Freeman presentation with slides From Mitofsky Debate is Here

[Melissa G]

at this link...http://www.appliedresearch.us/sf/epdiscrep.htm
Check it out all you Math and Stat Fans and Post your opinions Here!

[garybeck]

1. thanks, kick!!!

[CalmMan]

24. Kick?

Sorry, I'm new to this site. What does "Kick" mean?

[Melissa G]

65. Kick is shorthand and means the user wants to keep the post in

front of DUers and add some energy but without a lot of comment..
Sometimes they just use the donkey...
Welcome to DU Calm Man!:kick: :hi:

[kster]

2. KICK-N-RECOMMEND..NT

[JohnGideon]

3. Recommend and kick n/t

[Mnemosyne]

4. .

:kick:

[Amaryllis]

5. kick

[Mr Grieves]

6. great read!

thanks.

[Melissa G]

8. Hi Mr Grieves... Welcome to DU!

You are right! It is a great read!:hi:

[Chi]

7. Thanks for posting Melissa...

Nominated

[Melissa G]

9. Hi Chi, thanks for the K and N!

:toast:

[Febble]

10. OK, here's a response

it's a bit long, and it's taken me a while, but FWIW:

First the good news:

I do think Steve makes a good case for investigation. I agree with him that Ohio stinks, voter suppression is rampant, whether structural or intentional (and it is hard not to believe intentional), and that the exit poll discrepancy cannot be due to chance. Given the first two, it is reasonable to suspect that the exit poll discrepancy may have been due to fraud.

And I like his term "Precinct Level Discrepancy"(PLD) “Within Precinct Error” (WPE) has become synonymous with a very particular measure of PLD, which I think Steve agrees is not the best, although inevitably it is the one that will feature in his analyses as it is the one used in the E-M report. And it is important as it appears that precinct level was where the red-shift occurred.

So, thus far I’m with Steve. The case for investigation is good, and PLD is where the bodies appear to be buried.

Steve also presents some interesting observations. Although state level aggregates are difficult to interpret, it is of interest that completion rates show a bit of a tendency to be higher in redder states, and that PLD appears to be higher in states with higher African American populations. (I identified a possibly related pattern which is that WPE was higher in bluer states). State level is too coarse a grain with which to investigate potential causal factors in these relationships, but they are, nonetheless, interesting, and worthy of further investigation.

Steve also says in one slide that the only two causes of PLD are "non-response bias" and "count corruption". I agree with this as long as you take a fairly broad definition of "non-response bias" to include actual sampling bias i.e. people who do not respond because they are not even approached. There is evidence that the actual sampling may have been biased (PLD was greater where sampling rate was low). This point is important in interpreting some of his other observations.

Because here is where Steve and I start to part company.

He then goes on to say that interviewer effects cannot account for all the discrepancy - although he then says that it cannot be ascertained without looking at the data. This latter thing is true. You need a multiple regression model to determine whether interviewer effects can or can't account for all the PLD. But from what is publicly available we do not know whether it can or can't. And it is also true that some polling factors may be collinear with fraud. So although I accept that with the data publicly available we cannot be assured that interviewer effects account for all the discrepancy – without the data we can also not be certain that they do not. There is a difference between saying that “interviewer effects cannot account for all the discrepancy” and saying that “on the data given, we cannot know whether effects account for all the discrepancy”.

The completion rate by precinct partisanship argument is of interest, but IMO not at all conclusive. At the risk of sounding like an apologist for the exit polls, there are many reasons why the completion rate patterns are perfectly consistent with non-response bias as an explanation for the exit poll discrepancy (particularly if this is taken to include sampling bias). Firstly we know, from a scatterplot presented by Mitofsky at AAPOR (and again at this debate) that variance in completion rates was vast, and neither negatively nor positively correlated with Bush’s share of the vote. This means that staring at aggregate values tells you very little about whether those aggregates are meaningfully different. Secondly, total completion rate may vary independently from differential completion rate. For example, voters of from both groups may be more likely to respond in, say suburban than urban precincts; however, for a given total completion rate, there may nonetheless be a tendency for Bush voters to respond less. Thirdly, non-response bias may operate at the level of voter selection rather than voter refusal. If Bush voters were less likely to be selected, but equally likely to participate IF selected, completion rate would not vary by precinct partisanship; red-shift would nonetheless be manifest in the PLD. And the fact that PLD was greater where interviewing rate was lower (and therefore where there was more opportunity for non-random sampling) strongly suggests that bias in voter selection was a factor in producing redshift.

Steve presents a plot in support of his claim of “implausible” Kerry response rates which is simply wrong. I believe it was produced by Ron Baiman for UScounts votes, and plots a calculated value for Kerry and Bush response rates based on a formula that can be readily shown to give “impossible” values for response rates for voters for the minority candidate under conditions of sampling error alone. A full rebuttal of this plot is way to geeky for this post, but if anyone’s interested I can demonstrate that Ron’s formula can easily generate computed “response rates” for voters for one candidate that exceed 100%, even if the only source of error is sampling error. And it is not possible to conduct a poll without sampling error.

And now we come to the apparent rise in PLD in “high Bush” precincts. This is, I believe, an artefact of the WPE as a measure, and of the instability of aggregate measures where Ns are relatively small. I believe that Steve agrees with me that the measure I proposed, and called ln(alpha) is a better measure. However, even when WPE is used as the measure of PLD, Steve’s statement that “there’s no PLD at all in the Kerry strongholds” is simply untrue – the mean PLD may be zero in the Kerry strongholds, but there is plenty of it. Moreover if ln(alpha) is used as the measure, it is clear that the only sense in which PLD is”lower” in Kerry strongholds than in Bush strongholds is that in the high Bush category (incidentally, not “quintile” – which would imply equal numbers in each category) is that there is a shortage of extremely blue shifted precincts – and in fact one very blue-shifted precinct only escapes the arbitrary “high Bush” category line by a single percent. And in fact, as Mitofsky demonstrated at the Miami AAPOR meeting in May, there is no significant tendency for PLD to be higher at the Bush end of the spectrum than at the Kerry end. The correlation coefficient between ln(alpha) and Bush’s vote-share is insignificantly different from zero.

A few more miscellaneous comments:

The swing state thing is interesting. Could be evidence of fraud. Could be a spatial manifestation of the temporal phenomenon suggested in the E-M report whereby red-shift appears to be greater in years in which interest in the election is high.

Voting technology. I simply don’t think anything can be drawn inferred from this. As Steve draws attention to, the number of urban precincts with paper ballots is absurdly small – too small to do any statistics on. As with many of these analyses, it’s kind of interesting, but you need to know what is collinear with what else to draw any conclusions.

Reported Vote: Steve seems to have cottoned on to TIA’s case. My response is the same as my response to TIA’s. I won’t repeat it here, except to say that it strikes me as being in no sense a slam dunk for fraud, despite popular belief.

Confidentiality. I simply disagree with Steve that the data should be publicly released. Quite apart from being a violation of the ethical guidelines of the professional organisation of pollsters, it is, simply, unethical to make data public in a form in which respondents could be identified – and here we are talking about the sanctity of the secret ballot. And I also believe that E-M owe their interviewers (casual employees) a duty of care. If precincts can be identified, so could interviewers. The demographics of both responders and interviewers are important variables. It is simply wrong to allow these variables to be released in a way that make either identifiable. “Blurred” data on Ohio was released. Perhaps other sets could be similarly prepared and released. But confidentiality matters. I would be dismissed from my own job if I published data in a form in which my respondents could be identified. You just can’t do it.

So to summarise my response to Steve’s talk, for anyone who is interested, and doesn’t think I’m a freeper:

Freeman and Mitofsky agree that there are only two possible causes of the Great Exit Poll Discrepancy: Fraud and Non-response Bias (and I’ll qualify that by including selection bias in Non-response bias).

So whodunnit? Fraud or Non-response Bias:

Steve, as a good prosecuting counsel builds a nice case for Fraud as the killer of the Exit Poll. His evidence that a crime was committed is sound. His evidence that Fraud was the culprit is suggestive. Some is flawed. Some is wrong. But regardless of which is which, it is built into a solid-looking edifice. Things look bad for Fraud.

But even solid looking cases can come apart with a watertight alibi. And I think Mitofsky as defence counsel presented the alibi for Fraud.

Here it is:

If Fraud was responsible for the Great Exit Poll Discrepancy, then where the PLD is greatest, the boost to Bush’s vote should be greatest. After all, the point of fraud is to win the election. Cui bono?

If Non-response Bias was responsible for the Great Exit Poll Discrepancy, PLD should vary independently of Bush’s advantage in the vote count.

Mitofsky correlated a measure of PLD with a measure of advantage to Bush, namely the increase in Bush’s share of the vote relative to 2000 (a year in which mean PLD was near zero).

And there was no relationship. PLD is not correlated with boost to Bush’s vote share.

So, carefully as Steve’s case was built, I simply don’t think it stands.

Of course Fraud could still have stolen the election. But it looks to me that with regard to the Great Exit Poll Discrepancy, he’s innocent.



(edited for grammar and typo - I'm sure there are more)

[kster]

11. But this report coupled

with the fact that they barred our side from the attending the Baker/Carter commission is only strengthening Freeman's side.

"The lack of a fair and open process like that used by the Carter-Ford Commission was evident throughout. In the last commission, civil rights groups submitted research, reports, and testimony. This time around, civil rights groups were essentially barred from the process. The only input from the civil rights community (Barbara Arnwine from the Lawyers' Committee for Civil Rights Under Law) was essentially ignored on this critical issue. It speaks volumes that the public could not participate in a process that would effect their most fundamental right, the right to vote. Moreover, the Commissioners spent only a short time deliberating on these issues"

http://www.commondreams.org/news2005/0919-04.htm

[Melissa G]

13. Hi Febble, Thanks for the post. You are right it is long so it 's hard for

a poor typist like myself to address all your points but i'll take a stab at a couple. I agree that PLD is a better term but to my cursory read i'm not sure that he necessarily says that is where the bodies are buried. That seems to be your redirect.
I see statewide references in several of his slides so i disagree on the significance of PLD or WPE or FFF (Febble's Fancy Function) as enough to prop IMHO Mitofsky's weak assertions up.

FFF does not strike me as a watertight alibi by any means...I think as a theory it still fails the common sense likely to have happened test.

Let's look at what seems to me to be plausible as a possible theory...

Rove is a political strategist. When he stole 2000 the way he set about stealing was a statewide strategy. He admits this strategy if not the theft in a number of quotes.
He took a lot of heat for losing the popular vote so the Number 1 priority still had to be winning the electoral college but concurrently he HAD, unquestionably HAD, to win a decisive majority to continue to govern. He could not afford a second supreme court delivery of the white house and hope to get any program through. In addition, he was not going to be able to take any chance of having his statewide strategy have scrutiny so he had to win in excess of amounts that would trigger automatic recounts. He would also need to win in so many different tricky ways that if one of them was discovered, there would not be enough to put together proof that could show the election would be overturned by that action. There was a Florida example thread on this that I am too tired to link to but i'll find this later if you want it.

That is my personal theory of election Fraud and Freeman's data makes sense to me in my worldview. I have trouble with your assertions and finding the extra 10 million voters you seem to be comfortable assuming exist because of a IMHO an implausible function says they can. I also am uneasy because you have no theory of where these folks came from that I have been able to discern.. but maybe I'm just too tired when I'm reading and you can explain this to me.
Best,
Melissa

[Febble]

16. OK

Maybe it's a redirect, but I don't think Freeman would disagree*. The early projections that appeared to indicate a Kerry win were not due to faulty weighting or unrepresentative precinct selection - we know from the E-M report that the discrepancy was at precinct level. Freeman uses statewide data based on mean WPE (PLD). He may use other data. I didn't say that using PLD "propped Mitofsky's" assertions up. And I didn't say the FFF was an alibi, although it does undermine one of Freeman's points - that PLD was higher in more pro-Bush precincts.

What I described (metaphorically!) as an "alibi" was the finding that PLD didn't correlate with Bush's swing from 2000.

As for the rest - OK, if you think that other circumstances make Freeman likely to right, fair enough. But you'd still have to account for the non-correlation between PLD and swing.

But I agree, it doesn't rule out election-winning fraud. I do think it largely rules out popular-vote winning fraud.


*I sent a link to my response to Freeman, so I'll hope he'll point out if I've misunderstood. But I don't think there is any disagreement between Steve and me over the level at which fraud is indicated (if it is indicated!) And I think Steve agrees that the FFF is more appropriate measure of PLD than WPE (well he did the last time I talked to him), but of course mostly WPE is what he's got to deal with as that is what is given in the E-M report.

[Bill Bored]

14. I don't think I buy this one Feb:

"If Fraud was responsible for the Great Exit Poll Discrepancy, then where the PLD is greatest, the boost to Bush’s vote should be greatest. After all, the point of fraud is to win the election. Cui bono?"

This presupposes that the fraud was widespread or even universal. No one is claiming this AFAIK and no one has said this was necessary to swing the election, esp. the electoral college.

With double digit PLDs all over the place in both directions, varying by percentages that would dwarf those necessary to reverse the election outcome, how can one expect to see precincts where fraud occurred correlated with higher PLDs than those with accurate vote counts? The absolute values of the PLDs are close to an order of magnitude greater than those needed to swing the election. While this is averaged down to a "mere" 6.5% in the aggregate, at the precinct level, which is where you're looking, the PLDs are huge!

That said, there are in fact a number of precincts where the relationship that you describe is true i.e., there were Red Shifts in the vote count AND Bush did better than in 2000 in the vote count. They comprise 8% of the Ohio sample alone, and some percentage nationally. So why not INVESTIGATE them instead of expecting to see some statistically perfect scatter plot correlation and summarily dismissing the notion that fraud occurred? (I'm not saying you're doing this Febble, but Mitofsky and ESI sure are.)

If we had a well audited transparent election system (ESI said that Ohio's was "NOT auditable") then the presumption of fraud as a last resort would make sense, but since our system is largely unaudited and opaque, those who dispel the notion of fraud are in effect harboring those who would further privatize our elections and steal our votes. And we should make NO DISTINCTION BETWEEN THOSE WHO WOULD STEAL ELECTIONS AND THOSE WHO HARBOR THEM, should we?

[Febble]

15. hang on

It doesn't presuppose that.

"If Fraud was responsible...etc" what I said would be true. I didn't say

"If there was any Fraud...." etc.

And I don't think any one is expecting to see a "statistically perfect" scatterplot - cite, please!

As long as the exit polls are not used to make the argument that fraud was the size of the discrepancy (i.e. that Kerry won with the margin indicated by the "unadjusted" polls) fine.

And if you mean by your last paragraph that anyone (me?) who "dispels" the notion of fraud is, in effect, stealing your votes - well, I disagree.

