they know they need to be now.

As if anything else mattered more than our right to vote. As if we shouldn't have guarded it with our lives. Of course, then they didn't know we had major hurricanes in our future. Though we were at war; it should have mattered. We didn't realize how many of our lives depended on honest government and the good management of our resources. How at risk the mismanagement left so many people, and how many of us have now died because of it. I guess now we know.

Had we elected a decent president, we likely would not be at war now. I'll say that straight up, I think we'd be out by now, and if not, we'd at least have a plan. I know we'd have something constructive. Right now, we have nothing. Nothing but lies, that is.

We likely would have actually handled the thousands of disaster victims decently, humanely, and QUICKLY. Active mismanagement went into the building of the disasters, ice and food turned away, and on, and on. And imho, chances are, money and work would have been directed to the levees BEFORE the hurricanes hit. I keep on saying it, and I'm going to KEEP saying it; it was predicted to be a bad hurricane season. Not a surprise. Work on the levees WOULD have happened under a different admin, because there was NO reason for this admin to have cut the funding. No sane reason, that is. No real reason at all. Patently insane. LIHOP.

I wonder what else would be different now. All I really know is, a LOT more people would be alive today, and a LOT of suffering would have been avoided. When this admin won office, it all became unavoidable.

[Native]

2. kick

[OnTheOtherHand]

3. factually incorrect

"In two Ohio precincts, even if all voters who did not complete exit polls had voted for Bush, the total Bush vote count would have been less than the official count. In a third precinct, all voters who did not complete exit polls would have had to vote for Bush to equal the official count."

No, rather, if everyone who was "missed" or "refused" to complete an exit poll had voted for Bush, the proportion of Bush votes in the survey would be less than the official proportion.

This is an absolutely crucial distinction. Most people in most exit poll precincts don't even have the opportunity to fill out an exit poll.

[OnTheOtherHand]

5. let's try, well, reading DU on impossible Ohio exit poll precincts

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=397259&mesg_id=397543

Liam_laddie has reported that he interviewed a couple of Democratic Party Poll Observers from Cincinnati 4M. (Of the Ohio NEP polls we have been able to identify, Cincy4m has the biggest WPE. But it does not have the biggest WPE of all the Ohio NEP polls.)

According to the observers, the NEP interviewer at Cincy4m was totally out of his depth. He managed to conduct 31 interviews out of the 1754 voters (times 2, because they went by him twice) fanning in/out from the polling place (the polling place held 4 precincts). It was raining and he was stationed 100 feet away from the door, in a group of campaigners who were vying with each other for the attention of voters entering the polling place.

The observers said that "Republican types" just streamed by him. They didn't refuse to be interviewed. They didn't even notice him.

Maybe the guy was able to note the age, gender and ethnicity of every Nth person in the fanned mob, but I doubt it very much.

He was totally discouraged, took several long breaks and went home early.

Cincinnati 4M appears to be "Mitofsky Precinct Number 25" in the ESI report. (I'm not too worried about jeopardizing confidentiality here, given that (1) I am quoting widely available sources and (2) less than 2% of the voters were interviewed anyway.) Table 1 in ESI's data document reports a 68% Kerry share in the exit poll (that would be Kerry 21, Bush/other 10), with a possible range between 42% and 80%. So we can infer that the interviewer reported about 19 misses and refusals, for a total of 50 attempted interviews. Kerry's percentage would range from 21/50 = 42% to (21 + 19) / 50 = 80%. Looks like a match.

OK, so if an interviewer reports having attempted to interview 50 people out of 1700+ (for the four precincts sharing this polling place), and completes 31 interviews, with a result of 21 Kerry, 10 Bush/other (i.e., 68% Kerry) -- and the final result in the precinct is Bush 312, Kerry 211, other 3 (40% Kerry) -- and the same precinct gave 23% of its vote to the Democratic candidate for governor in 2002 -- who here is prepared to call this virtually irrefutable evidence of vote miscount?

[sunshinekathy]

26. Try Reading Appendix B - Factually it is Correct

Even though you may not like the way we define the words in our paper, the paper IS factually correct as you would realize if you read the derivations of the calculations in appendix B and understood them.

Nay-saying is very easy isn't it? You didn't have to spend months researching and writing the paper, or deriving the algebra, or even have to read the entire paper.

It would be much more difficult and time consuming to read the calculations before criticizing them wouldn't it?

That said, I DO appreciate your feedback and knowing how the paper is confusing still - and understandably so in this case by its use of this particular word.

However, I believe that we just used the exact same words as ESI did in their report, but I can double-check it for you. I had the same confusion myself when Ron first did this calculation because I couldn't see how he could do it without knowing the number which were not asked and the "nonresponders".

So it is confusing, I admit, but not if you read the appendix where the calcs are done.

[OnTheOtherHand]

37. I'm not at all sure that "nay-saying" is easier

Actually, Kathy, you might be surprised how many pages I have written on this issue that I have never done anything with because they depress me too much. But that is beside the point.

I've made extensive criticisms here; I will try to post my formal response within several days (after completing my response to your other paper). That gives you several days to sort through many of my concerns. Incidentally, you need to look a bit harder at ESI's Figure 2. Two symbols overlap -- which is a hint to a broader problem, because it is very likely that overlap accounts for the "missing data" in figures 2 and 3.

[RonB]

70. Is ESI Mislabeling its Graphs and producing Nonsense Reports?

OTOH,

I just checked Figure 1 of the ESI brief again. It says:

"The chart above shows the range of possible response rates for the Edison-Mitofsky 2004 Presidential exit poll compared to the official vote percentage by precinct. The gray bar REPRESENTS TOTAL POSSIBLE REPINES RATE , FROM ALL EXIT POLL NON-RESPONDENTS VOTING FOR BUSH AND ALL EXIT POLL NON-RESPONDENTS VOTING FOR KERRY." (CAPS MINE)

This is exactly what the algebra in Appendix B of our report does.

1) If by "non-respondents" they mean just "missed" and "refused" in your sense, this graph makes no sense - it says nothing about "the range of possible response rates" which are defined in E/M Jan. 19 report (as I recall - I don't have report with me) as share of exit poll that completed surveys out of the entire precinct.

2) IF you are right - ESI (and E/M) have been playing games with the public (per previous point) and with us - as we've been ask them to clarify the missing data, and other info in their graphs since August!

But then again folks who would produce a report with five out of forty nine data points missing (with no mention of this in the report) might be capable of this kind of "bait and switch" tactic as well.

I stand by the report based on ESI's stated definitions. If these are wrong, ESI (and Mitofsky) are to blame.

Moreover, regardless of this issue - I think this report makes clear that ESI's own "disproof of fraud" by comparing 2004 to 2000 precincts is bunk.

And also our Graph 3 that does not depend on the "non-responder" calculations shows the same kind of correlation with precinct partisanship that E/M (and unfortunately you and Liddle) refuse to acknowledge.

Best,

RonB

[Febble]

75. Ah, Ron -

Well, as I said - I was not entirely sure what ESI meant either, so I wrote for, and got, a clarification, which was that "non-respondents" were exactly as E-M described them in the E-M report - those selected who did not complete ("refusers" + "misses").

So, for whatever reason, it does appear that you have chosen the wrong interpretation.

And I believe you are wrong about the correlation - your conclusion does depend on assuming "model" rather than "actual" variance, if I understand you right, and I would argue that this assumption is not valid. There may be unexplained variance, but that doesn't make an error term based on model variance valid. It just means there is too much other noise to draw a conclusion.

Doesn't it? But I will give you the point that there is a hint of the relationship you describe ESI plots.

Finally: ESI do not, as I understand it, claim that their analysis is a "disproof of fraud" - merely that it suggests that fraud was not a major contributor to the exit poll discrepancy. They may be wrong.

But even if they were right it would not rule out fraud in Ohio - it would merely mean that fraud was not responsible for the exit poll discrepancy. There are many forms of fraud that would not result in an exit poll discrepancy.

[Febble]

79. PS: But, Ron,

if that was how you were interpreting ESI's chart (and I do agree it is confusing) did it not occur to you that it was giving you absurdly small precinct sizes?

Particularly as Kathy postulates at least some precincts of very large size in her paper, as her "counter-example 2"?

FYI - from the E-M report glossary:

Completion Rate
The Completion Rate is the percentage of sample voters who agree to fill out the questionnaire. The rate equals completed questionnaires divided by completed questionnaires plus refusals plus missed voters who were in the sample.

Refusal Rate
The Refusal Rate is the percentage of sampled voters who are approached by the interviewer, but who refuse to fill out the questionnaire.

Miss Rate
The Miss Rate is the percentage of voters designated to be in the sample that are missed by the interviewer because the interviewer could not physically approach the voter and ask them to fill out a questionnaire.

(My bold)

And from page 31:

completion rate for all exit poll precincts was 53%.

So it seems pretty clear that the 53-56% "completion rate" you cite from the E-M report in Footnote 22 is as defined in the E-M glossary, so I'm not quite sure why you would have thought it meant anything different.


BTW, I think the answer to your "missing data" question is simply overlapping points. All the data appears to be in the table, doesn't it?


Edit: re-formatted for clarity

[OnTheOtherHand]

81. you bear some responsibility, Ron

First of all, since you mention it: There is no evidence of missing data in their graphs at all; rather, in all likelihood there are overlapping points. In Figure 2 two of these points can even be seen, a circle and triangle at Bush 2000 = 54, Bush 2004 = 53. It is disappointing that no one is left to caution you against poorly substantiated allegations.

(I have no idea why Scheuren isn't answering your e-mails, although I am beginning to form an idea.)

The Figure 1 caption is misleading (it does not show "possible response rates" in the sense that we are accustomed to). But I cannot imagine how it could be read to mean what you have interpreted it to mean, either in isolation or in the context of other information. Nothing here claims to depict a range of possible vote proportions; all indications are that the ranges apply to the exit poll, not the official vote. The figure is titled Ohio Poll Ranges, is it not? The gray bars evidently represent the range of possible exit poll proportions. Compare Table 1, which says that the minimum and maximum percentages represent "attributing all refusals to" Bush and Kerry voters, respectively (although I don't think that is quite right, either, since I assume "misses" are included as well as "refusals").

On your interpretation, ESI was presenting clear evidence of fraud and... and... just didn't notice?! Didn't that, well, trouble you?