Edit: I did say, if you get as far as the end of my too-long response (didn't have time to write a shorter one), that Fraud could still have stolen the election. I agree that there is room in that noise for the degree of fraud required to swing Ohio, and that would have been enough. And especially there is room if some of that fraud is voter suppression which is clearly illegal, but plausibly deniable, probably. But there are plenty of people, I think Steve included, who interpret the size of the exit poll discrepancy as the size of the theft. I don't think this holds water in the light of that non-correlation.

[Melissa G]

17. Okay Febble, I'm hearing your Edit sink your water tight alibi against

fraud...
Are you saying because there are red-shifted precincts in which Bush did not do better than 2000, that there was no fraud? There was also the reverse. I hear your edit showing you agree that there is room in the red shift for the ten million vote shift.

Your previous assertion did not make sense to me as an argument.. That would be saying if i hear you right that because they did not shift all the precincts they did not shift some of the precincts... That would strike me as silly...

Do I understand you correctly?

edit for last question

[Febble]

21. No, I am not saying that

and I don't really understand what you are saying I am saying. This is what I am saying,and I will try to phrase it more clearly:

The zero correlation between redshift and swing to Bush means that redshift is not reflected in increase in Bush's vote share.

Overall. There could have been fraud in some precincts resulting in an unearned increase in vote-share for Bush. But if so, this must have been balanced by something pushing the other way. And as Steve says there are only two conceivable pushers - fraud and non-response bias - then either the thing pushing the other way was pro-Kerry fraud, or there must have been some kind of "rBr" in precincts where there was no fraud. Or some complicated combination. (And of course, if you need to invoke rBr to make the fraud argument we are in somewhat ironic territory) But the point is that the net effect of fraud on the exit poll appears to have been zero.

Well, let me be more specific: the R squared for the correlation is actually given on Mitofsky's slide (0.0000873), and the correlation is actually (insignificantly) negative so we can compute the correlation coefficient: -.0.0093 (square root of the r squared, negative). And we know there were 1250 precincts in the sample, so you can compute the confidence interval for the relationship. I make the 95% upper confidence limit for the correlation coefficient= .046, and if I'm really generous and use the 99% confidence limit I make it .063. That means that there is a 1 in a hundred chance that the true correlation is .063. Which is a very small effect size.

And that is the net effect. If fraud occurred it is more likely to have a lot in some rather than a little in all. Doesn't make any difference to the average, just to the variance, which is already in the equation.

Eomer has found a potential loophole, though. If the fraud was designed so that Bush's proportion of the vote was not allowed to fall below his 2000 proportion, maybe that would beat the system. If so, the distribution of the swing should be truncated at the lower end and thus positively skewed. This is a testable hypothesis. Mitofsky apparently offered to test any hypothesis Steve suggested. This might be worth doing.

(Edit: had a go at computing the maximum loss in votes, and thought better of it! I'll post back if I figure how to do it. Less than 100 million though!)

[Febble]

26. Sorry typo

Meant less than 10 million. I'm getting tired.

[Bill Bored]

37. Fair enough

First of all, the vote swing would not have to have been anywhere near the magnitude of the exit poll discrepancy to have changed the outcome of the election. A popular vote shift of about 1.5% would have been enough to do that. And yet we are debating the cause of a 6.5% exit poll discrepancy!

Citation: Mitofsky said that because there's no correlation between Red Shift and Bush's improved vote count in the polled precincts in 2004, that this "kills the fraud argument."

Now, did anyone SAY that such a correlation must exist for there to have been fraud (and I mean specifically counting fraud), or could there be other less-correlated mechanisms?

I never cited Freeman and I specifically exempted you Febble from my harboring election fraud perpetrators because you have done some good work to help find them. So don't take my rhetoric personally -- I am not TIA and it is not directed at you.

But as I said to you and Mark privately, Mitofsky and Scheuren at the very least need to take some Sensitivity Training! And I will stand by that remark. Here's why:

When they say things like "this kills the fraud argument" or "On last observation, even though we have found errors in the Ohio vote, we did not find systematic fraud and, in any case, the errors were not enough to change the declared winner." (emphasis added), this is rhetoric too.

If the argument is about exit polls, and they are saying they don't prove fraud, why not just leave it at that instead of dragging the actual unverified vote count into the conversation?

Instead Mitofsky seems to be saying that BECAUSE the exit polls don't prove fraud, there was no fraud. Otherwise, why isn't HE calling for more investigation, as both you and Freeman clearly have?

You see, it's not just about the problem, it's about the solution!

Mitofsky offers none: Everything's fine; the exit polls suck; they don't prove fraud; and my precinct selection was impeccable by the way; so can we move on because there's nothing else we can or should be doing except to hire better interviewers and somehow keep our servers from crashing on Election Night.

Freeman on the other hand is calling for more investigation and I think this is entirely appropriate, whether it's based solely on the exit poll discrepancy or not.

[Febble]

38. Points taken

I'd defend Mitofsky more than Scheuren though, because his interest is in what caused the exit poll discrepancy, not whether there was fraud (and whatever people think, the exit polls were not designed to detect electoral fraud, they were designed to project the winner, which they did - whether the winner should have been disqualified for cheating is a separate matter).

Whereas the ESI brief was to investigate the probity of the election, not the cause of PLD.

So I would pars Mitofsky's comment thus:

"The plot kills the argument that the exit poll discrepancy was due to fraud." And I pretty well agree with him, apart from some possible unfinished business with Eomer. So perhaps I should rephrase: it kills the argument that the exit poll discrepancy was due to fraud unless the fraud was cleverly designed only to prevent Bush's vote-share dropping below 2000 levels, and/or it occurred in virtually the same magnitude in virtually every precinct, including those in which the PLD was blue-shifted.

Frankly, Bill, I think Mitofsky and you are in agreement. That the exit poll discrepancy does not index fraud. It is me going out on a limb to say that as I read it at present, that plot actually argues against "massive" fraud - i.e. fraud on the kind of scale that Freeman and TIA have argued for - and indeed fraud on a scale to steal the popular vote.

But I do agree with you that the ESI finding does not rule out fraud in Ohio, although I think, as with the same finding writ large, that it puts fairly strict constraints on the kind and concentration of fraud that could have plausibly occurred. On the other hand Ohio was sufficiently close that I accept that it is a climbable barrier.

[Bill Bored]

47. OK, so you are refuting the TIA/Freeman 10,000,000 vote thing.

But to swing the popular vote, which is legally meaningless anyway, it would only take about 1.5 million votes. Now it's true that by insisting that the swing was 10-million, they are leaving themselves open to certain counter arguments.

But I'll settle for proving the 1.5 million if it can be done, and for that matter, just proving that there is no way to prove that wasn't done! Because that's bad enough, isn't it?

I mean I still think the Aug 6 PDB is bad enough for Bush to have been removed from office after 9/11. What can I say?; I guess I just like to see some accountability in my government!

Once again though, it comes down to how many precincts it would take to swing the vote and how many votes per precinct on average, and how best to make the switches undetectable. For Ohio, I think it was 6 votes per precinct. Nationally, it might be about the same.

I would not expect to see a scatter plot showing exactly how this could have been engineered.

Now, if someone wants to show how the outcome could have been reversed, using Mitofsky's own data, that would be interesting.

Also, not so sure the exit polls "predicted" the winner. At what time did the outcome in the polls match that of the vote count? 2 AM on 11/3 maybe, or was it even later than that? And how many votes were actually left to be counted in the swing states at that point? That's about as much of a prediction as going outside on a rainy day to see if you need your brelly, isn't it?

[Febble]

58. Well I tend to agree about the brolly

But the greater the discrepancy between the response data and the count the longer it will take for the algorithm to converge on a number within the required confidence limits.

As for the time, well, GMT, I was getting ready to open my champagne at about midnight, and drank it anyway in a miserable fug around 5.00am.

[Bill Bored]

70. Well I'd still be interested to know how many of the 1460

precincts in the national poll had red shifts AND also had Bush doing better than in 2000, since the lack of correlation between the two seems to be your main evidence for saying the popular vote could not have been switched.

As I said, it's not a shift of 10-million votes I'm looking for, it's more like 1.5 million. That's not many votes per precinct, or many precincts. I really think if you look at it in those terms, you'll see that an "audit" the size of the exit poll would likely NOT detect a fraud of the right concentration in an election this close, and given the large PLDs it would probably get lost in the noise too.

Assuming the origin of the scatter plot is at 0 Red Shift and 0 change in Bush's vote percentage from 2000, look at the upper right quadrant and see if there are any precincts there. If there are enough of them, there could have been fraud. In Ohio this number was 8% of the polled precincts. Nationally, it could be higher because Bush's margin was a bit bigger, but in any case, there should be some precincts that meet these two criteria.

Then, we can divide the total vote count by the 1460 precincts, and estimate the average number of actual votes each polled precinct represents. From there, you could calculate how many precincts it would take to swing the vote and by what percentage they'd have to be corrupted.

But if you are just looking for correlations, you wouldn't find them given the most likely methodology of the fraud which is:

1. concentrated;

2. equal on average in both polled and unpolled precincts;

3. as the bias index has shown, not necessarily correlated with precinct partisanship.

Right?

But you could find it by looking in the upper right quadrant in the uncorrelated scatter plot.

[Febble]

72. Quick and incomplete response

You'd be MORE likely to find correlations if the fraud was concentrated. As Eomer has pointed out, if it was completely uniform in extent and magnitude, it wouldn't show up in an Analysis of Variance because there'd be no variance.

More later.

Lizzie

[Bill Bored]

77. I think we're talking past each other.

I'm saying the fraud was concentrated and it was probably smaller than the variance attributable to other factors. Still large enough to affect the election outcome though.

You say that because there is no correlation between:
1. Bush's improvement in 2004 AND
2. The Red Shift at the precinct level,
that you don't see fraud.

I'm saying you don't need such a correlation for there to be fraud.

As long as there are a sufficient number of precincts in which both conditions 1 AND 2 above are true, their could be enough fraud to swing the election. It doesn't HAVE to be so correlated. It's not such an exact science.

In other words, it's quantity -- not quality.

Now the anti-fraud proponents came up this metric, not I. But even using that, they can't prove that there was no fraud simply by looking at a correlation. To rule out fraud (and only that fraud detectable by the exit polls!), they instead would have to show that there were not enough precincts that met conditions 1 AND 2 above for the outcome of the vote count to be wrong, and so far, no one has done so.

Want to start another thread on this when you have more time?

[Febble]

78. Well, maybe we are

OK. Maybe some of the redshifted precincts were redshifted because of fraud, and that resulted in an increase in Bush's swing. But if so, this was completely balanced by redshifted precincts which showed a decrease (or a smaller increase) in Bush's share - which doesn't look like fraud. It looks like rBr. And it was also balanced by blue shifted precincts that showed an increase in Bush's share - apparent rKr. And by blue-shifted precincts in which Bush's swing was smaller than average or negative. Pro-Kerry fraud?

Well, OK, maybe there was only pro-Bush fraud. And everything else was polling error and random fluctuations in his swing. In that case we have to postulate that rBr was greater in non-fraudulent precincts than in fraudulent ones. And the greater the extent of fraud you postulate, the greater that differential has to be. And I cannot think of any rationale (though I'm quite a creative thinker, and I may come up with one) why rBr should be more prevalent in non-fraudulent precincts but not in fraudulent ones. Wait - I have it - perhaps, in fraudulent precincts, people were hired to tell the pollsters they'd voted for Bush, to make sure hide the tell-tale extra red-shift on top of the rBr. Oh no. They didn't expect rBr, did they....

Honestly Bill, this isn't snark. Yes, there is room for fraud in the numbers. Not a lot, but a sporting chance of a couple of hundred thousand vote-switches. Although not all in Ohio, because that one's been done separately.

But I'll give you decent odds on Ohio. And better odds on NM (odds on, in fact). Unless Eomer has a brainwave, that's all I can do for now.


On your last point: I've been writing new threads in my head all day. I can't face it right now. Maybe when the plot is linkable. And like you, I think the exit polls are a complete distraction anyway.

What I'd like to see is a decent investigation into voter suppression, felon purges, vote spoilage, Blackwell's provisional ballot scam, rationing of voting machines, all of which has the potential for a slam dunk Civil Rights case. And it IMO it cost the Democrats the election. Because it cost Gore Florida, and if Gore had won Florida, he'd be president now.

[kiwi_expat]

85. I think it is important to say that uncounted overvotes cost Gore Florida.

I agree that voter suppression, including felon purges, cost Gore tens of thousands of votes. But Gore would still have won the count if all of the Florida CAST votes (including overvotes with clear voter intent) had been counted. That is a stronger statement.

Kerry also lost tens of thousands of votes through voter suppression -almost certainly enough to cost him the election. But it is not so certain that he would have won the count if all of the CAST votes had been counted.

Cheers!

[Bill Bored]

127. Distraction? Quagmire is more like it!

"Well, OK, maybe there was only pro-Bush fraud. And everything else was polling error and random fluctuations in his swing. In that case we have to postulate that rBr was greater in non-fraudulent precincts than in fraudulent ones."

I think this is only true if the PLDs were the same in all precincts and of course they were not.

But here's a possibility: The precincts aren't weighted by number of voters so the individual PLDs will not necessarily correlate with their effects on overall Red Shift. To be well concealed (and more effective) the fraud would probably have been carried out in larger precincts. Perhaps in these precincts, the baseline (non-fraud) PLDs were smaller due to larger n's.

Also, we know that in the multi-precinct polling places, interviewers could not stick to individual precincts so the PLDs in these places are not really precinct level anyway.

[Melissa G]

123. One of the rather tricky things the republicans are doing is changing,

which usually means consolidating, precincts.. The cost of the new expensive machines they are mandating is one of the reasons given for the consolidation...Although I'm thinking there was some just changing in the Ohio stories..anyone remember?

..using a baseline of comparing 2000 to 2004 stats on inconsistent precincts would strike me as peculiar... Do others remember this?

[kiwi_expat]

124. Another factor: increasing the number of precincts per polling place.

WPE(PLD) is affected by the number of precincts per polling place.

The entire POLLING PLACE is sampled in the exit polls. The results are then compared to the NEP *precinct* vote count to get the WPE.

[Melissa G]

125. Thanks kiwi! I'm concerned about strange variables creeping in that

distort the differences between 2000 and 2004..
Thanks for all your insights!

[Febble]

138. Good point, Melissa

It would be good to know how many of the precincts in the 2004 poll had been changed since 2000

[OnTheOtherHand]

19. umm, no one is claiming WHAT?

"This presupposes that the fraud was widespread or even universal. No one is claiming this AFAIK and no one has said this was necessary to swing the election, esp. the electoral college."

We can take "universal" off the table (I'm not quite sure what it would mean). But the gist of Freeman's argument seems to be that wherever the exit polls differ significantly from the official counts, likely the official counts are wrong. That entails vote miscounting in Alaska, Connecticut, Delaware, Florida, Minnesota, New Hampshire, New Jersey, New York, North Carolina, Ohio, Pennsylvania, South Carolina, South Dakota, and Vermont, and grave suspicions about Alabama, Mississippi, Nebraska, New Mexico, North Dakota, Utah, Wisconsin, and Wyoming (states with high T scores and/or absolute errors over 10 points). Fairly widespread, wouldn't you say? It had better be, since Freeman suggests a 10 million vote net discrepancy in the vote count.

If Freeman argues that the exit polls point to a 10-million-vote miscount, and Mitofsky stands up and disagrees, I think it's nonsense to say that Mitofsky is "dispel(ling) the notion of fraud" or "harboring those who would... steal our votes." By framing the issue in that way, you make it impossible for academics to even touch the issue, since only one kind of conclusion is allowed -- which is probably why most of us have the good sense to stay the hell away.

Freeman didn't even call for paper ballots. He just called for Mitofsky to release the precinct identifiers, because they might contain crucial evidence. Well, someone should go try to convince some judge of that point. Otherwise, if we're tired of hearing what Mitofsky has to say about exit polls, maybe we should talk about something else. Doh.

[eomer]

18. You need to muster more facts to support your case...

because the ones you are using are insufficient to reach the conclusion that you do.

Take two sets of data, each with some pattern of distribution and not correlated with each other. Call these two sets A and B.

Now we are going to lay another set of data, call it C, on top of each of the two first sets. The way we lay C on top of A and B is by adding it. In this way we have created two new sets of data:

A' = A + C
B' = B + C

Check for correlation between A' and B'.

You say that if A' and B' don't show significant correlation then it is not possible that they were created by adding C to A and B.

If I can construct an example where C is added to A and B and the resulting A' and B' show no significant correlation then I have disproved your hypothesis.

Since your logic doesn't involve any assumptions about the distributions of A, B & C, I can make them anything I want. So I will give A & B each a large dose of variation and I will give C relatively little variation. Now hold on for a minute while I check whether my example shows correlation between A' and B'. If it does, I'll increase the variation in A and B and decrease that of C and check again. I'll repeat this process until I get to the point where A' and B' show no significant correlation.

Clearly I can construct an example that disproves your hypothesis.

This means that you need to muster some additional facts to support your case. You need to say something about the amount of variation in A, B and C. Unfortunately in the real world we have no way of determining A, B and C - we only know A' and B'. So I don't know how you would be able to accomplish this, but it is necessary if you want to demonstrate the conclusion you are putting forward.

[Febble]

23. Sorry Eomer, we seem to be talking past each other.

I'm trying to be clear about this, and failing, or else I'm not seeing your point. I agree with your premise, but the way correlational analysis works is that you compute the ratio of explained (or "shared") variance to unexplained ("unshared") variance. As you point out, you have to start with the observed data. Variance that is due to independent factors will be unshared. Variance due to a common factor (e.g. fraud) will be shared.

So have your two sets of variables, each with variance (duh...). And you want to know the extent to which one variable co-varies with the other. Some of the variance will be unshared (independent variance) and some will be shared. And by figuring out the slope of the regression line that minimises the square of the residuals, we can give a value to that slope. And by computing the ratio of the residual variance to the variance predicted by the best fit model (the slope) we can say whether the shared variance is "significant".

Now, as we've discussed, you can't prove a null. If the slope is "insignificant" you can't say it isn't there. However, you CAN put confidence limits on its value. I've had a rough go at doing so in my response to Melissa. And it's very small. Someone might like to try to translate it into maximum net number of votes switched.

I've also thought about your interesting hypothesis regarding an algorithm to prevent Bush's vote going below a 2000 proportions - see my response to Melissa.

[eomer]

27. Yes, I think we are talking past each other.

Let me try making my example more real.

A = Swing before fraud is applied.
B = PLD before fraud is applied.
C = Fraud.

A' = A + C = Swing after fraud is applied.
B' = B + C = PLD after fraud is applied.

Now say that A and B have lots of variation (which I think they do) and are not correlatd (which I think they are not).

Make B have a mean of zero (which means that the exit poll has polling error at the individual precinct level but is accurate on average).

And make C have no variation whatsoever. Make it a constant number of votes switched from Kerry to Bush in each and every precinct.

Your correlation analysis applied to A' and B' will show no correlation, since A and B are not correlated and adding a constant on top of each won't cause them to be correlated.

Yet it is still true that fraud in the amount of C was applied to A and to B to arrive at A' and B'.

Sorry, I know I don't express myself in a way that makes sense to you. I'm pretty fluent in some other branches of mathematics but statistics has never been my thing. See if you make any sense of what I just said.

[Febble]

28. You are right

If there is absolutely no variance in C than clearly there will be no correlation. Although it would have to be a constant proportion, not a constant number, but no problem.

But as soon as C varies then A will covary with B because it A and B share variance in C.

So yes, if you want to propose that fraud of identical magnitude occurred in every precinct, then you are right.

But if C has a normal distribution - it varies in either magnitude, or in prevalence, it will cause A to covary with B to the extent that variance in A and B is due to C. And we can compute the proportion of variance in A that is shared with B and can be inferred to be due to a third, shared variable, C. And we can compute the upper confidence limit of the proportion of shared variance. And it is very small.

Tell me what you think of my skew hypothesis, or a better one, and run it by BillBored or Yowza to see if it is is practicable (or maybe you have that skill yourself - I don't know your background except that it is science).

[eomer]

29. Thanks, I need to ponder this a bit more.

It still seems to me that a certain amount of variation in C could get lost in the ups and downs of A and B respectively and therefore show up as insignificant correlation. Maybe I'll try to do some modeling and convince myself that you are right.

edit: typo

[Melissa G]

50. Hi Febble,In case you don't remember my lack of technical geekness

I restate it here..That said i follow all this pretty well and if i think i am missing something I am recently relying on Bill Bored for clarity. Eomer is stating my case accurately and with far more mathematical expertise than i could ever dream of so I have been silent but appreciative of both of your efforts.

That said I have concern about the way you two are talking about C and normal distribution but I can't articulate it yet...will ruminate further and check in when i have something useful..
Best,
Melissa

[Melissa G]

52. This is around what i wanted to say about C..

Thanks to TFC i can get in the verbal neighborhood if not the mathematical.. He writes..

"PLUS, it's also important to consider the fact that there are a number of kinds of fraud that wouldn't even show up in PLD. Consequently, exit poll bias could have accounted for three quarters of the PLD, and yet there still could be enough fraud to swing the election -- and that's with regard to the popular vote, let alone Ohio."

This is also why I consider Mitofsky as needing a theory of extra voters to say the exit polls are solid because otherwise i don't see it adding up.

I believe in constructing their verbal and mathematical edifices Freeman puts very little weight on FFF in his and Mitofsky and dear Febble put IMHO unjustified weight on it in theirs as an explanation.

This may also partly explain why we talk past each other a lot...


edit to say see TFC's post #44 and for some clarity

[Febble]

57. Melissa, I think there is some confusion here

about the "FFF". It is virtually irrelevant to the current discussion and to Mitofsky's new plot. Actually I think was a bit disingenuous of Steve to use Mitofsky's plot of WPE against Bush's vote share, and not use his plot of FFF against vote share to make one of his points, as he has certainly verbally agreed with me that the FFF is a more valid measure of PLD, than the WPE, but no matter. I think that's water under the bridge. Well maybe not. But it wasn't crucial to Mitofsky's new plot at all. It's only relevant to the argument that PLD was higher in high Bush precincts. And I remain convinced that this is not the case.

But the really critical plot is the new one which you can't very well judge because it isn't posted anywhere yet, although I understand it will be. It does use a version of the FFF, which we have christened Tau. But the result is not particularly sensitive to the measure used. I think Tau is probably the most valid measure we have devised so far however (although a Tau prime is in the works), so it was the one used here. Steve wouldn't have seen it before, so I don't know what he makes of it.

But it shows, as with the ESI plot, that swing from 2000 was not significantly correlated with PLD, as I explain in my response above. This is actually very important.

And while I'm at it: a word about scatterplots. Some people don't like them. I do. The reason I do is that unlike plots of aggregates like means and medians, especially when aggregates are presented with no indication of variance, they show every single damn data point. So when they are presented with a correlation coefficient, it is clear to the naked eye not only what the correlation coefficient means, but also whether the assumptions it is based on are met. I've seen plot after plot in this debate showing aggregated means for various groupings of datapoints. But scatterplots are the real McCoy. Every table, every plot of aggregate values hides a scatterplot. A scatterplot hides nothing. It is the data, warts and all. The "raw" data, in fact, in the case of the scatterplots Mitofsky has presented. Nothing up anyone's sleeve. Scatterplots are your friend. They tell you a heck of a lot if you stare at them hard enough.

[Melissa G]

68. About FFF and FFFV which I could call Mitofsky's new plot but I prefer to

reference as Febble's Fancy Function Variant 'cuz I like you better than him and I'll tell you why in a pm later..
I like Scatterplots too! Back when I was in statistics class I was particularly fond of staring at them as well. They are nice pictures and as you said they can be loaded with lots of data to peruse..

It is however helpful to have the data and the precise methods by which the pretty picture is derived otherwise it is just that.. a pretty picture..

For Mitofsky, or you, dear Febble, to tell us that the pretty picture proves No Fraud (or for that matter anything at all) without the back up data and methods seems unreasonable and kinda funny as an academic argument..

Ironic in fact..Kinda like the hidden vote counters telling me my vote is being accurately counted in the back room with the secret proprietary code.. which is, of course, where this whole discussion started.

[Febble]

71. Thanks, Melissa

for your support and for your PM.

I take your point and will try and explain, though it is difficult without the plots (which I think will be available online shortly, however). I don't know how clear Mitofsky was.

But I'll have a go anyway.

The data: is the "raw" responses and vote counts from the 1250 precincts from the total 1480. The E-M report gives the following reason for confining the analysis to this 1250:

Since election day we have examined information from all 1,480 exit poll precincts in our samples, including all of the exit poll data from election day. This includes presidential vote tallies; questionnaires by demographics; refusals and misses by demographic, etc. There were only 20 precincts where we were unable to get the vote returns from the precinct in our exit poll sample.

Note: The WPE values do not measure just the error of the exit poll in precincts that contain significant absentee vote. When absentees were greater than 15 % statewide, we removed precincts from this study that had the absentees merged with the precinct vote. In these precincts we cannot obtain counts of the election day vote separate from the absentee vote. Also, not included in this study are any precincts with fewer than 20 interviews as well as three additional precincts with large absolute WPE (112, -111, -80) indicating that the precincts or candidate vote were recorded incorrectly. Out of the 1,480 exit poll precincts, 1,250 were included in the analysis that follows.

The first plot showed, for each of those 1250 precincts, Bush's vote-share in 2004 plotted against his vote share in 2004. As you can imagine, there is a strong linear relationship. Precincts in which the PLD was redshifted were shown in red. Precincts in which the PLD was blue-shifted where shown in blue. A regression line was drawn through the plot. Most data points were close to the regression line, but some were above it, indicating that Bush had done better than his average, relative to 2000, and some were below it, indicating that Bush had done worse than his average. Of interest was the observation that blue dots were equally prevalent on both sides of the line, as were red dots. And in fact, regression lines drawn through blue dots and red dots overlapped exactly.

But that is fairly crude - simply dividing precincts in to blue-shifted and red-shifted (and clearly more were redshifted).

So the second plot took the residuals from the first plot. "Residuals" are the vertical distance of each data point from the regression line. So precincts with a positive residual are above the line and are precincts where Bush did better than his average increase. Precincts with a negative residual are precincts where Bush did worse than his average increase (and may even have decreased his vote share). So the residuals are a measure of Bush's gain (loss) relative to 2000.

Anyway, in this second plot, the residuals were plotted on the vertical axis, and "bias" (FFFV, aka Tau) on the horizontal axis. And the regression line was completely flat. In other words, there was no correlation between PLD and residual. And in fact the result is not dependent on using the FFFV. Same for FFF, WPE and something called Tau prime. It also makes no difference how you measure Bush's gain/loss relative to 2000, and there are number of ways you can do that.

Now, a flat regression line does not prove a null. But you CAN compute the confidence intervals of the regression. So you can put a probabalistic ceiling on the maximum relationship. And the upper confidence limit of that regression line, is very small. I put it in another post to you.

If there is a way that this plot can be reconciled with large numbers of votes switched to Bush as an explanation of PLD, I'd be more than happy to hear it. My ballpark estimate is that there is a 1 in 100 chance that it is more than about half a million. More probable (statistically speaking) values are closer to zero(including negative values).

Note that this would not include stuff that went on in the same precincts in 2000. This would include differential spoilage in Dem precincts. And of course it would not include voter suppression. There may be other loopholes. I will keep searching for them.

But it is nonetheless an important finding, that needs to be considered when hypothesising the kind and extent of fraud that is consistent with it.

I'll post links to the plots if they become available online.

[eomer]

75. Febble, I'm not with you yet (and I'm headed in the wrong direction).

In an earlier post I said I might do some modeling and see if I can convince myself that you are right. I've been working on that (it's going a bit slowly because I have to stay employed in between the more fun stuff that I do here).

But so far what I am finding is that the relative amounts of variation in A, B and C (from my previous example) seem to control whether you can safely infer what you are inferring (that correlation between A' and B' would necessarily appear as an artifact of the same fraud having been applied to arrive at each of them). In a model where A, B and C are all normally distributed and have about the same standard deviation as each other, it seems your inference is fine.

But if you make A and B have about 10 times the standard deviation of C then it seems that your inference is false. Somewhere in between the same standard deviation and 10 times the standard deviation (don't know where yet) you cross over from fine to not fine.

Please take all this with a grain of salt because I need to go back through my work and make sure it's correct. I wrote my own little application to do the data generation and another one to do the statistics so what I'm finding could be a result of a bug in my code (it's not infallible like evoting software supposedly is) or a flaw in my understanding (all too fallible).

I'll let you know when I feel more comfortable with my numbers but, unfortunately, have to go to work now.

Cheers,
eomer

[Febble]

79. Your model will work

I think, but it will have to make some pretty heroic assumptions.

Virtually ALL the VARIANCE in PLD (however we measure it) will have to be fluctuations in polling error, in both directions. It can't be sampling error alone, Ron Baiman, bless him, has demonstrated that, and it is pretty obvious from Mitofsky's AAPOR plots. So it will have to be some kind of polling bias.