And, as Febble and I have noted elsewhere, your interpretation clashes with what we know about Ohio precinct size. For that matter, the assumption that about half the voters in most precincts completed surveys -- which is roughly what would be required to produce Figure 1 on your interpretation -- flies in the face of what we all know about interviewing rates. Those precincts could all have had an interviewing rate of 1 (which obtained in about 11% of exit poll precincts), but it would not have been my first thought, fer sure.

Ergo, I regard your interpretation as untenable.

"I think this report makes clear that ESI's own 'disproof of fraud' by comparing 2004 to 2000 precincts is bunk."

You should either document that "quotation" or issue a public apology to ESI. It is pretty easy to refute arguments if you get to just make them up.

"the same kind of correlation with precinct partisanship that E/M (and unfortunately you and Liddle) refuse to acknowledge."

By my analysis, that correlation isn't even statistically significant by a two-tail test. (EDIT: it could be for WPE, but not for ln alpha.) How about yours? Hmm... you don't actually conduct any inferential tests. Isn't it fairly unusual to assert a "strong relationship between Kerry/Bush vote count share and exit poll discrepancy" without, well, demonstrating one? Really, who do you expect to convince with this sort of analysis? and what do you hope to convince them of? Even if we stipulate that the correlation is meaningful, what does it mean? is it "Virtually Irrefutable Evidence of Vote Miscount"? why on earth?

I don't expect to spend any more time trying to explain the problems with your paper, until my response comes out. At that point I will refer to whatever version is then available, and see how many of the mistakes you have managed to walk back. That is all.

[Febble]

4. I'd just like to make one point here:

This paper states, correctly:

The precinct level Ohio exit poll data provided in the ESI report16 shows mathematically impossible results in two precincts. In precincts #47 and #25 17, even if all of the voters who did not complete exit polls (the “non-responders”) voted for Bush, the official Bush vote share could not have been as high as it was. In another precinct, #7, all non-responders would have had to vote for Bush to get Bush's official vote count, despite a 34% Kerry vote among exit poll responders in that precinct.

Thus in these two precincts, the error cannot have been due to "differential non-response".

However, differential non-response is not the only - nor, possibly, even the most common - mechanism by which bias can occur in a survey. The other mechanism, to which face-to-face surveys are prone, is non-random selection: in other words the problem may not simply be that if one group of voters (say Bush voters) are less likely to be willing to participate, they may also be less likely to be selected. Indeed the evidence in the E-M report is much stronger for this mechanism than for literal "differential non-response", in that error was larger where the interviewing rate was lower, thus creating greater opportunity for departure from strict non-random sampling.

To ensure that I had not misunderstood ESI's meaning of "impossible" I contacted Fritz Scheuren who confirmed that his finding simply ruled out strict "non-response" bias in these two precincts, but would not, of course rule out non-random sampling.

It is therefore perfectly plausible, whether or not any believes it, that the error in these two precincts was due to non-random voter selection. Indeed the evidence would suggest that non-random voter selection was a contributor to the exit poll discrepancy.

This is not to argue that fraud was not responsible for the apparent polling error in these two precincts; merely that it is not the only alternative.

(On edit: OTOH is correct - this point was not necessarily clear from the report, which was the other reason I contacted Fritz Scheuren, who confirmed that OTOH's interpretation is correct)

[OnTheOtherHand]

6. just to review:

(1) I have no idea what it means to say that the results are "mathematically impossible"; that may not be wrong, but it certainly isn't analytically helpful.

(2) "even if all of the voters who did not complete exit polls (the 'non-responders') voted for Bush, the official Bush vote share could not have been as high as it was" -- that is simply wrong. In "precinct #25" there were literally hundreds of voters who did not complete exit poll surveys; the official Bush vote share could have been well over 90%. The author probably intended to limit this claim to the people who were actually counted as non-responders, but then the claim would also have to be limited to Bush's 'share' in the exit poll, not in the official count.

[Febble]

10. Doh!

I admit I was wrong. The passage in the NEDA study I cited as correct is not in fact correct.

My understanding is that by "impossible" precincts, the authors of the ESI study meant precincts where the discrepancy was greater than it would have been if all those selected for interview but who had either refused or had been "missed" (because the interviewer was busy with the previous responder) had voted for one or other candidate.

I asked Fritz Scheuren this question directly, and his answer was that this was so. "Non-responders" refers to those who were selected, but who did not respond; it does not include those who were not selected.

So what IS correct (I believe) is that the error in the two "impossible" precincts cannot have been due to Bush voters refusing or being missed at a higher rate that Kerry voters. However, it remains perfectly possible (and indeed likely, given the anecdote above) that the error could have been due to Bush voters being selected for interview at a lower rate than Kerry voters. Unwilling responders may avoid both responding when selected and being selected in the first place.

And while I'm here.....

A point regarding "impossible" response rates for Kerry voters in "Bush strongholds": Sampling error alone will tend to lead to extreme completion rate values for voters for the candidate for the less popular candidate in extreme precincts. This phenomenon arises from the fact that where a group voters are a small minority in a small precinct, the probability of getting completed responses from all of them can be quite high, even if the likelihood of completion for any given voter, regardless of candidate preference, is 50%. In this instance the true completion rate for that group of voters will be 100%. The probability of none of them at all completing is also quite high, in which case the true completion rate for that group would be zero percent. However, the the completion rate for the majority voter group is unlikely to deviate very far from 50%. Thus the variance in completion rates for the minority group of voters is much greater than the variance in the completion rates for the majority group of voters. And this means that "impossibly high" (as well as "impossibly low") completion rates are not impossible at all for the group of voters who are in the minority.

However, "true" completion rates cannot be computed, as by definition we do not know the voting intentions of those who refused or were missed. USCV/NEDA, in previous papers, have suggested an algebraic formula for estimating it. The formula works fairly well for precincts in which support for each candidate is fairly even, and/or where sample sizes are fairly generous. However, in extreme precincts, where sample sizes are small, NEDA's formula can actually produce literally "impossible" completion rate values for the minority voters, due to sampling error alone. By literally "impossible", I mean values greater than 100%.

The plausibility of computed completion rates, using NEDA's formula, would therefore seem to be an unreliable guide as to whether fraud occurred in precincts where one group of voters is in the marked minority.

[OnTheOtherHand]

12. sorry to pick on you ;)

"So what IS correct (I believe) is that the error in the two 'impossible' precincts cannot have been due to Bush voters refusing or being missed at a higher rate that Kerry voters."

That is correct (except for the typo) in the technical sense that I think you intend, but I think not in ordinary English.

The case of Cincinnati 4M points to a very high rate of unrecorded misses (and/or refusals). If the DU report is accurate, the interviewer had no way of distinguishing voters from the four precincts. Since AFAIK interviewers are instructed (depending on the polling place and sometimes the time of day) to interview at least every tenth voter, the interviewer should have attempted well over 100 interviews.

The interviewer in 4M apparently reported an above-average 62% completion rate (31 interviews out of 50 attempts), but that estimate can hardly be regarded as reliable on the evidence available to us.

(But even if we did assume that the completion rates were correct, you are right that the results still wouldn't be "mathematically impossible." There is actually about a one in 1000 probability of drawing 21+ Kerry voters out of 31 interviews in 4M by chance alone -- not that I think for a moment that the interviewer actually achieved a random sample.)

In the other 'impossible' Ohio precinct, Mitofsky number 47, the exit poll proportion is 43%, ESI reports a minimum and maximum of 36% to 53%, and the vote proportion was 32%. From the minimum and maximum, we can estimate that the interviewer reported a completion rate of about 83%. I think it is untenable to assume that this rate Must Be correct.

But it is also untenable (again, as you point out) to assume that an 11-point difference between the exit poll and official proportions is "mathematically impossible," no matter how many or how few non-respondents (by whatever definition!) there were. The average Ohio precinct had about 500 votes, so let's say that the precinct split 160 (32%) for Kerry, 340 for Bush, and the interviewer completed 70 interviews -- 30 (43%) for Kerry, 40 for Bush. I reckon about a 2.6% chance of obtaining 30+ Kerry tallies out of 70 interviews by random chance alone, even without bias, never mind fraud. Nothing remotely 'impossible' about it, at least on the evidence available to USCV.

[Febble]

14. Thanks

I seem to be having a bad day.

I take it we both agree that the discrepancy in "impossible" precincts can "possibly" be due to some form of polling error, if not literal "non-response" bias, as in more Bush voters failing to respond once selected.

Actually I was surprised that Scheuren categorized non-random sampling problems as "measurement error" rather than "non-response bias", neither of which I think are particularly good terms for what we are alleging, and which seems to have been a real phenomenon, at least in Cincinatti 4M. "Unrecorded misses/refusals" seems to be a better way of describing it. Thank you.

However, leaving aside what we call it (and my typos - got the wrong specs again) what seems clear is that if non-responders are not properly recorded as such, then completion rates won't tell us much about the possible limits of bias. And actually, if, under Cincinatti 4M type conditions, only those willing to respond end up being recorded as having been selected, completion rates may end up being higher under the very conditions where bias is more likely.

None of which, of course, rules out fraud as a contributor to the discrepancy, but it does tend to make the gun less smoky.

[OnTheOtherHand]

16. yes, I begin to expect a "U" relationship between completion rate and bias

where either very high _or_ very low rates signal likely problems. (But I don't mean to imply that completion rates are a good _proxy_ for likely bias -- we can say that the outliers are suspect, but we don't know what to think about the precincts that end up in the middle.)

And yes, there is plenty of room for fraud in the data. But also plenty of room for non-fraud. More smoke than light, one might say (Yogi Berra, perhaps).

Am I hallucinating, or on p. 10 does the paper use one-tail confidence intervals for comparisons that go in both directions? Umm, that might come out in the wash, I guess.* (By the way, they note that apparently there were an average 41.7 respondents per Ohio precinct, so my hypothetical example about precinct 47 perhaps should have been about getting 18 (still = 43%) or more Kerry voters out of 42 interviews. That's over 8%.)

*EDIT: Yeah, the idea of a "one-sided confidence interval" makes me queasy, but the basic finding here is just that there is red shift in a lot of Ohio exit poll precincts, and I certainly agree with that.

[mgr]

19. I think we have a clue:

"But I don't mean to imply that completion rates are a good _proxy_ for likely bias -- we can say that the outliers are suspect, but we don't know what to think about the precincts that end up in the middle."

It is strange that we only have Cinn-4M identified among the NEP precincts, but what has been described by Liam Laddie and Kiwi ExPat is consistent with what was reported by NEP's report in January. I think this is a good indication that control of the sampling design was undermined in Ohio, and that one cannot assume with any confidence that bias when obtaining voter responses did not come into play with any of the precincts, as is presumed in this report. There may be a couple, but I suspect the problems at Cinn 4M are repeated practically everywhere in Ohio.

I think the evidence suggests that we may want to place more confidence in Mitofski's assertions than we have previously; and use them as guidance of where to the Sinister Signpost might be. This does not change my opinion that little of value regarding the demographic trends that are NEP's purpose to measure can be gotten from the 2004 exit polling. Unfortunately, that confound has me as more worried, since it can be a base line to allow future perfidy.

Mike

[OnTheOtherHand]

25. yeah, we don't know how raunchy the demogs might be

The conventional wisdom is, probably not very, because even if vote share is off, demogs mostly aren't highly correlated with vote share, so they are probably off less. (This is the flip side of the argument that even if the gender balance was substantially off, it wouldn't affect the vote share very much.)

It's a good argument 'on average,' but it is hard to be sure that it applies in any one case. And incidentally on the demogs, the observed (or reported) completion rate for 60+ respondents was over 12 points lower than for younger respondents. Who can say what other discrepancies like that might be lurking, or might pop out in any given election?

I think we had better work hard on making elections directly auditable, because these indirect audits are hell.

[mgr]

40. Already its being used...

to show Bush's inroads into the Jewish electorate. But when you review the report, only three respondants indicate they are Jewish, and that translates to 70%? Pretty interesting weighing.

Mike

[OnTheOtherHand]

45. umm, you mean 3% indicate they are Jewish?

I'm not even sure what report you are talking about. But the final adjusted national NEP exit poll seems to say 3% of voters were Jewish, and 25% of them voted for Bush, up 6 points from 2000. I haven't checked whether the pre-adjustment numbers vary.

[mgr]

52. Unintentional ellipsis (sheepish)

Rushing through things today. Normally I vet this stuff by checking the original source after I draft it. Yes, I meant 3%.

Not a report, but a recent news article had it at around 28-30%.

The problem is that I evaluated the claim in

http://pillageidiot.blogspot.com/2004/11/hidden-jewish-vote.html

and found that for Los Angeles area it ran closer to greater than 80% support.

I suspect that this column was the source for the argument against Jewish support for Kerry in Florida TS counties that was used to counter the Berkeley study, until the symmetry issue was raised.

Mike

[OnTheOtherHand]

82. tell ya what...

One of my colleagues has spent some time pondering how different surveys will get different proportions of "Jews" or "Jewish" depending on how the question is presented. I haven't paid attention to the details. But I would be really cautious about any argument about "Jewish voters" (or, for similar reasons, "Hispanic voters"). Certainly it would be risky to generalize about either Jewish or Hispanic voters in Florida on the basis of voters in Los Angeles. But that's just a general point; specific args should be evaluated on their merits, of course.

[nashville_brook]

34. what you say reveals to me the level of uselessness in questionable

elections. the goal is to be beyond reproach. this is the cornerstone of our democracy. the process and the results should be crystal clear.

i live in a diverse neighborhood and talk every chance i get about this stuff. it is rare that i talk to anyone who thinks there is any honesty in the system. it's my sense that this contributes to the "why should I do anything," attitude.

why should i vote?
why should i bother to "play the game?"
why should i continue to be their fool?

apathy has been cultivated and it's working pretty dang well for the Sinster Ones.

[OnTheOtherHand]

39. yabetcha

Right, the process and the results should be crystal clear. That has to be our goal.

[kiwi_expat]

80. Bad weather and 100 ft limit affected most Ohio NEP precincts

"It is strange that we only have Cinn-4M identified among the NEP precincts..." -mgr

Thanks to brilliant work by skids, we know the counties for 45 of the 49 Ohio NEP precincts. We then identified 16 of the precincts. (See minvis' Ohio Exit Poll Raw Data thread
http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=329749 )

We know about the Cincinnati-4M NEP interviewer's problems from liam_laddie's discussions with the Democratic Party observers for that precinct.
http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=397259#397543

Two of the interviewer's main problems were rain and Blackwell's 100 foot limit. The 100-foot limit - which placed the NEP interviewers with the (mainly pro-Kerry) campaigners - affected most NEP precincts in Ohio, for at least part of the day. According to the E-M Report, the last NEP precinct to lift the restriction did so at 5:00pm.

[Scout1071]

54. Are you guys even speaking English?

Sorry, but my eyes began to glaze over. You took me back to school and the one subject I struggled with the most, math.

Great work. It all sounds very convincing!

[OnTheOtherHand]

56. sorry, that was Insider Baseball -- no trauma intended ;) n/t

[Scout1071]

61. No worries. Just thought you'd be amused that my brain actually hurt

while I was reading it. Slight tinges pulsing through my brain.

But, what can you do? Not all of us can be great at everything. And math is a weak point for me.

Carry on. I'll look forward to reading the abbreviated end analysis.

[RonB]

72. More on ESI/EM withholding data, missing data, dodging, and obfuscation

Febbble,

Since Scheuren won't talk to me - I guess I need to get this from you.



So now the ESI report is just an exercise in finding "non-response bias"!

Why produce a whole graph on this and make a big deal about it?

We know the E/M story that "selection bias" - that a majority of pollsters were Kerry partisans so they selected Kerry voters disproportionately - is the major (though covered up) reason for some kind of "pervasive discrepancy" that is not obvious from the data.

So what's the story with:
1) the clear correlation with partisanship shown in graph 3 (which has nothing to do with the non-responder issue)
2) the missing data in the ESI figures
3) the empirically and methodologically illogical ESI attempt to "prove no vote fraud"
4) Why can't you guys provide a serious EXPLANATION of the unexplained discrepancy instead of continuing to pretend that it doesn't exist!

Its bloody outrageous that this situation can go on for a year - they have all the data, all the definitions, and make selective releases with terribly ambiguously labeled (or not labeled at all) graphs, axes, etc.

We ask for clarification for months - they refuse to respond and now according to you guys what they "apparently" produced is not really what it looks like but something else that seems hardly relevant at all!

The correlation exits, the discrepancy is unexplained - this unethical behavior by ESI/EM et. al. is an outrage.

[RonB]

73. Pattern Issue Raised by Report May Be Even More Pertinent

IF OTOH and Febble are right - the three precincts may not be "mathematically impossible" to obtain but they still remain highly unlikely from a statistical point of view. What's the likelihood of getting something like 24% "excess selection bias" when picking every "n"-th person that comes out of the precinct - as pollsters were supposed to do!

In any case, the key issue raised by the paper remains, what about the PATTERN of the "non-responders" (in this new definition)?

This may be even a bigger problem to explain under this definition.

Why is it that in partisan Bush precincts "missers plus refusers" become more and more tilted toward Bush voters?

As stated in the report, one would expect just the opposite - that is more Kerry partisan pollsters in partisan Kerry precincts.

I don't think this definitional clarification gets ESI/EM out of the woods. Rather, it may make the need to explain the pattern even more pressing as the various (non-sampling error) factors probably have even more affect with regard to response than with "selection".

And note Febble (and OTOH) these data reveal a striking pattern!

Enough with the pretense that no explanation is necessary!

[Febble]

78. I don't think ANYONE here

certainly not me, is indulging in any "pretense" that "no explanation is necessary", and I know for a fact that OTOH, and myself, and many others, are continuing to investigate ways in which fraud in Ohio is compatible with the ESI finding (and there are many).

But I do dispute your "striking pattern" for reasons given elsewhere. It may be a pattern, but it falls short of statistical "striking"-ness. But it could, indeed, be because votes were shifted towards Bush, particularly in precincts where he was doing badly relative to his 2000 performance.

But I don't think that the patterns demands to be explained, because it simply isn't a statistically significant pattern, unless you ignore all variance apart from that ascribable to sampling error. There could well be sources of polling error that are essentiall random, but more than sampling error. I have suggested one, and the Cincinatti anecdote would seem to support it: breaks in coverage coupled with lumpy patterns of voter arrival. This would not produce systematic bias necessarily, but it would certainly add to the total error variance. And I don't think you can ignore the actual, data variance when computing your correlation.

[Febble]

77. Ron, I can see why you are cross

I found the report very frustrating myself, especially the mislabelled bar graphs and all.

But I am not ESI - I just had to email them.

I'll do my best to address your questions, but I can't speak for ESI.

I didn't really see the point of that bar graph myself, even when I'd understood it - it seemed just to put theoretical limits on the possibility of literal "non-response" bias - without taking into account sampling bias. But I suppose that was mildly informative.

Your other points:

1. I do dispute your correlation - I agree that there is a trend, but I don't think you can regard it as a "significant" trend by using "model" variance as you have done.

2. You'd have to ask ESI about "missing data". If I find out I'll tell you.

3. I've dealt with the apparent "illogic" elsewhere - the "illogic" derives from an misreading of the hypothesis which concerns fraud as an explanation for the discrepancy, not the existence of fraud. Fraud could well have occurred and not been responsible for the discrepancy. I think that is quite likely, myself.

4. No-one is "continuing to pretend" that the discrepancy "doesn't exist" - cite please? E-M's entire report was about the discrepancy, so is the ESI report. But the ESI report, and Mitofsky's new analyses, does suggest it wasn't due to fraud (which is not the same as proving there was no fraud. The fairly obvious alternative candidate is some form of biased polling, for which there is evidence in the E-M report - either non-response bias or actual sampling bias.

[paineinthearse]

7. Full disclosure

I fear this important news will be locked due to flaming. It is not inevitable. How? Anyone who is on the payroll of a political organization, PAC or elected official should identify their connection.

Sunshinekathy, your affiliation?

OTOH, your affiliation?

Febble, your affiliation?

thank you.