The less variance you attribute to polling bias, the more your fraud will show up in the correlation. I can't quite get my head around this without a model, but the more of the net shift you want to attribute to fraud, without producing a significant correlation between swing and shift, the more polling bias you will have to postulate. And you will also have to postulate that the magnitude of fraud was identical in precincts with a massive blue shift, regardless of swing, as in precincts with a massive red shift. And of course to fraud in precincts with no shift at all. In other words, every precinct in the poll will have to have fraud, in identical proportions. Or near enough. Which is at least refreshingly different to most hypotheses, which are, variously, for fraud that was greater in swing states, high Bush precincts, THE EAST (THE BEAST), or only in unpolled precincts.

But I'll be interested to see the results of your model. Models are good.

[rdmccur]

53. Wait a minute

I don't understand how A and B share variance in C if events A and B are totally independent of C (how can you argue that A and B depend on C??).
I would humbly ask for clarification.

[Febble]

54. TfC, Mellissa, Eomer, Bill, too

I've only got time for one post so I've posted to rdmccur:

First of all, rdmccur: if A and B are totally independent of C they won't share variance. If A and B are both partially dependent on C they will share some variance. if A and B a totally dependent on C they will share most of their variance. So the proportion of variance shared between A and B tells you the amount of variance in A or B accounted for by C. If C is fraud, it could index the amount of fraud. If there is NO VARIANCE in fraud, it won't tell you that. All statistics referred to here is ANalysis Of VAriance (ANOVA aka the General Linear Model).

More generally:

I'm not saying, and I know Mitofsky's not saying, that the exit poll evidence rules out the kind of fraud that wouldn't show up in the exit poll. Clearly it can't. When Mitofsky says the new plot "kills the fraud argument" he isn't saying it kills the argument that there was fraud in the election. He is saying it kills the argument that fraud accounts for the exit poll discrepancy (which is what Freeman is arguing), and the corollary of that is that the exit poll discrepancy is not an index of the magnitude of fraud. There are a couple of senses in which that is not completely sayable, but mostly it is, IMO.

However, it does NOT rule out the kinds of fraud that would not show up in an exit poll. TfC has made a good case for these, and it includes voter suppression. Vote spoilage WOULD show up in the exit poll, but not in the ESI plot nor Mitofsky's new plot, if it went on in precincts where it has gone on for years. And we know it has.

Regarding Eomer's constant. It is true that if the amount of fraud was completely uniform - happened in all precincts, and by exactly the same proportion - it would shift the mean PLD and not show up in the correlation. But (at the risk of echoing TIA) to believe that you'd have to believe: that fraud to the same degree happened in: precincts in blue shifted states as much as in redshifted states: in precincts with clean recounts; in precincts where the PLD was calculated with the precinct count; in precincts where the PLD was calculated with the county tabulations; in precincts with massive blue shifts, and with zero PLD. And I have to say my prior for that is low. Vanishingly low.

So there are a few caveats. But I am being pretty scrupulous here. As long as we assume that there were at least a substantial number of precincts in which there was no fraud (and I do think it is somewhat circular to have to assume that even blue-shifted precincts were fraudulent to the same extent as red-shifted ones); and/or the degree of fraud was not constant (little is in life) then the new plot puts very severe constraints over the net amount of new, vote-switching fraud that can have occurred in 2004. I've computed the upper confidence limits of the correlation, and the upper confidence limits on the net (notice my word net here) number of vote switches is going to be well below 1.5 million. My guesstimate is a few hundred thousand (and that's the UPPER confidence limit - 99% - more likely figure is lower). Now this would be enough to swing key states, but not enough for the popular vote. And if Steve thinks he can account for 10 million in the light of the new plot then I'd like to see the math. This isn't some kind bazaar where you name a ridiculous price then come down. Or if it is, it's not math-as-we-know-it.

So this is where I think we are at:

Differential spoilage could still account for some of PLD, and not show up in the new plot.

The plot puts upper limits on the net amount of vote switching, which could plausibly still be in hundreds of thousands but not millions. It could also be zero - the centre of the confidence interval is zero. So it makes popular-vote stealing scale theft vanishingly unlikely IMO, pace Eomer's two theories. I don't buy the constant theory (though I gave it a whirl!). The stopping-Bush's-vote-share dropping below 2000 levels is intriguing, and does make a testable prediction, although the test wouldn't be conclusive either way. Geeks: could it work?

Nonetheless, if we assume that a limited amount of vote-switching is compatible with the plot, and that some forms of fraud won't show up in that plot, and add voter suppression and vote spoilage, and target it in key states - yes, we haven't ruled out election-stealing fraud.

So I'll back Mitofsky's statement at this stage and say that the new finding kills fraud as an explanation for the exit poll discrepancy (except for the grand geek algorithm maybe). It doesn't kill fraud as explanation for the fact that Bush is still president.

[rdmccur]

60. I need some

time to digest this.
Thanks.
I'm only thinking that there may be 'hidden' assumptions (not necessarily wrong) in these arguments that may call them into question. These kind of assumptions occur at a very basic level; I've seen it happen it mathematics often.

[OnTheOtherHand]

61. yes, you are quite right

Scientific progress -- and social progress -- often come from challenging hidden assumptions. Many of us have gotten somewhat caught up in polemic even despite ourselves, so you may see what we don't. (Or if you don't today, you may soon.)

[Febble]

63. I agree with OTOH

Hidden assumptions are the very devil. Eomer unearthed one I had ignored. I don't actually think it undermines the conclusion, but it was worth checking out. Please tell me if you find more.

Oh, and see this post:

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=397074&mesg_id=397793

I see you went over to PI, presumably on my directions. If you want to google me for additional information my name is Elizabeth Liddle, aka Lizzie.

[Time for change]

94. But Febble

The exit poll discrepancy was 6.5%. Can you tell us how much exit poll discrepancy due to fraud could have been consistent with Mitofsky's analysis? Certainly it's not zero.

[Febble]

108. It's not an easy calculation

So I made generous assumptions. Taking the most generous boundaries for all guesstimates, I put the 99% confidence interval at around half a million net gain for Bush. It can't be all in Ohio, as we have a separate plot for that, but it could be concentrated in some other swing state.

As I say, it's broad brush, but you'll have to trust me that I've been generous. If anyone else wants to have a go, you have to take a stab at the variance by eyeballing the plots (which I hope will be available online) and make various assumptions (which you can get help with from the E-M report) about how representative the precincts were of the country as a whole. Then you have to extrapolate the result of your calc to the final margin in the count.

I don't expect anyone to do that, but if you trust me, I'd say max, half a million. Zero is perfectly consistent (but possibly wrong). the lower the number the greater the probability that that was it.

Obviously it is theoretically possible that it was all fraud, but only if it was targetted very precisely, in the right quantities, in the precincts where Bush did badly, and avoided like the plague in precincts where he was doing well. The probabilites assume a random pattern - or at least, a pattern orthogonal to Bush's increase. If the fraud was tied to Bush's increase, then you can increase the probabilities of fraud.

I'm saying this to be scrupulously honest. I can't actually see a method myself that doesn't violate something else we already know. So if you want to pursue this, the next step is to hypothesise a method by which fraud of this type could be implemented. If it generates testable predictions, maybe they can be tested. Mitosfky has offered to test any hypothesis for Steve.

But remember that around 60% of the vote counts used in the bias calculations were collected at the precinct. So most of it can't have been central. I think you'd need a fiendishly complicated algorithm, ON THE MACHINES.

[Time for change]

109. Ok, thank you. So half a million if random

But I don't think that it was likely to be random. In order to be consistent with Mitofsky's findings it would either have had to have been targeted at places where Kerry wasn't doing as well as expected, and I suppose that this could either have been purposeful or not, but if the latter, then "targeted" would not be quite the right word for it.

I think that someone with more knowledge (not statistical) about this than I would have to offer an opinion on the likelihood of targeting precincts like that.

And another question would be whether Mitofsky has released enough information on his analysis that it could be independently evaluated. Do you have a read on that?

[Febble]

110. No

it would have to be targetted where Bush was not doing as well as expected.

And half a million is the upper 99% confidence interval. A very generous upper. There is a 1 in a 100 that it would be higher than this. There is a much higher probability that it would be a lot lower. The middle of the confidence interval is about zero. I'm NOT saying it shows there were half a million votes lost (quarter of a million votes switched). I'm saying that any more than that is pretty incompatible with the data, and the greater likelihood is that it was near zero.

And by "random" - I don't mean thoughtless. I mean orthogonal to swing. In other words fraud would have to have been specifically targetted at precincts where Bush was doing badly relative to 2000. Targetted in swing states, or targetted in blue states, or targetted at DREs or targetted at op-scans, unless the target was pretty collinear with Bush's real increase/decrease in vote share, it would be orthogonal to swing.

And if fraud WAS targetted in precincts where Bush was doing relatively badly, that would tend to show up in a positive skew in the swing, and the plots suggest, if anything, a negative skew (more precincts a long way below his average than a long way above.

No, there is not enough information in the public domain to replicate the analysis independently. You'd need the response counts, the vote counts and the vote counts for 2000. So the other defence against the plot is that the data is invented.

But if Mitofsky is inventing data, there is no point in demanding its release!

(I don't think the data is invented).

There may be other loopholes. I continue to think. I am wondering what eomer is going to come up with. I don't think there's a lot of wiggle room, though.

[Time for change]

111. Yes, of course that's right, it would have had to have been targeted at

areas where Bush was doing badly relative to 2000.

As I'm sure you know, I strongly suspect, on the basis of a good deal of evidence (IMO) that Cleveland was the source of tens of thousands of lost votes for Kerry, perhaps even 100,000. It was widely believed that Bush was doing quite badly there, and if I am correct about this, even with a great deal of fraud he still did not do well there relative to 2000. So I think that Mitofsky's analysis and that of others leaves Cleveland as a potential source for vote switching fraud -- though I believe that most of the fraud there occurred by other means.

I understand your point about half a million being the upper 1% bounds of your estimate.

And I was not accusing Mitofsky of dishonesty by the question that I asked about independent verification. As you well know, scientists sometimes come to different conclusions when they look at the same data, and that different a priori perspectives can sometimes affect this.

[Febble]

112. Well, I think that is a good reason

(taking your last point) for reporting with a clear methods and results section, which was conspicuously missing from the E-M report. I hope that the method and result of those plots is clear

(see Mark Lindeman's http://inside.bard.edu/~lindeman/slides.html">webpage)

I think the axes are fairly self explanatory.

Interesting point about Cleveland. And of course, large fraud concentrate in one area would only contribute a small amount to the exit poll (as polled precincts are widely distributed). This is why I say that it is important to listen to what the exit polls are telling us about where fraud ISN'T in order to check out where is left!). As I keep quoting, from The Cat in the Hat:

When you mislay a certain something,
keep your cool, and don't get hot,
calculatus eliminatus is the best friend that you've got,
calculatus eliminatus always helps an awful lot,
the way to find a missing something is to find out
where it's not.

If it's not in the exit polls, maybe it's in Cleveland!

[Melissa G]

115. Hi Febble, I wanna restate that I appreciate your care

and diligence at looking at things different than you believe...I think the exit polls are helpful as you say in finding where the votes are not... The reason I like TIA and others around here who converged in Nov's work and (I know you are not a fan of this line of thought)is that we started looking for where the 9-11? million new Bush votes could have come from. For some of us that was the puzzle and still is...
Another possible reason why we talk past each other some times.

The first numbers I wish we had reliably confirmed are actual votes cast at polls, the actual votes absentee and mail in. I wish I could have confidence in the universe figure of votes cast and confirmed by signatures and poll workers.

My theory is that figure is a place that would tell us that a significant number of votes are a figment of computer memory either voting machine or tabulator..

then I would have a sense of the the figures to look for vote switching which we already know are quite small like 1.5 as an unadjusted figure maybe even smaller when subtracted from a reliable actual vote cast figure... Hiding in big deep pockets like Cleveland and hiding in small wide pockets like rural Texas, it seems to me an election could be easily carted away without the clear correlation you keep searching for..

But there still is LOTS of evidence in the exit polls that these votes were not there due to all the folks who say they did not put them there..


I still have not heard a really successful answer from anyone about this phenomena... Here is where I find some of the negative correlations..he lost some due to death.. He lost some due to gender.. he lost some due to Nader voters..etc.. No good theory of where he found those new votes...hordes of Evangelicals Still seems weak.

I know the infamous game was a stab at solving that question but for me the question still stands unanswered..

and I not sure how staring at scatter plots can make it go away.. Any insight?

Thanks,
Melissa

[Febble]

118. Thanks, Melissa, again

I'm not sure I quite understand your problem.

As I see it, the problem is to reconcile a fraud explanation of the exit poll discrepancy with the absence of any correlation between the discrepancy and the degree to which Bush improved his vote share.

If the discrepancy WASN'T due to fraud, it is actually quite easy, as there are lots of potential ways for a poll to have bias.

But because we have reason not to believe the count, we should, as Steve says, have grave suspicions about what the discrepancy indicates.

The question really, now, is whether we can largely ascribe the exit poll discrepancy to polling bias, and concentrate on finding out why there were so many irregularities, particularly in Ohio, and New Mexico, where relatively little fraud could have had a big impact.

[Time for change]

119. I have two thoughts on this

although I'm not sure I understand all your points.

The first thought is about Cleveland. I think that there are many reasons to believe that Cleveland could be or is the source of enough fraud to account for well over half of the Bush victory margin:
http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=104&topic_id=4998616. If this was done the way that I am surmising here it would be perfectly consistent with Mitofsky's analysis without even involving a few thousands of vote switches. A talk I had with an ACT member from Cleveland last night was thoroughly consistent with this scenario. He and his organization worked all election day in Cleveland, making sure the beleaguered voters were well fed, and just trying to convince them not to leave the long voting lines. Yet we're told that turnout in Cleveland (which voted 83% for Kerry) was 51%, compared to 73% in the rest of Cuyahoga County.

Then there is the lockdown in Warren County rationalized by the bogus national security alert. Warren County did not have enough voters to make that big of a dent in the Kerry lead. But some people have said that they were able to control the central tabulators from adjoining areas from the room where they were "counting the votes" in Warren County -- though I haven't seen any documentation of this. Anyhow, it this is true, large numbers of votes could have been added to large areas of the state by manipulations of central tabulators, again without being inconsistent with Mitofsky's analysis, as long as the added votes (to Republican precincts) were in proportion to the votes that had already been counted in those precincts.

Just some thoughts. There isn't proof for these things. But I think that this would go a long way towards explaining some of the mysteries.

And I agree that if these scenarios are correct we can thank Febble for helping to point us in the right direction.

[kiwi_expat]

121. 6 of Ohio's 49 NEP precincts are in Cuyahoga (Cleveland)

But I wouldn't use the exit poll data to determine which to manually audit. There are glaring vote anomalies in some Cuyahoga non-NEP precincts. That is where we should be doing manual audits.