[OnTheOtherHand]

9. I am (as my profile states) a professor at Bard College

-- a former participant in the USCV stats list

-- met Mitofsky once, have corresponded with him via e-mail, have posted scatterplots he provided

-- met Scheuren once and spilled water on him

-- I suppose my political contributions are on the record Out There (they are Democratic), but I am not on any political payroll (nor Mitofsky's, nor ESI's)

I think affiliations are really a red herring here. My opinions of sunshinekathy's arguments are evident, but I certainly don't think that she is willfully misrepresenting facts in support of an agenda. The arguments should be assessed on their own merits.

[Febble]

11. Most of mine is in my sig

But to clarify:

I am presently a Research Associate at the University of Nottingham, UK, and my field is cognitive psychology (incidentally a heavily statistical field).

I am a member of the British Labour Party and have been for about 25 years. I have been on the verge of tearing up my Party card on many occasions, but have not yet done so as I believe in working within the Party for change rather than outside it. I voted Labour at the last election, and hope to use my Party vote to get rid of Tony Blair as leader at the earliest opportunity.

I have done analyses of election 2004 issues for USCV, and I have also done a series of analyses for Warren Mitofsky on the exit poll data. It was one of my analyses that was presented by Mitofsky at the Freeman-Mitofsky debate. I was paid to do those analyses. The work of mine that is posted on the USCV website was voluntary.

The reason I became interested in Election 2004 issues is that I desperately wanted Kerry to be president, just as I desperately wanted President Gore in 2000. Given the fact that Gore should have been president, my default setting was to assume that Kerry should have been too. In a sense, it still is.


(edited for typos)

[Febble]

38. I forgot to add

my name is Elizabeth Liddle.

You can google me if you want.

[Nothing Without Hope]

8. Recommended with thanks. n/t

[kster]

13. Great work sunshinekathy your under attack,you must be on to something

K&R

[OnTheOtherHand]

15. actually, I defended her -- but her work has problems. n/t

[nashville_brook]

20. bingo!

ding ding ding

[Febble]

21. In answer to your question

on PI:

http://www.dailykos.com/user/Febble

http://www.dailykos.com/user/Febble/comments

I mostly post on general issues on DailyKos, but on the election issue, this is a better forum.

I'd also, despite what you may think, rather like to know how Bush managed to win the election, and I've simply come to the honest conclusion that the exit polls don't give us much of a clue.

Analyses of mine reported here:

http://uscountvotes.org/index.php?option=com_content&task=view&id=70&Itemid=43

and here:

http://uscountvotes.org/index.php?option=com_content&task=view&id=65&Itemid=43

are more likely to point us in the right direction I think, as are many other studies and investigations into the anomalies in Ohio, New Mexico, Florida, and elsewhere.

[DemReadingDU]

17. recommended

[bleever]

18. Recommended.

Thanks, sunshinekathy.

[autorank]

22. The Truth wil out...

They can run but they can't hide. The motivation for being a "debunker" of your own work, i.e., Mitofsky, is so odd it's not worth commenting on.

The truth of the matter is that the election was stolen.

If you'll betray a high leve sting to prevent a nuclear attack on your own country (i.e., the outing of Valerie Plame), then we can presume you will do just about anything.

When there's suspicion of a crime for BushCo, then there is reason to take the evidence seriously.

This is serious evidence.

Bush is a criminal. Those who defend him are enablers.

[OnTheOtherHand]

23. is someone defending Bush? n/t

[nashville_brook]

24. here's my thing

the mathmatical nit-pickers would improve their standing with the group by offering positive help once in a while. i've never seen that.

[Febble]

28. Have you had a look here?

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x395044

Or here?

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=398267&mesg_id=398267

Or here?

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=398460&mesg_id=398460


You may not have found it helpful but they were certainly positive.

Here's my thing:

If someone comes to the honest conclusion that people are looking in the wrong place (metaphorically) for fraud, then wouldn't it be helpful for that person to suggest the right place to look?

Now, you may disagree with my conclusion that the exit polls are tellin you where NOT to look, which is fair enough. Nonetheless it is my honest conclusion. And what that conclusion leads me to suggest are the following:

That voter suppression methods were probably the most effective
That suppression of votes in Dem precincts, or multiplication of votes in Rep precincts would escape detection by exit polls
That concentrating fraud in large precincts, especially where Bush was doing badly relative to 2000 would be most effective, and stand more chance of not showing up in the correlation presented by Mitofsky in his debate with Steve Freeman.

These strike me as fairly positive conclusions.

Not that they were original to me - they were suggested in the course of positive and constructive online conversations with DU posters including Land Shark, OTOH, eomer, anaxarchos, BillBored, TimeForChange, and others.

[foo_bar]

36. "telling the truth" isn't capitalized, unlike Divine Truth

Welcome to the age of reason:

As Mystery and Miracle took charge of the past and the present, Prophecy took charge of the future, and rounded the tenses of faith. It was not sufficient to know what had been done, but what would be done. The supposed prophet was the supposed historian of times to come; and if he happened, in shooting with a long bow of a thousand years, to strike within a thousand miles of a mark, the ingenuity of posterity could make it point-blank; and if he happened to be directly wrong, it was only to suppose, as in the case of Jonah and Nineveh, that God had repented himself and changed his mind. What a fool do fabulous systems make of man!

http://www.infidels.org/library/historical/thomas_paine/age_of_reason/part1.html

[kster]

42. With all the 2004 election theft FACTS

we do not need to be debating any longer, its what happened in 2004 and now its time to do something with all the evidence and FACTS.

Let the enablers debate amongst each other. We need to be spreading the truth far and wide.

30 Votes later this thread rolled the truth right past them. Great Work,everyone!

[nashville_brook]

57. the fact that the votes are so high, and the "debaters" are "debating"

all by themselves speaks for itself.

[autorank]

64. Democracy is a wonderful thing. No rigged vote counting here!

And it will be increasingly uncomfortable for people to apologize for a Bush victory and Bush bad behavior (bBb). Why? Read my new avatar. We have Reid in action. What a sight to see!

Give 'em Hell Harry

[autorank]

63. Resign now Bush. Your time is up. Who favors that?

I do and I think about 99.9% of DU does also.

[katinmn]

27. Thanks. There's so much evidence the election was stolen

that it's mind-boggling.

The only way to prevent it in the future is to throw out electronic voting machines. That's not likely to happen since states and counties are still buying them.

However, we still have time to demand voter verified paper ballots and random audits!

Where's the Senate companion bill to the Holt bill?

[sunshinekathy]

31. You're welcome & there are other solutions too

Here are the two primary solutions to ensuring vote count accuracy

http://uscountvotes.org/ucvAnalysis/US/election_officials/Audits_Monitoring.pdf

One solution needs to be done at the state and local level, the other is to build the national election data archive and work with local volunteers to collect all the data we'll need.

We can really use your help, either as a volunteer or by donating money so we can hire a fulltime programmer and other staff we'll need to implement the national election data archive.

http://uscountvotes.org/ucvData/US/How2CollectData.pdf

Please read more on our web site, sign up for our announcement email list, and donate if you still have a paying job and haven't quit it to work full-time on this like I have.

I appreciate your appreciation very much. Some days it seems like I'm criticized from all sides.

Best,

Kathy

[katinmn]

33. I'm working with this group locally

http://www.electionintegritymn.org/index.php

When it comes time to get MN state data, this is de place!

Thank you, Kathy, for your efforts.

I am in the same boat as you are. I quit my office job last December and now work from home. I did that so I could have more flexibility in working for change.

[nashville_brook]

35. me too! it's the best thing ever!

here's to homeys!

:toast:

[katinmn]

43. Whoot-whoot!

Hom-eeeee!

[Marleyb]

29. Rove and the media knew the exit polls are the smoking gun...

otherwise why would exit polls be "adjusted to reflect official vote tallies"? Isn't that admission that yes, the exit polls and outcome are supposed to match or you have a problem? But you can't just fix the problem by changing the numbers you don't like!


"Rove said he felt sick, then got mad when he started reading exit polls on Election Day as Air Force One returned from a final campaign swing. Surveys of voters just leaving polling places around the nation tilted toward Kerry early in the day and through much of the evening, causing early optimism for a Democratic recovery of the White House.

That faded through the night as exit polls were adjusted to reflect official vote tallies. But the exit poll was still tracking toward Kerry as the president's plane landed at Andrews Air Force Base in suburban Maryland after Bush's quick Election Day stop to thank supporters in crucial Ohio.

"I was on Air Force One, and we were literally on final approach into Andrews," Rove said in another Sunday morning broadcast interview. "The phone connection kept cutting out. I was holding a piece of paper on my knee, trying to scribble it down, holding the phone in the other hand. I got sick as I wrote them down, and then when I looked at them, I got angry, because they simply could not be true."

"I mean, it had us 19 points down in Pennsylvania. It had us 17 points down in New Hampshire. It had us 1 point up in Virginia," Rove said. "I mean, you looked at these numbers, and you realize, this is just insane."

http://www.cbsnews.com/stories/2004/11/07/politics/main654122.shtml

[nashville_brook]

30. and you don't even need math to SEE the LINES; the OBSTRUCTION

we were at our neighborhood pub watching returns and as they showed the lines -- from a HELICOPTER view -- people went nuts. can you imagine showing up to that? on your lunch hour? after you've been slaving at your job all day...with the kids in OVERTIME at day care or whatnot.

if there were any GOOD FAITH in 2004, they would added voting days and MADE DAMN SURE those peoples' votes were counted. their voice was stolen. that was the first strike.

the changing of the polls late at night -- stirke TWO.

the homeland security clampdown -- STRIKE THREE.

no algebra needed.

[Febble]

32. Actually, I agree

with 1 and 3.

2 is my only problem.

Well, I'd at least like 3 investigated.

[sunshinekathy]

41. New Updated version & One small screwup - E/M date is Jan 19th

One small screwup in the press release. I didn't notice that the person who wrote the first draft of our press release put the wrong date in for the E/M report.

It was Jan 19, 2005 not Jan 21.

My bad - but a relatively small error.

The Ohio exit poll report has been updated just now too, so please use this latest version if you're printing and distributing it to anyone who needs to see it (I can think of a few people).

http://electionarchive.org/ucvAnalysis/OH/Ohio-Exit-Polls-2004.pdf

[Febble]

44. Kathy, I'm staring

at Ron's Appendix B, which states that Kerry's minimum vote share is equal to (K + B)/(K + B + N) where K is the number Kerry responders, B is the number of Bush responders and N is the number of "non-responders".

Now: if N is the number of those who were selected for interview but who refused or were "missed" - i.e. noted for selection, but were not approached because the interviewers were busy, then this formula gives Kerry's minimum share of the the voters sampled, not his minumum share of the vote.

Similarly, Kerry's maximimum share of the sample is given by:

(K+N)/(K+B+N), NOT his maximum share of the vote.

Only if N= the total number of non-participants do these formulae give Kerry's minumum and maximum shares of the actual votes.

So I will assume that Ron does mean that N is the actual number of non-participants, not "non-responders" in the sense that ESI use the term.

But he then appears to use ESI's min and max figures, which refer to the miniumum and maximum share NOT of total vote (as I understand it from Fritz Scheuren) but of the exit poll sample (including responders, refusers and "missed" voters - those selected but not approached).

This was the point that OTOH made upthread, if I have understood him correctly, and you referred him to the appendix I have just been trying to figure out.

In both the opening paragraph (of the abstract) and in the appendix, it would appear that Ron regards ESI's minimum and maximum figures as referring to vote share NOT sample share. My understanding is that this is NOT what ESI meant. Perhaps you should check it out.

[OnTheOtherHand]

46. it's not just your understanding

The numbers in the report simply cannot be made to add up. Baiman seems to be firmly convinced that there are approximately 75 voters per Ohio precinct (42 exit poll respondents and 33 non-respondents, on average). Once it is acknowledged that this cannot possibly be true, the idea that ESI was describing minimum and maximum vote shares -- as opposed to minimum and maximum exit poll proportions, as the table rather clearly indicates -- collapses, and I'm not sure what is left of the argument. The observation that there are lots of precincts with red shift?

[sunshinekathy]

47. Stop fabricating - no where did Ron say anything of the kind

[Febble]

49. Appendix B:

If K and B are the number of Kerry and Bush voters who responded to the exit polls and N is the number of those voters who were non-responders (who were either not asked or refused to answer the exit polls) then:

Kmin = K/(K+B+N) is the minimum Kerry vote share where no non-responders vote for Kerry, and:
Kmax = (K+N)/(K+B+N) is the maximum Kerry vote share, where all non-responders vote for Kerry.

And Ron then uses ESI figures for Kmin and Kmax which are the minimum and maximum values for Kerry's share of voters selected NOT Kerry's vote share:

ESI provided the Kerry vote share, k, Kmin, and Kmax, so we can use them to calculate the actual Kerry vote share of non-responders in the same precincts. Based on this expression for X, the Kerry share of non-responders, and assuming non-respondent samples of 3056, we can generate estimates for the probability of the non-responder split necessary to produce the reported official Kerry vote, given the exit poll result (or the one-tail P-value for X),using the data provided in Table 1 of the Kyle et. al. ESI brief, in Table 3 of the report

So the Appendix and main text do seem to agree with each other - but not with ESI, and yet ESI's figures are used to calculate "the actual Kerry vote share".

[Febble]

51. Ah, here we are: Page 6

Data suggest that the average precinct in Ohio had more than 32.8 non-responders

and:

Page 10:

Assuming precinct sample sizes of 30 exit poll respondents gives us conservative estimates of standard deviation for precincts in Graph 3

Adding responders (sample size) to "non-responders" we get 62.8.

So if Ron thinks that his "N" is ALL those who voted but did not respond to the exit poll, it would appear he assumes that the average number of voters in a precinct is 62.8. (Not sure where OTOH got his numbers from, but I got those from Ron's paper).

If Ron assumes that, he must be wrong (because the precincts were way bigger than that). Alternatively, he must assume, correctly, that that number is the number of voters selected. But in that case, the Kmax and Kmin are not the max and min of Kerry's vote share but his share of the selected voters. So someone's confused.

Edit: just noticed the "more than" 32.8 responders. Not sure what "average precinct...had more than...." means, so now I'm not sure what Ron thinks. However, the fact is that ESI's definition of "non-respondent" is not the same as Ron's and yet it is ESI's figures that Ron has used.

Edit2: Here we are; footnote 22: "The average exit poll response for Ohio was 2042/49=41.7 – see www.exit-poll.net/electionnight/
MethodsStatementStateGeneric.pdf and the national average exit poll completion rate was between 53% and 56% (see E/M Jan. 19 report, p. 37) so a low estimate for an average non-response rate in Ohio is 44%. This would imply that the average number of non-responders would be 0.44 x 41.7 / 0.56 = 32.8".

So it seems Ron DOES think there was an average of only 32.8 "non-responders" in each precinct. Which is correct, as long as you don't then think that "non-responders" includes all non-interviewed voters in the precinct, which Appendix B implies

[OnTheOtherHand]

59. yes, 41.7 + 32.8

I would say I was being facetious when I said that Baiman seems "firmly convinced" that there are 75 voters per precinct -- except that as I also said, if he doesn't believe that, the entire argument about "virtually impossible election results" appears to vanish in a puff of smoke. Which I suspect is the correct outcome.

[sunshinekathy]

48. Please read the Appendix B - it explains what N is

[Febble]

50. Well, it explains that N is

the number of those voters who were non-responders (who were either not asked or refused to answer the exit polls)

But this is not ESI's definition of "non-responders".

Let:
C be voters who completed an exit poll interview
R be voters who are selected for interview but who refuse to participate
M be voters who are selected for interview but are not approached as the interviewer is busy.
E be Everyone Else (people who were not selected for interview at all, i.e. the majority in all but fairly small precincts).

C = K+B (as defined by Ron)

Ron's definition of N = R+M+E.
ESI's definition of N = R+M.

That E makes a large difference.


Edit: As I have posted on the my other responses, Ron appears to believe that the mean of C+N is around 63 (C=30, N=32.8). If so, N cannot possible = R+M+E. It must = R+M. But in that case the formula does not give Kerry's vote share as stated, but his share of the selected voters. There is a very large missing E.

[sunshinekathy]

60. Thanks - I've forwarded your comments to Ron

and asked him to get on DU and discuss it with you.

He probably won't do so until tomorrow I'm sure.

I asked Ron the same question and he assured me that ESI meant everyone that did not answer the exit polls, not just the nonresponders in the typical sense - I didn't check, but another member of our list agreed with Ron too on this - I was too swamped working on other things and my History and Proof papers.

I'll let you work it out with Ron.

[sunshinekathy]

68. On second thought - Ron is definitely right

Even if, as you insist, ESI has a different meaning for the word "nonresponders", Ron is clearly applying the result of his derivations in Appendix B correctly, given the data that ESI present in their table, where clearly they claim to give the Kerry minimum and maximum possible vote share for precincts, and the actual official kerry vote share.

It is pretty hard to misinterpret the data in ESI's table, unless they've really misled us, so, in spite of your having a disagreement with the way that Ron uses the word "nonresponders" and his use of the letter "N" in his derivations which apparently you would like to have him use another letter instead or mean a different thing than he does for "N", Ron clearly did correct derivations in Appendix B for what he was trying to do, and clearly it allows him to make excellent use of ESI's data to determine the Kerry vote share for all those who did not answer exit polls.

I think it is very clever. David Dodge I think helped Ron develop the idea for those calcs originally.

[Febble]

74. Well in which sense is Ron right?

Is N the total number of voters who did not participate (i.e. all voters minus responders), which you seem to suggest, in which case Ron really DOES appear to think that there are only around 70 votes in a precinct, which is clearly wrong?

Or is N the total number of voters who were selected and did not participate, which is what ESI means, and which would make Ron right about the numbers of non-responders, but would make his inference about what they mean wrong?

I've no problem with the use of the letter "N" - I'd just like to know what Ron thinks it means, because it strikes me that he is confused.

And if the basis on which the paper is based is a confusion about that N, then, Houston, I think we have a problem.

I'm not saying there was no fraud in Ohio, just that until Ron can tell us what the calcs mean, I don't think we can conclude that there was fraud in ohio from those calcs.

[sunshinekathy]

84. No - you were right - Ron was wrong

I woke up this a.m. and double-checked for myself, the column definitions in the ESI data that Ron was entering into his formulas.

While his formulas were brilliantly derived, he was wrongly interpreting the ESI data.

I had much earlier, before releasing the paper, asked Ron if he was sure that ESI meant ALL the voters in the precinct who had not answered the exit polls rather than just the nonresponders in the sample, and Ron had wrongly assured me that ESI meant ALL the voters in the precinct who had not completed exit polls.

From now on, I will not let any of Ron's work go out without checking every detail of it myself. Ron is so self-assured and stubborn about his own work lately that he has consistently refused to seriously consider things I have to say when I spot something that is wrong, so I "should" have known not to trust him on this too. I accept blame for publicly releasing this error.

However Ron has become difficult for me to correct lately when he errs. For instance, his use of the hypergeometric formula in this paper is also not logically correct in this instance, but he refused to consider my reasoning on it, so we left it in the appendix because it really isn't germane to anything in the paper anyway. Ron told me this a.m. "You're unethical" because I refused to include another known error into this paper that he wanted me to include in it. And he is still angry at me for not including another calculation of his into our prior "history" paper that is not used in a logically correct way - i.e. multiplying individual state probabilities together - i.e. a multiplication of integral probabilities similar to what Freeman originally did and is more intelligently not doing any more) With this paper Ron wanted to state in it that the vote fraud is taking place in precincts where there are more Republicans and that the vote fraud is "administrative and political" even though vote fraud clearly shifts the precincts with higher WPD into the higher Republican precincts and Ron had not done the analysis to see where the precincts might have originally come from and zip can be concluded from the data about how the vote fraud is being done, or, in fact, even if it is vote fraud or just innocent miscounts. If refusing to publish his errors is "unethical" according to Ron's definition, then baby you haven't seen anything yet, as compared to how "unethical" I'm going to get in the future in refusing to publish errors.

It's too bad that I did not catch this worst-of-all gross error that Ron made and was too trusting when he assured me that ESI was giving the maximum and minimum possible VOTE rather than the max and min values of exit poll results in their data.

I will from henceforth not publish Ron's work without checking every detail of it myself and not trust what he tells me. Ron is leaving for the weekend he says he cannot fix the paper until he returns, so we are temporarily withdrawing the paper until Ron has a chance to redo it. Which I shall do now.