Let's go to Cleveland!

[Time for change]

134. Thank you for the link Febble

but it appears not to work.

[Febble]

135. Damn

Here:

http://inside.bard.edu/~lindeman/slides.html

I must check my links before posting.
I must check my links before posting.
I must check my links before posting.
I must check my links before posting.
I must check my links before posting.
I must check my links before posting.
I must check my links before po........

[Time for change]

137. Thank you -- got it and book-marked it

[Bill Bored]

128. a fiendishly complicated algorithm, ON THE MACHINES?

Or just enough corrupt ballot definition files?

[Febble]

133. Presumably that too

if that affects the count at the precinct, because for about 60% of precincts, the poll count for each precinct was compared with the count at precinct level, collected at the precinct on election night at close of poll. For the remainder it was compared with the precinct count at county tabulation level.

So if you can devise an algorithm that only kicks in when Bush's vote falls below a certain proportion relative to 2000, then there's a loophole. Trouble is that would predict that the distribution of swing would be positively skewed, over all, and eyeballing the plot it looks the other way, if anything. But it was in a few key places, maybe....

[eomer]

144. Finally, a response to your response.

Hey Febble, as you know I've been doing some modeling to see whether I agree with your assertion about fraud and correlation between PLD and swing (pardon the delay but blame it on Wilma).

Here is your assertion:

If Fraud was responsible for the Great Exit Poll Discrepancy, then where the PLD is greatest, the boost to Bush’s vote should be greatest. After all, the point of fraud is to win the election. Cui bono?

If Non-response Bias was responsible for the Great Exit Poll Discrepancy, PLD should vary independently of Bush’s advantage in the vote count.

Mitofsky correlated a measure of PLD with a measure of advantage to Bush, namely the increase in Bush’s share of the vote relative to 2000 (a year in which mean PLD was near zero).

And there was no relationship. PLD is not correlated with boost to Bush’s vote share.

So, carefully as Steve’s case was built, I simply don’t think it stands.

I developed a model that shows that fraud added to both PLD and swing can be the cause of the positive mean of PLD and still not produce any significant correlation between PLD and swing.

The model starts with these 3 sets of data:

• Set A, a proxy for pre-fraud PLD, consists of 1250 values that follow a Gaussian distribution with mean of 0 and standard deviation of 4.0.
  
• Set B, a proxy for pre-fraud swing, consists of 1250 values that follow a Gaussian distribution with mean of 0 and standard deviation of 4.0.
  
• Set C, a proxy for fraud, consists of 1250 values that follow a Gaussian distribution with mean of 5.5 and standard deviation of 1.0.
  

(The model is abstract - the 3 sets of data are proxies for PLD, swing and fraud but are not intended to simulate their actual values, means, variance, etc.)

From these 3 sets of data we develop 2 more:

• Set A', a proxy for post-fraud PLD, consists of 1250 values each of which is calculated as the sum of the respective value from A and the respective value from C.
  
• Set B', a proxy for post-fraud swing, consists of 1250 values each of which is calculated as the sum of the respective value from B and the respective value from C.
  

Here are statistics for one run of the model:
mean of A': 5.41935
mean of B': 5.69783
variance A': 17.93862
variance B': 16.69672
std dev A': 4.23540
stdDev B': 4.08616
regression of B' on A' (slope): 0.00531
covariance: 0.095108

The regression line is virtually flat. We could make it even flatter by upping the ratio of variance in PLD and swing to that of fraud or less flat by lowering the ratio.

The model seems to show that it is possible to have fraud be the one and only cause of the positive mean of PLD (and also cause a positive mean for swing) and at the same time not show up as correlation between PLD and swing.

A key assumption that allows the model to be a counter example is that the variance for fraud is lower than that for PLD and for swing. Effectively the fraud proxy gets lost in the noisy data to which it is added but only from a correlation point of view. It is still willing and able to sneakily move the mean of both the PLD proxy and the swing proxy from about zero to about 5.5 but without showing up as correlation.

BTW, I'll be offline this evening (and I guess it's already getting late over there) so I'll check back in tomorrow.

[Febble]

145. OTOH and I have been working on similar lines

but OTOH has taken it further than I have.

Why not compare notes when you are back. I'll email the URL of your post.

How's the clearing up going?

[eomer]

147. The clearing up is almost done (thanks for asking)...

I'll finish up this weekend. Still no electricity and no water (the well pump is electric) but other than that we're almost back to normal. I drive a Prius (like a good liberal would) and turned it into a little generator by hooking a 600 watt inverter to the 12 volt battery (i.e., not to the big high voltage one). The Prius is smart enough to start the gas engine only now and then when it needs to recharge the battery. Works like a charm.

[Febble]

148. Cool. I'd like a Prius

but the UK isn't ready for that yet.

But I've got my bike.

[eomer]

150. Hmm... I've been out liberaled (bike beats hybrid, rock beats scissors,..

N/T.

[OnTheOtherHand]

146. it seems to me that with fraud variance that low

indeed the Mitofsky argument fails because all the "points" are moving together -- there is no way for a difference between fraud precincts and non-fraud precincts to emerge, because there are no non-fraud precincts.

However... is anyone arguing for uniform (or even almost uniform) fraud? Surely Freeman isn't?

I don't know whether I even bothered to state _that_ assumption, so I will go back and add it. More later....

[eomer]

149. Hey OTOH, how's it going?

My point is that the assertion has been put forward as if it is a mathematically necessary conclusion while I have the feeling there are elements missing in order to reach that level of certainty.

One missing element (the one I pointed out) is knowing something about the relative variance. The variance of A, B and C should ideally all be the same if we want to reach Mitofsky's conclusion.

Another missing element is knowing whether the distributions of pre-fraud PLD, pre-fraud swing and fraud are all normal. They really should be if we want to draw Mitofsky's conclusion (or at least it seems to me). The fact that we don't know whether they are all normal means there is another missing but necessary element.

If those flaws exist then there may be more (it only takes one counter example to prove that the logic is insufficient).

BTW, sorry if any of this is a rehash. Still catching up.

[OnTheOtherHand]

151. well, fair enough

I don't think anyone intended to put it forward as a mathematically necessary conclusion, although I can't speak for all the people involved. So far I stand by what I said in my little essay -- if there is fraud in there, it is well disguised. (Perfectly possible especially for narrowly targeted fraud. Indeed, fraud could avoid the exit polls entirely -- but then, why oh why do we keep talking about them?)

Normal distributions aren't a necessary assumption, as far as I can tell, but I don't claim to have canvassed all imaginable distros.

A lot depends on context. ESI did an obvious test, it came up negative, they said 'no smoking gun,' and they are taking flak for it. I think ESI made the right call -- although I don't see any way that the Ohio exit poll data could "rule out" decisive fraud. On a national scale, well, we'll see. (I am falling asleep over here, so I had better not try to sum things up!)

[eomer]

152. Let me rephrase that...

I think when I said "a mathematically necessary conclusion" I really should have said "a valid inference". Those two mean the same to me but the first sounds more rigorous when really it is not. I do believe the asserters have represented their assertion as a valid inference. If there are conditions that are necessary for the inference to be valid and those conditions haven't been demonstrated then... well you know what I mean.

On the part about one or more of the data sets not being normally distributed, I wasn't thinking so much in terms of some other distribution. I was thinking of fraud that was systematic and therefore did not follow any statistical distribution.

I played around with some systematic (or algorithmic) definitions of fraud and found some that produced correlation and therefore are not counter examples to the assertion. For example, setting the fraud proxy equal to 6 in every third precinct did, as you would expect, create significant correlation between PLD and swing.

One systematic version of fraud, however, did not create correlation. I took every precinct in the first half of the data and looked at the swing. If the swing was negative I set the fraud equal to the swing times negative one. If the swing was not negative then I set the fraud to zero. In the second half of the data I set the fraud always to zero.

The first run of the model as just described produced a regression slope of -0.0095. That surprised me so I ran it again and got a regression slope of -0.0443.

I'm not sure this scenario is a counter example yet because (I think) it would produce a situation where some lessor portion of the PLD was caused by fraud rather than all or most of PLD being caused by fraud. Maybe it could be combined with some other type of fraud that also does not produce correlation. Or it could point to part of the PLD being explainable by fraud and the rest of it being polling error, which is what I think is more likely. This latter possibility would be a partial concession by both sides of the debate.

[OnTheOtherHand]

153. yours works if

the mean swing is 0 (or it mostly works -- the expected corr does seem to be around -0.04, which would obviously sneak under the radar in Ohio). At that level, it induces about 1 point of fraud swing (2 points on the margin). It degrades as mean swing tilts toward Kerry (at mean swing = -3, the corr is about -0.08 -- but if we can settle for 1 point of fraud swing, we could presumably tweak the algorithm).

(Why would we tilt mean swing toward Kerry? Because Freeman argued that the data indicate that Kerry won by 6 to 7 million votes; a more conservative estimate would be that Kerry won by 3 points or perhaps about 3.6 million votes. If we can set that aside, then mean swing can be close to 0 -- but then we are tacitly conceding, I believe, that a lot of exit poll error is exit poll error, not vote count error.)

My variant was to apply fraud, normally (but fairly narrowly) distributed, in all precincts where actual swing favored Kerry by more than (arbitrarily) one point. That degraded pretty quickly as the proportion of fraud precincts decreased; yours might be more viable.

I do think it is unlikely that we will get all systematic PLD to be explained by fraud. Even if we can make it work with respect to the scatterplots, I don't think it will make sense in real life. So, if I am right, the q becomes how much fraud we can "work in," and what constraints we have to accept in order to do so.

[eomer]

154. Yes, that sounds like the right q to me.

BTW, can you give me a link to your essay?

[OnTheOtherHand]

155. here ya go

http://inside.bard.edu/~lindeman/slides.html

You'll see that I've tried to work in some of the stuff we've talked about. The essay is reaching its outer limits -- originally it was just intended to explain the scatterplots. Now that it is being read in the context of arguments about smoking guns and rulings-out and argument-killings, it's getting harder to get the emphases right.

[bleever]

12. K and R! Thanks!

[CalmMan]

20. followup reporting on all this?

I was there, alarmed, and looking for followup. I did a google search and this was all i found. How come this stuff isn't being reported?

[CalmMan]

22. Not just for Stats and Math fans

BTW: This stuff is not just for Stats and Math fans. The material Freeman presented should be pretty comprehensible to anyone.

[Febble]

25. Except stat and math fans?

I jest. But just because something is comprehensible doesn't mean it's right. There are defensible arguments that quite a lot of it may be wrong.

[Time for change]

30. Thank you posting this Melissa

Yes, I agree that Freeman did a very good job of presenting this.

There was a large discrepancy between the official results and the exit polls, and only fraud or exit poll bias, or a combination of both can explain it. If fraud explains most of the PLD, that means that Kerry really won the popular vote (and almost certainly Ohio and the electoral vote).

Here is what I find convincing explanations that it was most likely fraud:

1. PLD significantly higher in swing states. Yes, as Febble says, it is possible that greater exit poll bias in swing states could be responsible for this. But certainly fraud is a more plausable explanation for this.

2. PLD lower with the use of paper ballots. Yes, as Febble notes, it is possible that this finding is not statistically significant, and it is possible that when controlled for other variables that the relationship could disappear. However, Mitofsky did not present the statistical signifcance of this result, and as far as I know, no data has been publically released that would allow one to calculate the significance of the lower PLD when controlled for other variables.

3. No other Mitofsky exit poll has been as far off as in this election, where we have secret vote counting by computer on a massive scale. Yes, the exit polls were off in 92 by a comparable amount, but still I find it significant that in no other election have they been this far off.

4. Mitofsky's report (from earlier this year) presents little or no evidence that bias accounted for a good proportion of the PLD, as Freeman meticulously and lucidly explained. Yes, it's possible, as Febble points out, that when multivariate analysis is performed that something could be found to change this. But this was not presented in Mitofsky's report, and as far as I know the data is not available for public analysis anywhere else.

5. Higher PLD where Republican Governors are in charge is additional evidence.

6. So is higher PLD where voter complaints are more numerous -- though I would like to know more about what kind of complaints these were and what kind of statistical analysis was done on this. I do find it a bit confusing that Ohio wasn't near the top with regard to these complaints.


On the other side of the ledger I will mention two things:

1. It is not at all clear to me the point that Freeman is trying to make by correlating the percent of blacks with PLD (I wasn't at his presentation, so maybe this would have been clear had I been there). If the point is that black voters are suppressed, I don't get it, because then they wouldn't have shown up in the exit polls. In other words, I don't see what the percent of black voters could plausibly have to do with PLD.

2. The finding of no correlation between Bush swing (from 2000) is the best point (and maybe the only one) that I have heard Mitofsky make on this subject. However, it is not clear to me how much swing would have had to occur before it could have been statistically detected. Also, I haven't seen this data in written form anywhere, so it is difficult to know what to make of it. Of course, as OTOH has pointed out, Freeman was not familiar with this data either, at the time of the debate, so he would have been quite handi-capped in trying to combat that point.

[anaxarchos]

31. Your point #1....

... African-American voters historically have a much higher % of residual vote ("spoiled ballots"), thus they can show up in the exit polls but not in the vote total. There is a lot of material on this being intentional disenfranchisement.

[Time for change]

32. Ok, that makes sense -- but are you sure that was his point?

Definitely, black voters use less well maintained voting equipment, and that would show up as PLD in precincts where there were a lot of black voters. But my understanding was that this effect would be way to small to account for the vast PLD shown in this election. Enough to have cost Gore Florida in 2000, for sure. But not the vast difference demonstrated in this election in Ohio or nation-wide.

[anaxarchos]

33. 2 Points...

1) Let me quote you what I quoted Febble:

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=330828#333264

Posts #35, 40, 57....

1.3 to 1.5 million votes in 2000 is not "small". Klinker's point is that equipment, economic status, literacy, etc., by themselves, don't account for the residual vote.

2) These precincts are a "marker" - They have been "fraud" ridden since Reconstruction and are a good place to look for "additional fraud" now, although I am on record up thread as saying 2004 isn't this simple.

[Time for change]

36. I agree it's not small

It's just relatively small in comparison to the much larger exit poll discrepancy in 2004. The total exit poll discrepancy was 6.5% nation-wide.

Let's say that 1.5% of the votes were spoiled in 2004. Even if two thirds of those were Kerry votes, and one third were Bush votes, that would be a net loss to Kerry of 0.5%. That's pretty small in comparison with the total discrepancy of 6.5%.