I do want to mention, that if Ron had been using the correct definition for ESI's data, his analysis would have been exceedingly brilliant. His ability to derive useful algebraic formulas is wonderful. Ron's mistake was an innocent one - due to his not understanding the definition of "nonresponders".

Thank you for catching this and pointing it out to us.

[sunshinekathy]

85. Stupid posting on my part

I had neglected to check Ron's use of ESI's data by reading the definitions for the column headings in ESI's table of data.

My bad.

I had spoken to Ron over the phone and questioned him on the definitions of ESI's data, because it did seem a little odd to me, but apparently Ron was not really listening to what I said, or had not carefully checked ESI's data definitions like he told me he had.

It was my fault, however, for not doublechecking for myself.

One of my flaws has always been being too gullible.

[mom cat]

53. There is certainly enough evidence for a thorough, impatrial investigation

If these results had been behind Hugo Chaves' election, George Bush would have invaded Venesuela by now screaming "election fraud" all the way.The election stank to high heavens and has left a severe distrust of our government.
It is just unimaginable that the exit polls would have been so wrong all of a sudden. There is no explanation I have heard that can pass the sniff test that can justify the last minute turnaround...not one.
I do not believe that George Bush really won. For the sake of democracy itself, this election must be investigated.

[nashville_brook]

55. the sniff test is all that matters to real Americans and they've already

made up their minds that both elections were stolen and WORSE, that no one came to their aid.

[OnTheOtherHand]

58. "real Americans (have) made up their minds..." -- source please? n/t

[autorank]

65. Yes ... and what a statement it is to see 55-50% Disapproves for Bush

I see "approve" move between 35% and 41% on the various polls. Those numbers are artificial. Why? Because the f'ing news media is still not doing it's job. All they have to do is report a story for more than a day or two. But then again, the long arm of Federal benefits - punishments has reached out to greet the CM (corporate media).

That's all right, watch CM devour itself when some of them realize that fair and thorough reporting and investigation beat Faux-like fantasies.

We will win. The Truth is on its way.

[Bill Bored]

62. So what have we learned from this thread so far?

1. First of all, the biggest mistake both sides of the Exit Poll Debate seem to have made so far is NOT READING ENOUGH DU! We've had people on the ground in Ohio checking this stuff out for months now, identifying the precincts, looking at the ballots and the ballot order rotations, how the votes could have been switched, hidden, deleted, etc. And still, it's all about the exit polls and various forms of bias and the poorly worded ambiguous presentations of Mitofsky and ESI and those who want to refute them before they've even been corrected to the point where they are intelligible.

2. The second biggest mistake on both sides, seems to be conflating the issue of whether the EXIT POLLS prove there was vote counting fraud, with the issue of whether there WAS vote counting fraud! I for one am more concerned in the latter than the former.

But to be clear about it, I think both sides can probably agree that there could have been vote counting fraud that was not detectable in the exit polls. In that case, this debate is a waste of time except for those with academic interest in the public opinion research field.

I mean come on fellow fraudsters!: EXIT POLLS??? Is that all ya got?

So let's keep digging and working on proving that there WAS VOTE COUNTING FRAUD, and NOT just whether or not the EXIT POLLS say so. Then maybe we'll get somewhere.

[yowzayowzayowza]

66. If I may...

:applause: :applause: :applause: :applause: :applause:
:applause: :applause: :applause: :applause: :applause:
:applause: :applause: :applause: :applause: :applause:
:applause: :applause: :applause: :popcorn: :applause:
:applause: :applause: :applause: :applause: :applause:

Imho, a ridiculous amount of time, energy and passion have been wasted blaming machines and juggling dubious numbers.

[kster]

67. We also learned that sunshinekathy

just got 40 votes for this thread and is one away from the no 1 spot in the greatest threads.

:applause: :applause: :applause:

:applause: :applause: :applause:

:applause: :applause: :applause:

:applause: :applause: :applause:

:applause: :applause: :applause:

:toast: to sunshinekathy

[nashville_brook]

69. sunshinekathy rocks!

:bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce::bounce:

[sunshinekathy]

71. Our goal is to collect & publicly release ALL election results data

Thank you for the smiles everyone. :-)

If you are obtaining the precinct-level, by vote type detailed election data that will enable us to pinpoint exactly the precincts in Ohio (not just the exit polled precincts) then GREAT.

see this description of what we need people to collect to restore democratic elections in every state:

http://electionarchive.org/ucvAnalysis/US/election_officials/ElectionArchive_advice.pdf

They've basically proven vote tampering (as much as is possible to do so) in Washington state Snohomish county as well using detailed election data.

See this paper's short section on Washington State:
http://electionarchive.org/ucvAnalysis/US/Presidential-Election-2004.pdf

The goal of NEDA has never been to analyze exit polls ad infinitum. In fact its only been necessitated due to folks like Mitofsky, ESI Blumenthal, and others who know who they are, who've put out shoddy analysis and wrong hypotheses on this, who've caused us to have to respond to disprove their work. It has, I agree, wasted a lot of time that should have been spent putting a system into place, like the one we've planned with many people helping in the design, since November, to quickly gather, publicly release and analyze all the detailed election results data immediately after elections, BEFORE any candidates concede so that we get the right ones into office.

To do that we need to build this system that requires high speed dedicated Internet servers w/ lots of disk space - and we need a full-time system administrator and at least one, if not more full-time programmers, and then we need people to coordinate with volunteers and election officials to obtain the data, etc.

If we do this, we can PROGRAM the analysis eventually so that it is done instantaneously. We have the legal right to all the data, although not one of the 3300 separate county and township election officials releases it yet in the detail we need it. Don't you love the way our system totally covers up all evidence of vote tampering now? The election results are conglomerated so evidence of vote padding cancels with evidence of vote subtraction in two different ballot types.

Unfortunately this project cannot be done without funding and not even the mere 250 people that we need to hire one programmer have pledged $10/month. And the almost $500/month that people have pledged altogether so far is going to $250/month in co-location for the 2 servers plus $500/month for part-time system administration and other donations recently have had to go towards $2,000 in legal costs to get past the IRS to obtain our nonprofit status and I've worked unpaid for over 18 months on this issue now and I've had to put hundreds of dollars in each month in addition to supporting myself.

Well, so I'm not organizing the writing of ANY more academic papers until we get some funding - and I'll focus instead on getting a bare bones election document archival up and running - which will be a rudimentary tool if enough volunteers help collect data and others do the analysis by hand until we can get the statistical programming done.

Please go read a few papers on our site about this project.

http://electionarchive.org

Good luck with all your work in Ohio. Here is what we need from you for the national election data archive, but don't do it now as our system has been temporarily disabled due to a security flaw and a sick system administrator:

http://uscountvotes.org/ucvData/US/How2CollectData.pdf

[mom cat]

86. What an incredible job you are doing!

I am broke as hell, but I am going over and pledging. :yourock:

[Bill Bored]

95. Kathy, for better or worse the perception is

that a lot of time is being spent on exit poll analysis.

Now I happen to think that your random auditing paper was excellent, and frankly, we should all work on putting the principle of the hypergeometric distribution into LAW to audit our elections.

There are still lots of folks who think vote margin doesn't matter because e-voting can manipulate the count with no limit. While this is true, it's also true that a random audit of 2% or 3% would detect such a massive swing in the count. So if we have those initial audits, and they are error free, we will only have to worry about CLOSE elections, and you've shown that more auditing can be used to make sure their outcomes are correct as well.

So while the archive may be quite useful, wouldn't it be even better to be able to know, through the appropriate percentage of random audits, who won every race with mathematical certainty?

If there are paper trails in 25 states, we need to audit them. If there are none in the other 25, we need to have them! Why not spend more time on that agenda? I'm not saying you aren't doing so, but the existence of the archive alone will not change election laws. While the archive may have scientific validity, it will not have legal standing.

On the other hand, some elegant mathematical proof that there is no basis for confidence in our election results at this time might raise some eyebrows, even among Republicans. And I bet even OTOH, Febble or even Mitofsky and ESI could sign onto that. (Well, maybe not ESI.) ;)

Just my 2 cents.

[Febble]

76. Absolutely right

actually worse: I honestly believe that the exit polls have been seriously misleading the investigation.

Read rightly, I think they could help, not by proving either that there was fraud or that there was no fraud, but by showing us the kind of fraud that is most and least likely. But it seems you've managed to do that anyway.

Congratulations!

[paineinthearse]

83. .

:kick: as it appears anything >24 hours old - (started Wed Nov-02-05 06:25 AM) falls off the "greatest" page.

[sunshinekathy]

87. Analysis temporarily retracted - call for ESI to likewise retract

On November 2, 2005, NEDA released an analysis of the 2004 precinct level Ohio exit poll data entitled “The Gun is Smoking: Ohio Exit Poll Data Provides Virtually Irrefutable Evidence of Vote Miscount”. The analysis used data provided in the Election Sciences Institute (ESI) report of June 6, 2005.

On November 3, 2005 NEDA realized that its interpretation of the definition for the term “nonresponders” to the exit poll, as used by ESI, was incorrect.

What this means is most likely that the Ohio exit poll data is:

1. inconsistent with voter exit poll response explanations as put forth wrongly by Mitofsky in his Jan 19th paper

2. highly suspicious and very consistent with vote fraud explanations

but does "not" contain "virtually irrefutable" evidence of vote fraud.

In a positive sense, notice that NEDA withdrew its incorrect analysis within 24 hours, whereas ESI and Mitofsky, whose analysis has clearly been proven mathematically to be balderdash, have yet to show the integrity to retract their own analysis, which they have wrongfully let stand since June, and presented in October to the American Statistical Association fall conference in Philadelphia.

Anyone who has used our press release from yesterday, please substitute this press release instead:
http://electionarchive.org/ucvInfo/release/ESI-hypothesis-illogical-press-release.pdf

It is very important to get the word out that the Mitofsky/ESI analyses are bunk.

http://uscountvotes.org/ucvAnalysis/US/exit-polls/ESI/ESI-hypothesis-illogical.pdf

This above math proof still stands as 100% correct (despite Febbles claims to the contrary), has been reviewed by PhD mathematicians and physicists, and the "under construction" in its appendix should be removed from the appendix as we've checked it and found it correct.