AND, I think that voter suppression, in Ohio, probably accounted for another 1% or so net loss for Kerry, as a low estimate. The exit poll discrepancy in Ohio was 6.4%. Fraud due to voter suppression in Ohio wouldn't even have shown up as an exit poll discrepancy. So you'd have to add that to the 6.4% to get the true extent of the problem.

[anaxarchos]

46. Not to belabor this but...

2 more points:

1) If you followed my references, you are underestimating both the residual vote in 2000 and the Democratic plurality (perhaps for 2004 as well - MIT/Caltech initially claimed residual vote rate was cut by one third but that was only based on initial numbers and 39 states). Much more importantly, it is not uniformly distributed either by precinct or even by state. It ranges from negligible to 15%. More, the US Civil Rights Commission bit the bullet and called it a form of "electoral fraud" in many cases. This is clearly part of the issue.

2) A much more subtle point: One of the original claims for DREs was a radical decrease in residual votes. In 2004, victory was declared in at least Georgia and Florida. The rates went from 3.5% and 2.9% in 2000 to .4% for each state in 2004. Both M/CVTP and press accounts had predicted a significant resulting blue-shift, at least locally... Where is it?

[Time for change]

59. Ok then, can you tell me what you estimate to be the total net loss of

votes to Kerry from spoilage in the 2004 election?

Regarding your point 2, I don't think that a decline in residual votes would result in a blue shift. It would result in a decline of the red shift due to residual votes (but maybe that's what you meant).

[anaxarchos]

62. A reasonable number would be 1.8 to 2.0 million votes...

..."spoiled" in 2004 (the numbers are still not available). That would be a residual vote rate of 1.6% which is consistent with the CalTech/MIT preliminaries and also the claim that DREs reduced spoilage by 1 million votes (spoilage in 2000 was 2.1 to 2.4% depending on how you count). That gives a Kerry PLURALITY in "spoiled ballots" of 1.2 to 1.4 million votes (50% black, of which 90% would go to Kerry; 25% Hispanic; 25% White at better than 3 to 1 and 60/40, considering income level) if 2004 was similar to 2000.

As to whether you would get a blue shift or a "decline of the red shift" in Florida and Georgia, it is really a question of what you anticipated of the exit polls in 2004. As a practical matter, you would expect roughly 200,000 more votes counted in Florida and 100,000 in Georgia and these would be overwhelmingly Democratic. If nothing else changed (big if), the exit polls would not change either as they were already "counting" this vote. The point is that this is additive to any red shift in the 2004 exit polls.

If you go county by county, the story gets even worse because there is no statewide uniformity in the distribution of "spoilage" either...

[Time for change]

67. Ok, I get something close to that

Using your figures, and assuming 2 million spoiled votes (which I won't argue with) I get a plurality of 1.15 million for Kerry. So, that's about 1%. But didn't the Florida 2000 count of spoiled ballots come up with a ratio for Gore that was quite a bit less than that?

With regard to the other point, the amount of red or blue shift isn't based on expectations, but rather the difference between the exit poll and the official count, regardless of what the expectations were.

[anaxarchos]

69. The Gore Florida "votes" from 2000 are all based...

... on given criteria for determining the "intent of the voter". There were many more "spoiled ballots" which did not meet any given set of criteria ("one hanging and one indented chad", etc.) and still others for which intent is not discernable at all. The intention of the voter is deduced by demographics and thus has no bearing on the election.

On your second "point", if you prefer the term "apparent blue shift", be my guest. The issue is still as I described it. Febble is exactly right on this point. This has been going on for years. As long as there has been research on this topic, the non-random nature of "spoilage" has been understood (you can follow the references back from what I cited). It has "appeared" in political "lore" well before that. In certain states (primarily but not exclusively Southern), the spoilage rate has often been twice what we just described.

Now, you can take many different positions on the exit polls. You can take the position that they have been relatively accurate despite this obvious source of error (in which case, eliminating it creates an "apparent blue shift") or you can take the position that exit polls have "always favored" Democrats (and this is why?) and this time they were "remarkably accurate" in areas in which "spoilage" was dramatically reduced or you can adopt yet another position. I am merely "avoiding controversy" because I have not looked at the issue.

Whatever position you take, my question remains the same: Where is my "apparent blue-shift"?

[Febble]

73. Well, thanks anax

We do sometimes seem to agree.

My beef is with the argument that the exit polls index NEW VOTE-SWITCHING FRAUD. I.e. new since 2000.

I don't think they do. Or at least there is a low max.

But not for spoilage in the same precincts as forever. And this is where it becomes a Civil Rights issue. I think, and I've always thought, that there is a huge Civil Rights case staring us in the face, and has been staring us in the face ever since we saw those black faces under those umbrellas in Ohio. And I think the exit poll evidence is a HUGE distraction from that. Even if you find TIA's evidence mathematically convincing (and while you are entitled to your view, it is the mathematician as well as the social scientist in me that is unconvinced) the fact remains that there is SIMPLY NO ARGUMENT that black Americans were systematically disenfranchised in 2004, as they always are, and that Kerry was the net loser (though frankly that worries me less, given his lack of fight - even if he was convinced he'd lost, he should have stood up for the rights of his black voters. And even for Bush's black voters).

So enough already with the snark about me on PI?

I'm not spinning a bloody thing.


(edited for crass typo)

[anaxarchos]

97. We have agreed on one point...

... important, though it may be. I have not nominated you for a Nobel Peace Prize. Such broad statements of personal vindication might be a little premature. You might even regret our agreement in light of what I conclude from it:

1) You might have noticed that I do not regard the exit polls as the primary prop for my conclusion that the 2004 presidential elections were "stolen". I think the numbers for a Bush victory do not and can not hold. You can call this the "political logic" of the election if you like. The exit polls merely provide supporting evidence. Thus, I am naturally suspicious of any intrinsic exit poll analysis which suddenly centers on hypothetical response errors. Given the larger issue, this sudden corruption of supporting data is altogether too convenient.

We have never debated this "political logic". We were supposed to do so in the "game", and we have approached the threshold again multiple times, but we always seem to get "diverted" by an issue (often obscure) intrinsic to polling. I know that OTOH, at least, is aware of the issue because despite the fact that he has abruptly dismissed the possibility, he keeps running into it. Up thread, he mentions (to his horror) that Freeman put up a political "scenario" that could have come directly from "the game". At another point he mentions that Freeman seems to have "glommed" on to an argument from TIA. In both cases he is referring to this very same "political logic". I can see how it would be worrisome if Freeman, TIA, and I are colluding on this issue... it is even more worrisome if we arrive at the same conclusion independently.

Now, this "political logic" rests on three legs: The polarization of the electorate before the election (and thus the difficulty of producing large "shifts"), the very large turnout (which was not anticipated), and Bush's 3 million vote majority (which seems impossibly large).

Through our agreement, we have just increased the Republican margin from just over 3 million votes to just under 4 million votes - Bush's majority plus Kerry's overwhelming share of 1 million votes which "ordinarily" would have been cast but not counted.

Instead of providing an alternative to fraud which can explain the 2004 election, we have dramatically increased the threshold that must be overcome in order to explain that election WITHOUT FRAUD.

2) We have complicated our discussion of the exit polls. After all of the hypothetical response errors which we have discussed, we have suddenly agreed on a real source of error. We can even regard it as a form of response error, perhaps uRb (unintentional response bias - "I thought I voted..."). But, having agreed, we have also spontaneously eliminated one third of it in 2004. Not only doesn't the remaining "spoilage" even begin to explain the red shift nationwide in '04 but the "apparent blue shift" we would expect in locations where this source of error is reduced or "virtually eliminated", fails to appear as well.

In Florida and Georgia as we look deeper at individual counties where spoilage totaled upwards of 7% in 2000, none of the expected effects are evident:

Rise in relative turnout? Nope. Rise in Democratic plurality? Nope. And so on...

Does this surprise you? It surprised me and I was already convinced of fraud.

3) We have just provided a testimonial to Mitofsky (admittedly against his will). We have just produced a plausible explanation for much or most of the deviation in the exit polls for 3 of the last 5 presidential elections (oddly enough, not by rejecting "fraud" but by acknowledging a form of it). In the process, we have made the 2004 exit polls stand out all the more starkly.

And that's it.

I would expect "the mathematician as well as the social scientist" to at least take pause in light of the above. I'm willing to let that determine the apparent rate of spin...

[Febble]

106. No Nobel Peace Prize expected

if we ever reach agreement, it should probably go to you.

I respect your political arguments for a stolen election. They were my own, actually, though not as detailed. But seeing as pre-election polls in the rest of the world (e.g. the UK) were overwhelmingly for Kerry, and indeed Kerry was widely anticipated to win, Bush's victory just didn't make any sense to us over here. Even the Tories were rooting for Kerry (and expecting him to win).

I may have misunderstood some of your arguments - I find your style a bit oblique and I never really knew what the "game" rules were - on the rules I thought we were playing to, OTOH's computation based on the NES data seemed like a winner - yet clearly according to your rules, he lost. Well, we lost. I was playing too, though you might not have noticed. But I'm not concerned with games. I'm concerned with whether there was massive fraud in the 2004 election.

But what I think you are saying is that differential vote spoilage could have accounted for red-shift for years. I agree. I think it almost certainly did. We know it happened, and presumably some of those precincts were polled precincts. However, I don't see that it can have been the sole, or even main cause of red-shift over the last five elections, although it alone was enough to lose Gore Florida. But the net red-shift at precinct level in 2000 was actually very small. Whereas in 1992 it was large.

But what is also interesting about the vote-spoilage story is that it is one source of red-shift that WOULD NOT impact on the non-correlation between Bush's net gain in 2004 and precinct-level discrepancy, as presented by Mitofsky last week.

I've been saying this for quite a while, although obviously only recently in relation to the new finding. I certainly made the point with regard to the ESI finding - that it would say NOTHING about red-shift due to differential vote-spoilage in precincts where it had been going on for years.

So yes, I'll take pause - but in my defence, I'll say I'm there already. But I don't see how it can account for increased red-shift since 2000 without also accounting for increased vote-share since 2000.

[Febble]

107. OK, having paused

here is a plausible scenario:

Say spoilage reached a max in 1992, then there has been a steady decline in spoilage sense, so it was just rumbling along in 2000, enough to lose Gore the election, but with minimal impact on the polls. So in 2004 they try a new trick: vote-switching.

By this reckoning it makes sense: we say the discrepancy USED to be caused by vote spoilage, but as vote spoilage has declined (has it? That would be worth checking maybe - does spoilage from year to year correlate with WPE?) they turn to a different form of theft.

OK. But.
H
ere is where we run into a problem, firstly with the ESI finding, and now with Mitofsky's new finding, both of which found a zero correlation between the precinct level discrepancy and the Bush's increase in vote share. Now, if the the precinct level discrepancy is due to spoilage, it won't correlate with Bush's increase in vote share, because it would have happened in the same precinct as last time. So spoilage could account for the redshift in Ohio (although there was very little net redshift in 2000 in fact). But NEW fraud, the new vote-switching stuff we are postulating to account for the sudden rise in 2004 - that should produce a correlation between Bush's increase in vote-share and the size of the discrepancy. And it doesn't.

So we are left with the problem: if there was NEW vote-stealing going on, why doesn't the track it leaves in the exit poll correlate with increase in vote-share?

This is not spin. Or a game. It is a perfectly serious question. I'd like to know the answer.

[anaxarchos]

113. Honest answer...

We are biting off way too much at once....

Follow the logical thread I just had with TfC. The answers to most of your practical questions are in there. Residual voting rates have been fairly stable (officially between 2.0 and 2.5%) in modern elections until 2004. There is some "conservative inference" included because there is no consistent reporting (13000 voting jurisdictions, etc.). If I was "feeling controversial", I might argue that "spoilage" is, in fact, much larger than that.

The guys at CalTech/MIT Voting Technology Project are shills for the machine companies, but they are smart shills. They declared "victory" way too early but there is no reason NOT to believe that spoilage was reduced by one third in 2004 or that "1 million votes" were counted that would not have been in previous years. Interestingly, they only hold DREs responsible for half that gain. The rest is attributed to increased interest in the 2004 election (i.e. more people "looking").

As far as the exit polls go, the error is real. It is, presumably, reflected in every exit poll (but because it has been a 'constant' for so long, it makes me suspicious). I don't really believe it accounts for "all" of the red-shift in 3 of the last 5 elections. The magnitudes are correct but the geography is not. It must account for "much" of it, though... I think '96 is a special case and, in our discussion, will become a "red-herring". I have no problem contrasting the other 3 years with '04.

The mechanics of this type of "fraud" are subtle, imprecise, and decentralized. It is known what types of voting technology produces what kinds of error rates given certain maintenance schedules, voting processes, etc. Then those factors are "innocently" or "unintentionally" introduced. The Republicans teach Workshops on this (with the intention of "avoiding spoilage", of course). Often, it can be a budgetary process alone and it can happen at several levels of government.

As for the rest of your question, let's ask it again once we have common assumptions.

Finally, all physical problems are "games" or puzzles, by definition. Only meta-physical problems are not. They are "religions".

[Febble]

114. From the bottom up

Fine. I like puzzles. But I'm not really interested in winning or losing, only in seeing the puzzle solved.

I'd like to put some new data points into play:

Mark Lindeman has put Mitofsky's new plots on a website, here:

http://inside.bard.edu/~lindeman/slides.html

together with some explanatory text. You can click on the plots to examine them more closely.

Now, one approach to these plots is that they are fakes. I'd like us to assume that they are not - that they represent real data. I have good reason to think that they do.

I think they are compatible with a number of things, but incompatible with others. But I'd rather we started with a simple understanding of what they mean (and maybe they are worth their own thread).

My understanding is that if the Precinct Level Discrepancy (PLD to use Steve Freeman's term) is due to fraud, and if the effect of fraud is to boost Bush's share of the vote, then there should be a linear association between the degree of PLD and the degree of the "swing" (to use a UK term) to Bush. There appears to be no such correlation.

Now, the swing was calculated from 2000 as a baseline. In 2000 the overall PLD was small, although we know there was spoilage that may have resulted in some of the redshift (there was a small red-shift). So the challenge is: if, in 2004, an increase in red-shift and an increase in Bush's share of the vote was achieved by the same means i.e. some kind of fraud, how can we account for the finding that the two effects are not correlated?


(edited for grammar)

[anaxarchos]

116. We have just shifted again...

Would you prefer I continue on the residual vote issue or suspend that momentarily to address Lindeman's post (which is unrelated)? I will do either, while we are "at pause".

[Febble]

117. Sorry,

I'd ended my last post with a question, and I thought you were referring to that with your answer.

The two issues are actually relatedm because spoilage is one form of vote-corruption that would produce red-shift and might not produce a correlation in 2004 between swing and bias.

But by all means continue with the residual vote issue.