I do not know why Febble cannot understand that our counterexample disproves the validity of the hypothesis which ESI and Mitofsky are using to wrongly "rule out" or "kill" the vote fraud hypothesis. There are a countably infinite number of such counterexamples to their hypothesis. And to quibble about the way that we've re-stated the ESI hypothesis (by copying it from their paper and adding a few words to clarify how ESI's analysis applied it) does not change one iota the fact that we've proven mathematically that the entire ESI/Mitofsky analysis since June is poppycock.

I suppose some people are more motivated to find the flaws in others' work than their own. I discipline myself to stay open-minded to both possibilities. Our corrected analysis will take a while to come out because both Ron and I have trips planned.

[Febble]

88. I stay open-minded too, Kathy

as I hope others here will testify. In fact, probing my own cases for holes is what I seem to spend my life doing, sadly. And I do realise that you don't know why I don't understand why your counter-example disproves the ESI hypothesis - and perhaps I can understand that you can only assume I'm simply being arrogant. Or worse.

But the truth is that it's just as hard for me to understand why you don't understand why it doesn't!

Funny old world.

[OnTheOtherHand]

89. Kathy, this is becoming shameful --

I do not know why Febble cannot understand that our counterexample disproves the validity of the hypothesis which ESI and Mitofsky are using to wrongly "rule out" or "kill" the vote fraud hypothesis.

Whom are you quoting when you quote "rule out"? ESI does use that phrase once, but in pretty much the opposite of the way you claim: "To rule out important errors or irregularities within the election system, further analysis is needed."

As for "kill," you are taking an anecdotal report of what Mitofsky apparently said in Philadelphia (I don't remember either way, but I don't dispute it) and imputing the interpretation that he claims to have disproven fraud -- AND, apparently, that ESI has too. Here is something that I am reasonably confident Mitofsky said in Philadelphia, because it was part of his slide presentation, which he sent me: "This is not a discussion about whether there was fraud in the 2004 election. There may or may not have been fraud."

I could go on, but you ignore me anyway.

[sean in iowa]

90. OTOH, that is a misintepretation of Kathy's statement

I'm confident that the term "vote fraud hypothesis" here refers to vote fraud as an explanation of the exit poll discrepancy, which is obviously what ESI and Mitofsky claim to rule out.

Kathy has never stated, to my knowledge, that ESI or Mitofsky claim to have ruled out fraud entirely in the election system. She has maintained that the exit polls alone provide powerful evidence of fraud, and that their analyses fail to rule out fraud as an explanation of the exit poll discrepancy.

I'm sure she can speak for herself, but I wanted to see such an obvious misinterpretation corrected quickly.

[OnTheOtherHand]

91. well, I'm glad it is clear to you

but I have read the NEDA paper repeatedly, and it contains multiple references to 'ruling out vote fraud,' period. Moreover, if you can find any textual evidence that ESI rule out vote fraud even as an explanation of the exit poll discrepancy, I will be much obliged. I think their text clearly contradicts this reading. (As for Mitofsky, we can only sit around speculating about which "fraud hypothesis" he thought was "killed," or whatever he said -- but that can hardly have much bearing on what we should actually conclude from the analyses he presented.)

And I would not say that Kathy has maintained merely that ESI and Mitofsky's analyses "fail to rule out fraud as an explanation" yadda yadda, but rather that they are bunk, poppycock, and I have forgotten what all else. No?

(This "fraud as an explanation of the exit poll discrepancy" formula is pretty woolly, actually. If you think that the exit polls point to fraud in general, are you also obligated to think that they point to fraud in New York and Vermont?)

[sean in iowa]

93. "...if you can find any textual evidence..." - it's in ESI's 4-page

report.

Here is a link to the ESI report. The textual evidence is a quote from page 4, which quote is also displayed in large blue letters:

Our study indicates that the non-reponse rate theory is much more likely than the fraud accusation theory to account for most, if not all, of the observed discrepancies between the exit polls and the actual results

My understanding is that when social scientists say that x is "much more likely" than y in a paper's conclusion, they are for all practical purposes, ruling out y.

As Febble wrote in an earlier thread (reply #66 to "New Math PROOF..." on 10-31-05):

I do so {admit that the ESI and Mitofsky analyses don't "rule out" fraud} rather in the spirit of one who says "no, I can't rule out alien abduction as a reason for my missing car keys, but the more likely explanation is that I yet again left them in the garage door".

As for the NEDA paper containing references to to "ruling out vote fraud," I think that in a paper entitled "The Gun Is Smoking: Ohio Precinct-Level Exit Poll Data Show Virtually Irrefutable Evidence of Vote Miscount," it's obvious that their arguments are in the context of the argument over whether the exit polls are or are not smoking-gun evidence of fraud. "Exits not a smoking gun" was after all, the phrase ESI used to describe its report (Link).

[Febble]

97. Sean, as you will know

the sentence of mine you cite was followed by:

"I exaggerate slightly. And truly, I am working on this algorithm. But either I'd make a poor vote-switch hacker or it's damn difficult."

And I should have added "new, voteswitching" before the word "fraud". Sorry. I try to remember.

I don't think the analyses do anything to rule out many kinds of electoral injustice (I am not entirely happy about calling all of these fraud at this stage but they may have been) that we know happened in Ohio and elsewhere and which could have cost Kerry the election.

In short, I don't think the exit polls tell us much about fraud either way, except, as I said on my calculatus elimatus thread, that they put useful constraints on what could or could not have been done.

[sean in iowa]

100. We are talking about "ruling out"

fraud as an explanation of the exit poll discrepancy. Which the ESI analysis clearly purports to do, for all pratcical purposes. The textual evidence OTOH asked for has been provided.

Now to my quoting you to drive home the point that when social scientists conclude a paper by saying x is "much more likely" than y is, for practical purposes, ruling out y. You admitted to exaggerating "slightly" by saying that you view vote fraud as a less likely explanation of the exit discrepancy in the same way that you view aliens stealing your car keys as a less likely ecplanation of your lost car keys. And that you are looking at the problem to be sure of things, but my gosh, it looks pretty grim for the hypothesis that fraud explains the exit discrepancy.

If the exit polls "don't tell us much about fraud either way" then the hypothesis that they offer evidence of fraud has been ruled out.

[Febble]

103. This is getting very curly

and contrary to Kathy's assertion, my mind is never fixed. If I find evidence that seems to support one conclusion, my confidence in that interpretation tends to become gradually eroded as I consider alternative interpretations that are consistent with the evidence.

I agree that the way you have described ESI's conclusion is probably accurate, and I tend to agree with that conclusion, not just on the basis of ESI's study, which was on a single state, and thus had a fairly poor signal-to-noise ratio, but because of what is, in effect, a replication of that finding by Mitofsky on the 1250 precincts across the nation.

I think I can with confidence state that my own interpretation of this finding is that fraud cannot account for the whole discrepancy. I also believe it cannot account for more than a tiny fraction of it unless we postulate that it was deliberately targetted in a way that was a function of Bush's likely performance. In other words if its distribution was not orthogonal to Bush's "true" performance (I am choosing my words very carefully here, and avoiding the word "random") perhaps it could account for rather more of it. However, my own attempts at producing a fraud algorithm that would produce a major benefit to Bush and yet not a positive correlation between Bush's performance and redshift have so far not been successful - and as I say, I have been trying hard. Which is why I dropped into aliens-stole-my-car-keys mode (actually I am convinced that aliens regularly move my car keys.)

But that may simply be because I am a crap fraudster.

[OnTheOtherHand]

102. no, I don't think that is the case

Speaking as a card-carrying social scientist -- well, OK, I don't know where the card is -- I don't think it is reasonable to interpret "indicates... much more likely... most, if not all..." as, "for all practical purposes, ruling out" anything. That's not what the words mean, and that's not how social science works. (If ESI intended to, for all practical purposes, rule out fraud, then the "not a smoking gun" language of the press release would be, at best, a strange euphemism.)

Frankly, it doesn't matter much what you or I think these words mean, but if I were you, I would not hold my breath waiting for dozens of survey researchers and statisticians to put pressure on ESI to recant their "bunk."

Febble's current assessment of the fraud debate has no bearing on how we should interpret an ESI brief about Ohio from five months ago. (And the fact that even Febble got sucked into language about 'ruling out fraud' underscores, to my mind, how NEDA's analytical frame warps the entire discussion.) Moreover, even if one thinks that ESI's conclusions (never mind Febble's) go beyond the evidence, that isn't grounds for trashing the entire analysis. If social scientists operated in that manner, no one would venture conclusions at all. We lay out evidence, we try to say as clearly and honestly as we can where we think the evidence points, and then we wait to see if someone else can do better.

We started out talking about the smoking-gun paper, but we aren't now -- at least, I'm not. Kathy's reference to a counterexample refers to the "mathematical proof" paper. In any case, there isn't much logical relationship between an argument that the exit polls aren't smoking gun evidence against fraud (which seems to be the, fairly trivial, argument of the NEDA mathematical proof paper) and one that the exit polls are smoking gun evidence for fraud (which seems to have been the argument of a NEDA paper now withdrawn -- we'll see what it says next time around).

[sean in iowa]

105. OK

As for ESI's "much more likely" language purporting, for practical purposes, to rule out fraud as an explanation of the exit poll discrepancy, I will once again drag Febble into the mire, reply #103 of this thread:

I agree that the way you have described ESI's conclusion is probably accurate, and I tend to agree with that conclusion, not just on the basis of ESI's study, which was on a single state, and thus had a fairly poor signal-to-noise ratio, but because of what is, in effect, a replication of that finding by Mitofsky on the 1250 precincts across the nation.

And just to be sure we're on the same page, here is my post (reply #100) the quoted statement replied to:

We are talking about "ruling out"
fraud as an explanation of the exit poll discrepancy. Which the ESI analysis clearly purports to do, for all pratcical {sic} purposes. The textual evidence OTOH asked for has been provided.

Now to my quoting you to drive home the point that when social scientists conclude a paper by saying x is "much more likely" than y is, for practical purposes, ruling out y. You admitted to exaggerating "slightly" by saying that you view vote fraud as a less likely explanation of the exit discrepancy in the same way that you view aliens stealing your car keys as a less likely ecplanation of your lost car keys. And that you are looking at the problem to be sure of things, but my gosh, it looks pretty grim for the hypothesis that fraud explains the exit discrepancy.