[anaxarchos]

120. Actually, quite the opposite...

In 2004 we have "corrected" one third of the vote corruption that we knew to have existed in 2000, quite apart from any "additional fraud". I don't see that either. Sigh... I guess I'll just have to read it.

[Febble]

126. Well I need your argument

spelled out in nice clear statements. Sorry. I'd like to know exactly what you are saying.

And what is it you are going to read?

Really anax, I just don't do oblique! I'm just not sure what you are talking about.

I think you are saying that one third of the vote corruption existing in 2000 has been corrected. Let me be crass here and assume that is what you are saying and extrapolate. If so, then any red shift due to spoilage in 2000 should be down to baseline in 2004. But we do, in fact have a large red-shift in 2004, which, if I read you right, is unlikely to be spoilage (or spoilage should account for less of it). So something new is happening.

But that something new, the thing that is producing red-shift in 2004, that isn't spoilage, does not appear to be correlated with "swing" to Bush since 2000.

But I may be talking about something quite different. In which case please disregard the fact that this post appears as a response to yours.

[Febble]

136. I posted a bad link

This has the relevant findings:

http://inside.bard.edu/~lindeman/slides.html

At the risk of being inadvertantly off-topic again, I'd argue that there are two ways in which undervotes and overvotes could selectively disenfranchise Democrats.

One is if, in a given precinct, more Democratic votes were spoiled than Republican votes. There is some evidence that this happens (the Harvard paper suggests that ballot design is critical) and that demographic groups with poorer education or non-English speakers (i.e. poorer people, more likely to be Democrats) may have more trouble with badly designed ballots. But that will show up in the exit polls. As will machines designed only to spoil Democratic votes (if there is such a thing).

However if: machines in Democratic precincts are less well maintained, or are nefariously set to randomly spoil votes, regardless of who the vote is cast for, then again, Democratic votes will be selectively lost, benefiting Republic - but it won't show up as a red-shift because the proportions of each kind of vote will remain as in the poll (thanks to TfC for pointing this out).

Which is why I am arguing that the plots shown in the link are consistent with any amount of fraud if the fraud was of the latter type. So your own argument would hold - Bush stole the election, because he couldn't possibly have won it. But the exit poll argument wouldn't. The exit poll discrepancy would just be an epiphenomenon.

In short, I am not arguing (and actually have never) argued against fraud. I am arguing simply that the exit poll discrepancy does not index it.

But I'd like to be educated more about the precise ways in which machine fraud/disenfranchisement could have been perpetuated. Because of course we don't have voting machines in the UK.

[anaxarchos]

139. Really Febble,

...you must learn patience. I didn't like the scenario you proposed so I'm thinking (and putting up shutters)... OK, that's enough.


The science of it:

It seems to me that we have a perfect laboratory for your various theories. We have vote corruption in 2000. We know the direction of it. We understand causality. We also have a great advantage in residual vote - it either exists or it doesn't. In 2000, there was a lot of it on a precinct basis or there wasn't any. There are many precincts which in 2000 had residual vote similar in magnitude to what you might expect from "fraud" in 2004. I don't know about the datasets (haven't really thought about it), but if you have enough data:

Redo your scatter plots. Use only precincts which had significant spoilage in 2000 and have very little or none in 2004. Redo your regressions. If you have correlation, you or Lindeman or Mitofsky (I'm unclear on whose work this is) can shout from the rooftops that you've proven your method. If you don't get correlation, then the scatter plots are shit. Toss them out... unless you are really, REALLY attached to them. Then you can look for counter influences...


The politics of it:

Since you are from the UK, the United States is unique among the world's countries (including South Africa). EVERYTHING in U.S. politics (and history) is about race... and EVERYBODY swears that it isn't. If you have enough data, start with African-American precincts.


The religion of it:

This is the extent of our chat, I'm afraid. I am about to get hit by Hurricane Wilma. EVERY time I talk to you or Lindeman, I get smacked by a hurricane. If I wasn't an atheist, I would think god was trying to tell me something.

[Febble]

142. I have patience to spare.

Good luck with Wilma.

Let me know when you are done with her.

And I know it is about race.

[kiwi_expat]

122. re. "The Game"

You have suggested, in other posts, that Rove did indeed get his 3 million "Evangelicals", but that they would not have been sufficient to give Bush a 3 million popular vote advantage (let's not say "win").

Are you referring only to fundamentalists who did not vote in 2000 but did vote in previous elections? Because there would also be a large group of fundamentalists (perhaps almost half of those who voted in 2000) who voted for Born-Again Al Gore in 2000, but chose Bush in 2004.

Thanks in advance for your clarification.

kiwi

p.s. Just for the record, I personally doubt that Bush got 3 million more popular votes than Kerry, but I think it is possible that he got slightly more popular votes than Kerry. I guess that makes me a "naysayer". ;-)

[Febble]

129. Do you mean

that you think it is possible that he got more electoral votes that Kerry?

I think that might be a typo. If so, I agree.

[kiwi_expat]

130. I meant popular votes - but I would also say that about Electoral votes if

we were talking only about those cast votes that could be detected in a count/recount - not including voter suppression etc.

I was originally trying to state very clearly that I think that Bush might have narrowly beat Kerry in the popular vote.


ON EDIT: To be totally truthful, my real opinion is that Bush POSSIBLY (narrowly) beat Kerry in Ohio; and PROBABLY beat Kerry in the popular vote - and maybe not so narrowly. But, I really don't care about the popular vote, and I wish we would not make such a big deal of it.

[Febble]

131. OK

That's interesting. I can't make anything produce a popular vote win, but I could still find room for an electoral vote win, at a pinch.

I'm open to offers though.

[kiwi_expat]

132. Please see the comment I added, on edit ,to post #130. n/t

[anaxarchos]

140. Looking at the Survey questions, eh?

Good, me too... Rove was looking for 3 out 4 million who sat out 2000 from the right. From a policy standpoint that was what the "turn-to-the-right" was all about. It really is significant that we haven't heard a word about constitutional amendments banning gay marriage, etc. since.

The theory was that he would freeze the rest of the 2000 vote (despite his right turn) and drop the fundies in like an alien invasion. I think he got them too.

I agree also that he got some erosion from the fundies that voted for Gore... But there weren't enough of them to overcome the turnout which was HUGE.

BTW, if you are reading the rest of this thread, the 3 million popular plurality that has to be explained is now nearly 4 million (due to the drop in the residual vote rate).

If believing Bush could have (outside chance) gotten popular parity or a narrow win (which is what Rove hoped for) or lost the popular contest and won the electoral votes makes you a "naysayer", then I'm one too. I wouldn't have been nearly so certain of fraud on the day after the election. I still wouldn't have bought in to Florida and I would still be pissed about all of the rest of the shit that they do in every election, but I wouldn't have been so surprised.

But 3 million (now 4 million)? Shit, it could be 30 million and it wouldn't get any sillier.

[kiwi_expat]

141. Possible 5 million fundamentalists voted Gore2000 and Bush2004 ??

"I agree also that he got some erosion from the fundies that voted for Gore... But there weren't enough of them to overcome the turnout which was HUGE." -anax

I tried to see if I could come up with some numbers that might be big enough to make a difference.

I used two pieces of data that I happened to have handy: the percent of Ohio (raw) NEP respondents who said they were "Born Again" and some unweighted NES-study figures for 2000 and 2004 for participants who said they believed the Bible was the "actual word of God". (The NES data was provided by OTOH, with a lot of cautions.)

564 of the 2042 Ohio NEP respondents called themselves "born again".

The 2000 NES (non panel) data showed 46% of the participants who voted in 2000 and believed the Bible was the "actual word of God", voted for Gore.

The 2004 NES (non panel) data showed 30% of the participants who voted in 2004 and believed the Bible was the "actual word of God", voted for Kerry.

It is important to emphasize that this data is unweighted and probably has other problems as well. But we can at least use it to see if fundamentalist Gore voters MIGHT have had a significant impact in supporting Bush in 2004.


So if, say, 30 million fundamentalists voted in both elections (and 46% voted for Gore in 2000 but only 30% voted for Kerry in 2004), that would mean 4,800,000 fundamentalists switched from Gore to Bush.

I'm sure that figure is way too high, but it does indicate that we can't dismiss the fundamentalists who voted for Gore.

Cheers!

[kiwi_expat]

143. More realistic: 2.4 million fundamentalists voted Gore2000 and Bush2004

I checked the Ohio raw NEP data to see how what percent of born-agains voted for Kerry. The answer is 36%

Also, a more realistic number of fundamentalists who voted in both elections is 24 million. Calculation available upon request. :-)


So if 24 million fundamentalists voted in both elections (and 46% voted for Gore in 2000 but 36% voted for Kerry in 2004), that would mean 2,400,000 fundamentalists switched from Gore to Bush. Still a VERY crude estimate, of course.


I hope you and your family and friends are OK!!

[Time for change]

76. I certainly agree with your main points

Spoilage has been going on for years, certainly since before the Mitofsky exit polls have been in operation, and they certainly have created red shifts. The degree of red shift that they have created is open to some question, but perhaps is as large as 1% or so, which would account for about half of the red shift in 88, 96, and 2000, but only about 20% of the red shift in 92, and even less than that in 04. So, you are quite correct that we should then see a relative blue shift when the problem is partly corrected, as was supposedly the case in Florida and SC in 2004. But that would be difficult to evaluate, I think, first because I'm not aware what the FL and SC polls showed in the other years, and second because the relative blue shift, I believe, would have been too small to be statistically significant.

In any event, I certainly do agree that a good portion of the exit poll discrepancy in 2004 was due to fraud, especially in Ohio, PA, and FL. As I said in another post, even if one quarter of the overall exit poll discrepancy was due to fraud, when added to other kinds of fraud that would not be detectable in the exit polls, that fraud very well could have been enough to swing the popular vote to Bush, and this would be even substantially more likely in Ohio.

But I'm still confused about the Florida 2000 thing. The Miami Herald recount showed 13,153 spoiled ballots for Gore and 12,158 for Bush. Certainly enough to make a huge difference in the 2000 Florida election, but the ratio is far less than what you are surmising. There were also nearly 24,000 ballots that had no mark whatsoever -- but what reason would there be for believing that these ballots were characterized by a substantially different ratio than the ones that were counted? I'm not saying you're wrong about your assumptions of distribution, but I just don't understand the discrepancy.

[anaxarchos]

80. Minor points...

Residual Vote rates (according to CT/M VTP):

1988 2.5%
1992 2.0%
1996 2.1%
2000 2.0%

There is also a minor controversy over what is "residual vote". My cut is that actual "spoilage" is perhaps .3% to .5% greater than that indicated by the table above. It is a minor point.

Either way, a pretty good argument can be made for "spoilage" accounting for "all" of the red-shift in '88, '96, and '2000 but NOT in '92 or 2004.


For total spoilage in Florida, Georgia, and South Carolina (among others), look here:

The Civil Rights Project, Harvard University - Democracy Spoiled: National, State, and County Disparities in Disenfranchisement Through Uncounted Ballots

http://www.civilrightsproject.harvard.edu/research/electoral_reform/ResidualBallot.pdf

There are some very nice county maps of those states included as well. South Carolina is not really at issue because we still have no report on residual vote in 2004. In Florida and Georgia however, it was "virtually eliminated".

The difference between the Herald's 39,000 spoiled ballots and Harvard's 181,000 is as I have said.

You will have to be more specific about what you don't understand in what I have quoted you. In general, half of all spoiled ballots come from black precincts, Hispanics are 5 times more likely to have spoiled ballots than whites, and spoiled ballots are many times more likely to come from low-income, Democratic, neighborhoods than anywhere else. I quoted the U.S. Civil Rights Commision, The Harvard Civil Rights Project, The Special Investigations Division of the Committee on Government Reform of the U.S. House of Representatives, and others. I cannot account for the Miami Herald (despite the fact that I read it every day).

I would not conclude, just yet, that "the relative blue shift, I believe, would have been too small to be statistically significant", as you have said. That is precisely where we are going (although maybe we won't get there today). What we are doing first is concentrating the error. Under 1% nationwide becomes 2.5% in Florida (and 3% in Georgia) and then, ...it is still distributed very unevenly, statewide.

[Time for change]

83. This is what I don't understand

1) You say that you believe that there is a good argument for spoilage accounting for all the red shift in 88, 96, and 2000. Yet, using your own figures from post # 62, we get a loss to Kerry of only 1.15 million net votes nationwide (as I note in post # 67), which is less than 1%. The total red shift in those years was about 2%.

2) Also, as I note in my previous post (76), when the Miami Herald did a recount of all the under-votes in Florida they obtained a ratio of Gore to Bush votes among the spoiled ballots that was considerably less than what you are postulating. I'm not saying that you're wrong, but I don't understand the discrepancy between what you are postulating and the results of the recount that was performed.

[anaxarchos]

84. We may be drifting "off topic"...

1) Look at the table. In '88, '96, and '00, we can make a pretty good argument that a residual vote rate of 2% to 3% (depending on how we count) yields a Democratic plurality of 1.3% to 2% - close enough... My post #62 is about 2004, a very different year. The claim for 2004 is that DREs dramatically reduced spoilage in 2004 (by one third).

Charles Stewart, RESIDUAL VOTE IN THE 2004 ELECTION, VTP Working Paper, February 2005

"Seventeen million more people voted in 2004 than voted in 2000, a 14% increase. Approximately one million of these “new votes” can be attributed to reforms in voting machines and administrative practices over the past four years." (Quotes in the original)

Available Here:

http://www.vote.caltech.edu/media/documents/vtp_wp21v2.3.pdf

2) I don't know how to answer your perceived discrepancy except to quote the research again... that puts us into an endless loop. I have a better idea.

Klinker has a blog called PolySigh (cute name), available here: http://polysigh.blogspot.com

You can ask him the question...

You can e-mail Christopher Edley directly at Berkley. He is the Dean of the Law School.

Not being a wise guy... just don't know how to resolve this.

[kiwi_expat]

86. As you say, the Miami Herald's recount was only of UNDER votes.

The subsequent newspaper consortium recount was of under and OVER votes. Gore did much better on the OVER votes. He got enough overvotes, showing unambiguous voter intent, that would have given him the Election.

Gore would have had a 46,000 margin of victory if (net) overvotes were included which were obviously intended for Gore/Bush but also showed a third party candidate. (www.gainsvillesun.com, November 24, 2001)

[Time for change]

89. I know that Gore did much better with the over-votes than under-votes

But I didn't think it amounted to anywhere close to 46,000. Are you sure about that? I couldn't find it at the web site you cited.

Anyhow, I didn't think that over-votes are typically thought of as spoiled ballots. The reason for them is entirely different than the reasons for an under-vote, and I didn't think that they would have the same association with poor and minority neighborhoods.