If the exit polls "don't tell us much about fraud either way" then the hypothesis that they offer evidence of fraud has been ruled out.

It would appear not all card-carrying social scientists agree about the practical effect of the language used in their conclusions. And, without snark, when do social scientists ever write a paper conclusion saying explicitly that they have "ruled out" anything? Even natural scientists are, I think, quite cautious about writing "rule out" in a paper, even with much finer empirical measures. The phrase "much more likely" is, I think, (and do correct correct me if I'm wrong), about as confident as the language of social-science papers gets.

No, I am not holding my breath waiting for other statisticians and survey researchers to call on ESI to retract their "bunk." I do hope that other such scholars who have no previous investment in this debate can be recruited to take a hard look at the redshift-swing hypothesis. I'm sure we all do.

As for your final paragraph, the NEDA mathematical proof attempts only to vitiate the ESI hypothesis that if fraud is an explanation of the exit poll discrepancy, we should expect a redshift-swing correlation. The NEDA paper (which we will hopefully see again soon) goes to the second argument you state in your final paragraph, that the exits are in fact smoking-gun evidence of fraud. So the logical chasm you describe does not exist. NEDA handles the two arguments separately.

[OnTheOtherHand]

106. fair enough -- point by point --

It does seem that Febble and I somewhat disagree about this, and to some extent I think that her current views are bleeding into her interpretation of ESI's intent back in May and June.

I am also starting to think that there is something inherently confusing about the "rule out vote fraud as a cause of the... exit poll discrepancies" language that has us chasing each other in circles.

I agree that social scientists tend to be cautious about their causal claims. Whether ESI's language is "about as confident as the language of social-science papers gets," well, I really don't think so, no, and even if it is, I don't see how that is any defense of the NEDA paper. I think if ESI had intended to rule something out, they would have. Even social scientists do sometimes manage simple declarative sentences. And I don't understand why, if social scientists manage to be cautious about their language, that would justify their critics in imputing ideas they deliberately didn't express.

And there is a fair amount of that. At least as of the 10/31 version (that was the third, I believe -- I haven't tried to keep up since then), one could learn that "Warren Mitofsky has been particularly active in defending the accuracy of the vote counts." Mitofsky is an exit pollster, and he talks about exit polls. I think it's a fair guess that he doesn't think Bush stole 10 million votes, but what we know for sure is that he doesn't think his exit polls point to that conclusion. To say that he defends the accuracy of the vote counts is at best a freehand inference.

OK, back to ESI. I think they might, very freely, be said to 'rule out' fraud as the cause of the discrepancies. It would be insane to rule out fraud as a cause, the view that NEDA attributes to ESI. (One can't even rule out fraud in Cincinnati 4M, but one can observe that the exit poll there wasn't likely to be very accurate regardless.)

I'm not aware that anyone at ESI (or even Mitofsky) had a particular axe to grind about how the Ohio analysis should come out -- but yes, I hope that other folks will think hard about this result and what it does and doesn't mean. It's kind of neat that we live in a time when we can quickly lay out ideas for millions of people to take a whack at, if they like.

We are completely talking past each other WRT the last paragraph, I think. You wrote, "As for the NEDA paper containing references to to 'ruling out vote fraud,' I think that in a paper entitled 'The Gun Is Smoking: Ohio Precinct-Level Exit Poll Data Show Virtually Irrefutable Evidence of Vote Miscount,'...." They write about ruling out vote fraud in the other paper. That was my point. It's not a logical chasm, just a point of fact. (They may write about "ruling out" in the smoking-gun paper too, but if so, it isn't a regular riff.)

It's true, however, that I think that pointing out that fraud remains logically possible despite the ESI analysis is a far cry from proving that the ESI analysis is "bunk" and should be withdrawn. A correct conclusion would be somewhere in between, but in my opinion pretty far from the "bunk" end of the continuum.

[sunshinekathy]

92. They misrepresent more than not

DU discussions on this issue have been basically destroyed by their twisting the facts IMO. It wastes time and makes the conversation unproductive to always have to point out the plethora of misrepresentations and truth twisting. Conversation on ProgressiveIndependent.com seems to be less fettered by such time-wasting unproductive distortions.

The post that you just responded to distorted not just what I said, but also what ESI and Mitofsky have said. I don't think he bothers to find out any facts at all before he opens his mouth. He obviously did not read the ESI paper or the reporting of what Mitofsky said at the Oct ASA debate.

When people's positions are not supported by the facts, and they refuse to give up those positions, distorting the truth seems to be all that is left to them.

[Febble]

96. Not only is OTOH's position supported by facts

he actually cites his sources.

Actually, it is rather a waste of my time to be continually "pointing out" your own omissions and interpolations. If you want to post a "Math Logic Proof" that appears to demonstrate that a hypothetical counter-example to in validates a statistical test of the probability of that counter-example occurring by chance, then fine.

But don't expect those who proposed and conducted that test, and (mis)quoted in your paper to sit idly by without "pointing out" the absurdity of your case.

Give me the odds of your counter example occurring by chance, and I'll take you seriously. Otherwise I'll stick with Mitofsky's and ESI's estimate.

Or alternatively I will regard their estimate as a useful indicator of the kind of deliberate fraud targetting that might have occurred. As I have been doing and will continue to do.

You were right to withdraw Ron's paper when it was shown to be in error. But you simply have not shown that Mitofsky's or ESI's "hypotheses" were in error, because the method you have adopted is entirely inappropriate to the nature of those hypotheses, and appears to betray a fundamental mis understanding of the nature of parametric statistics.

[glitch]

94. Very good. I consider it a sign of professionalism to admit mistakes.

Too many people in the business world don't have the confidence.

:applause:

[Febble]

98. And I couldn't say this better myself:

http://progressiveindependent.com/dc/dcboard.php?az=show_mesg&forum=120&topic_id=919&mesg_id=924|Argh. There seems to be no getting through to people who are emotionally attached to their own positions irrespective of the evidence!>

[yowzayowzayowza]

99. Corrected ...

link.

[Febble]

101. Thanks! I stand corrected!

I am fully prepared to admit that my link was faulty.

[OnTheOtherHand]

104. oh, just quit with the obfuscation, already! (grin) n/t

[sunshinekathy]

107. ESI's analysis has been proven illogical and useless

Quote from ESI's paper:

"We conclude that the data do not support accusations of election fraud in the Ohio presidential election of 2004."

end quote.

Quote from Mitofsky:

"This kills the vote fraud argument"

end quote.

Please read this proof that the ESI and Mitofsky hypothesis is invalid.

http://electionarchive.org/ucvAnalysis/US/exit-polls/ESI/ESI-hypothesis-illogical.pdf

Vote fraud can have ANY relationship with the variables that ESI analyzes, so they can conclude nothing at all about vote fraud one way or the other like they and Mitofsky have both done, or anything about the causes of exit poll disparities from their analysis. i.e. ESI's analytical methods are useless and meaningless when it comes to determining anything meaningful.

Anyone with any logic skills can easily see that the ESI/Mitofsky analysis is illogical bunk after reading through the math logic proof.

I recommend that anyone take the proof, along with either the Mitofsky or ESI analysis to any math department at a university and ask the department chair to have someone in his department who is qualified in math logic to verify that, if you don't believe me.

However, most people, even without any math logic training will be able to understand, by reading the counterexample in NEDA's proof and ESI's or Mitofsky's analysis, that the ESI/Mitofsky analysis is pure bunk -- that is if they tend to be logical.

If in doubt, though, please take it to any math department.

[Febble]

108. You need to make a clear distinction:

I would maintain that ESI and Mitofsky have said that their evidence does not support/"kills" the case that fraud was responsible for the exit poll discrepancy. You may disagree with whether their evidence does so. That is fine.

But I would also maintain that neither ESI nor Mitofsky have said that their evidence says anything about whether fraud occurred, many types of which would not have shown up in the exit poll. What both ESI and Mitofsky are considering is the narrower question:

Does the exit poll evidence support the case for fraud?


You quote ESI as saying that:

"We conclude that the data do not support accusations of election fraud in the Ohio presidential election of 2004."

What data were they looking at? Exit poll data.

And at the end of their report the authors say:

Our study indicates that the non-response rate theory is much more likely than the fraud accusation theory to account for most, if not all,of the observed discrepancy between the exit polls and the actual results. The more detailed information allowed us to see that voting patterns were consistent with past results and consistent with exit poll results across precincts.

-snip-

To rule out important errors or irregularities within the election system, further analysis is needed. In fact, the election records themselves need to be examined, and ESI has been engaged in that effort as well.

(my bold)


Thus it is perfectly clear that the data they are referring to is the exit poll data. They leave wide open the possibility that otherdata, e.g. the election records, need to be examined before "important errors or irregularities" can be "ruled out". So far from the exit polls "ruling out" fraud, ESI specifically argue that it does not "rule out" fraud - that other data need to be examined before fraud can be "ruled out". I'd argue therefore, that the actual premise of your paper, expressed in the mis-spelled title "Mathematical Proof that Election Sciences* Institute's Test to Rule Out Vote Fraud Is Logically Incorrect", is, er, "balderdash".


You also quote Mitofsky as saying "this kills the fraud argument" - but fail to cite source or context.

I believe the context it was the Freeman Mitofsky debate (I have no idea of your source - were you there?) In that context it is clear that "the argument" that is "killed" is the argument that fraud was the cause of the exit poll discrepancy. I happen to know that one of his slides actually said:

This is not a discussion about whether there was fraud in the 2004 election. There may or may not have been fraud.

This is a discussion about the exit polls and whether they shed any light on the claims of fraud.

(My bold)

Doesn't get much clearer than that.

So your paper at the very least needs to acknowledge, as at present its very title does not, that the hypothesis you are attempting to deBUNK concerns the relationship between the exit poll data and fraud, not the existence of fraud per se.

However, once you make that distinction, your logic largely falls apart.


* The name of the organisation is Election Science, not Election Sciences. I also note that you still have not corrected your citation of me in what is now footnote 5, nor provided source or link, despite repeated requests that you do so. Failure to correct verifiable citation errors when they have been pointed out does not strike me as being evidence of the "due diligence" you demand from others.