In 2000, the biggest reason for over-votes, I believe, was when voters using the butterfly ballot (Palm Beach County) weren't sure whether the Buchanan or the Gore spot on the ballot was the right spot for Gore, so they picked both of them. But there were only a little over 5 thousand of over-votes in Palm Becah County that were Gore/Buchanan. Then there were an additional 10 thousand that were Gore/someone else other than Buchanan (I don't understand what caused those), and there were about 4 thousand that were Bush/someone else.

Well, maybe there were lots of overvotes in other counties as well, but I don't know what would have caused a Gore preponderance in those counties.

[kiwi_expat]

92. Here is the entire Gainesville Sun article

Gainesville Sun

Saturday, November 24, 2001

Al Gore was the genuine choice of a majority of Floridians
By Mark Mayfield
Mark V. Mayfield is a Gainesville attorney.

'Bush wins media recount" was the headline, in various forms, in nearly
every newspaper after the National Opinion Research Center's (NORC) recount
of votes in Florida. The headlines imply that Bush actually received more
votes in Florida than Gore. Nothing could be farther from the truth.

Under any scenario where all of the votes are counted, Gore won. The only
scenarios where Bush won were those where significant numbers of votes were
simply not counted.

The Associated Press made the point clearly: "Under any standard that
tabulated all disputed votes statewide, however, Gore erased Bush's
advantage and emerged with a tiny lead that ranged from 42 to 171 votes."

Gore's margin of victory swells to over 46,000 if the overvotes are
carefully examined. Such an examination of the 113,000 overvotes shows that
more than 75,000 chose Gore and a minor candidate and just 29,000 chose
Bush (and a minor candidate). Common sense demands that we admit that most of these voters were not
supporters of either Patrick Buchanan or the Socialist Workers' Party.

Furthermore, many overvotes were entirely legal. They simply weren't counted
because a voter may have punched in Gore's name and written it down to be
certain the counter got the message. If all of Florida's counties had
error-checking machines in the precincts to prevent such overvotes, Gore's
margin of victory would have been beyond any doubt! The media consortium
paid practically no attention to these ballots. Why? To conceal the
strongest evidence by far that Florida's voters preferred Gore.

Beneath the deliberately misleading headlines is the inescapable fact that
Al Gore was the genuine choice of a majority of Floridians as well as the
victor by more than 540,000 votes nationally. As the Associated Press report
states, "In the review of all the state's disputed ballots, Gore edged ahead
under all six scenarios for counting all undervotes and overvotes
statewide." In other words, Gore got more votes than George Bush.

Gore won under a strict-counting scenario and he won under a loose-counting
scenario. No matter how you count it, if everyone who legally voted in
Florida had a chance to see his or her vote matter, Al Gore would be sitting
in the Oval Office today.

After Sept. 11, many seemed to feel it was their patriotic duty not to do
anything to call into question the authority of the commander-in-chief.
However, if we're supposed to support a war effort against an enemy that
hates us because of our free, democratic process, shouldn't we first ensure
that we really are living by that democratic process?

We've already witnessed the shredding of the Constitution in the name of the
war on terrorism (secret searches, indefinite detentions, military tribunals
with no judicial oversight, attorney-client privilege reduced), must we also
endure the dismantling of the very foundations of our democracy? It should
never be considered unpatriotic or treasonous to demand that our leaders
live up to the rule of law. The right to question the fairness and accuracy
of an election, even more so a Presidential election, should be sacrosanct.

Instead, we are witnessing the media at best giving short shrift to the
results of the recount and, at worst, deliberately misleading the public
about the results.

There were, at a minimum, serious irregularities in the election of 2000. At
worst, there may have been illegalities committed by a number of state and
national government officials. Yet all we hear are, "Get over it" or "It's
time to move on."

While we must surely try to return our lives to the best semblance of normal
in the wake of Sept. 11, we must not let those events go without recourse.
We must fully investigate and understand the events that lead to the fiasco
that became the Florida recount of 2000. If that points to the fact that the
wrong man is occupying the White House, we must face that and deal with it
appropriately. We do our principles and ourselves no good by hiding or
"spinning" the facts, ignoring or misstating the truth.

George W. Bush was not the choice of a majority of Floridians (or of all
Americans). Therefore, to "restore honor and integrity to the office" he
should either resign the presidency, allowing Al Gore to take over
immediately, or he should be removed. This would be unprecedented, no doubt.
It would likely cause great turmoil too. The country will survive, as it has
after other crises, because it is the right thing to do.

[Time for change]

93. Very interesting -- thank you

I wonder how many of those 75 K over-votes actually had Gore's name written in, and the same for Bush. I don't think that even the Florida Supreme court would have allowed the counting of over-votes where the name wasn't written in. I do know that there were plenty enough of those for Gore to win Florida easily, but I'd really like to know how many there were. I'd also like to know how they happened. If there were only 5 K Gore/Buchanan (i.e. butterfly ballot related) votes in PBC, where (and why) did the other 70 K come from?

[kiwi_expat]

95. The only over votes that could have been included in the vote count...

were the over votes that had a punch/mark for the SAME candidate as the candidate whose name was written in.

Gore would have won Florida by about 100 votes if those were allowed.


I don't know where all those over votes with minor candidates came from. But it does sound as though inexperienced voters might have caused a disproportionate number.

[yowzayowzayowza]

98. Actually the NORC study found that ...

Gore would win a full count of all overvotes by any standard:

http://www.democraticunderground.com/articles/01/11/p/12_gorewon.html
http://www.washingtonpost.com/wp-dyn/articles/A12623-2001Nov11.html

...yet in all cases Gore's winning margin was less than the * declared victory margin. Our assembeled numerological geeks: wadar the oddz of that? Things that make ya go Hmmmmmm.....

[kiwi_expat]

103. Gore narrowly won all legally-acceptable overvotes scenarios.

The additional (net) legally-acceptable under and over votes would have given another 600-700 votes to Gore - and thus would have had him win by 100-200 votes.


The Washington Post article neglects to mention that
Gore's margin of victory swells to over 46,000 if the overvotes including minor candidates are included. Unfortunately, those overvotes (e.g., showing Kerry and a minor candidate) could not have been legally included in any vote count, because the intention is not totally unambiguous.

[Time for change]

104. That may be but

If non-felons who were close computer matches of felons hadn't been illegally purged from the voter roles, Gore would have won by tens of thousands of votes.

And as Kiwi pointed out, if one was to count over-votes meant for Gore (though that couldn't be proven) he would have one by thousands. And certainly thousands of the Buchanan votes in PBC were meant for Gore.

[Febble]

41. And there is Jim Knapp's

Analysis of provisional ballots by demographic in Cuyahoga.

Rejected provisional ballots would show up as red shift. And by analysing the numbers of accepted provisional ballots, Jim Knapp has nice corollary to the evidence that black voters were more likely to be issued with provisional ballots.

http://www.dailykos.com/story/2004/12/15/143151/76

And again, if this has been going on for years it wouldn't show up in the scatterplots of redshift by swing, yet would contribute to redshift.

[Melissa G]

34. I don't believe he is using this single point to account for other than

some of the deviation.

[Time for change]

35. Ok, that makes sense n/t

[Febble]

39. But what's wrong with

spoilage accounting for some of PLD? I think it probably did, and if it was in precincts where it has happened for years, it wouldn't show up in the correlation between swing and redshift. It would just redden the polls slightly every year.

We know it happens. We know it happened in 2004. It would be weird if some of the PLD wasn't spoilage.

[Time for change]

49. I agree most definitely

It's almost certain that spoilage accounted for some of the PLD. All I was saying was that, it is also almost certain that it accounts for a relatively small proportion of the PLD. That was my only point.

Why is that important? Well, for one thing it means that you need to look elsewhere to explain the rest of the PLD. It also means that spoilage alone didn't account for enough to swing the 04 election to Bush. Also, spoilage is not considered fraud per se, so if that's all we have it's not terribly convincing. And, accounting for as little PLD as it probably did, it's a little surprising to me that it would be statistically identifiable.

[foo_bar]

40. it seems ready made for the internet

1. PLD significantly higher in swing states.



But certainly fraud is a more plausable explanation for this.

The most complete explanation is a combination of fraud (and its invisible cousin Incompetence, as in Palm Beach's butterfly ballot), exit polls being historically inaccurate, the assorted smoking guns in Figure 1, and so-called rBr being more prevalent among independents and Dems than broken-glass repubs.

Slightly off-tangent, there's a semantic conflict in the completion rate theory: the people refusing to be polled aren't Reluctant Republicans, they're Reluctant Bush-voters (reluctant precisely because they aren't Republicans but crossover Dems or indies, or mild-mannered Republicans if such a thing exists). I realize it's unpopular to conjecture that Gore voters could actually pull the lever for Bush in 2004 ("even my fundie minister voted for Kerry!"), but that would be as good a reason as any not to brag about it.

2. PLD lower with the use of paper ballots.

Since it's a similar discrepancy to the rural vs. urban WPE (which correlates highly with hand-counted paper (if Figure 1 is hand-counted ballots and not optical scans)), I'll buy Febble's explanation. It probably doesn't explain 100% of the story, despite our universal desire for a grand unification theory.

3. No other Mitofsky exit poll has been as far off as in this election, where we have secret vote counting by computer on a massive scale.

This appears to be a circular argument. You suspect the machines because of the exit polls, but you suspect the exit polls because of the machines? There's plenty of valid reasons to Fear the Machine, but the breakdown OTOH did of Vermont (the biggest "red shift" in the exit polls) had the Optiscans favoring Kerry by a double-digit spread over paper ballot precincts (since optical scan precincts correlate highly with population density, which in turn correlates with Democrats; the equivalent fraudster explanation would have Kerry rigging the machines).

As for the first part, here's a Mitofsky exit poll when he ran VNS in 92:

Eleven of twelve states had an error in Clinton's favor; seven had errors on the margin of six points or greater.

http://www.mysterypollster.com/main/2005/01/the_war_room.html

But don't believe your lyin' eyes:

Why is it that exit polls since 1988 have ALL overstated the Dem vote?

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=114922#115013

4. Mitofsky's report (from earlier this year) presents little or no evidence that bias accounted for a good proportion of the PLD, as Freeman meticulously and lucidly explained.

I agree, but absence of evidence isn't evidence of absence. The fact that Mitofsky hasn't presented a compelling argument for why he screwed up isn't a mystery to me.

5. Higher PLD where Republican Governors are in charge is additional evidence.

Then why were the highest State Level Discrepancies solid blue?

6. So is higher PLD where voter complaints are more numerous -- though I would like to know more about what kind of complaints these were and what kind of statistical analysis was done on this. I do find it a bit confusing that Ohio wasn't near the top with regard to these complaints.

I believe this is the source:
http://www.msnbc.msn.com/id/6364287/

It was a call-in line (866-MYVOTE-1), not to be confused with VerifiedVoting's 1-866-OUR-VOTE (!). The EIRS one does have Ohio at the top followed by Florida and New Mexico (in terms of incidents/population). That's where the skeletons are buried, not the states with the highest exit poll differential. Ralph Nader and DUer IdaBriggs tested the latter hypothesis, and it failed:

Before:

CONCLUSION: These numbers are NOT within a reasonable margin of error, and further investigation is required.

http://www.invisibleida.com/New_Hampshire.htm

After:

Sometimes "odd numbers" reflect reality, and in New Hampshire, apparently that's just the way things are: larger population centers appear to be trending conservative (Republican), while the rest of the state appears to be trending liberal (Democrat). There are a few more things that should be double checked (just as a matter of eliminating the obvious), but it appears the will of the majority of people was appropriately represented in New Hampshire.

http://www.invisibleida.com/

[OnTheOtherHand]

42. great post -- a note on Republican governors

Both are true: the WPE is higher in states with larger Dem margins, and in states with Rep governors. But only the former finding looks to be statistically robust; Freeman seems to have a slightly different list of governors in order to even get one statistically significant result. (The four Rep-gov states with the top WPEs, according to my list, are CT, VT, NH and NY; I'm just not liking these as epicenters of vote fraud, although NY would be a good choice if one could pull it off.)

[Time for change]

45. Are you saying that Freeman misclassified the party of

some of the Governors in his analysis?

[OnTheOtherHand]

56. actually, I think probably not, but I still can't replicate the result

His histogram shows 22 Dem govs with median WPE 4.5 and mean WPE 5.0, and 28 Rep govs with median WPE 7.9 and mean 6.7. I agree with all that (except to note that he isn't using the "IM WPE" measure, so that could be confusing). So if there is any difference in our data sets, it should be fairly subtle.

But SPSS gives me a t score of -1.2 (whether or not equal variances are assumed -- actually -1.20 or -1.17), which is nowhere near p = .04. I half think Freeman may have copied and pasted the swing-state result from several slides earlier and then forgotten to change the number. Or possibly he used a non-parametric test to exploit the fact that the medians are farther apart than the means. (I thought a Mann-Whitney might do it -- better, but still only z = -1.6.)

At any rate, I don't think the result is statistically or substantively robust (and in general Secretary of State would seem to be a more appropriate variable).

[Melissa G]

96. Secretary of state is generally appointed by the governor ..no?

That is is why one would use the Gov's R or D label as the appropriate variable.

[Time for change]

48. The internet?

1. Yes, I agree that there was most likely a combination of fraud and exit poll bias. But a 50-50 split would mean enough fraud to swing the election -- see my post 44.

2. There were only 5 urban precincts with HCPB. Not much power there. I agree that this finding was probably not statistically significant, but still I think it's suggestive. It certainly doesn't fit Mitofsky's characterization as evidence against fraud, since there's no difference in PLD by voting equipment.

3. I don't see the circularity of my argument. 2004 is the year that we have the biggest ever red shift. It is also the year that we apparently have the biggest ever opportunity for fraud, in that a good portion of our votes were counted in secret, by convicted felons who just happen to have strong ties to the Republican Party. I think that we need to strongly consider that this isn't a coincidence. I didn't say that I suspect the exit polls (though I'm willing to admit the possibility of some exit poll bias). I'm saying that I primarily suspect fraud, and that this fraud accounts for a larger red shift than we've ever seen (and yes, I know that there was a red shift in 92 that was about three quarters of the red shift we see in 2004.)

4. I agree that the absence of evidence for exit poll bias in Mitofsky's original report is not proof that there was no exit poll bias. However, what we know is that the red shift that we see is explained by either fraud or exit poll bias or a combination. Therefore, any lack of evidence for exit poll bias has to be seen as a plus for fraud.

5. I think that OTOH explained this one - though I didn't understand his comment about having a different list of Governors.

6. Nice map. What I mostly meant by my comment on this was that it would be nice to know what kind of voter complaints were involved. Of course, we already know from EIRS analysis that there were a great many vote switching complaints involving touch screens in Florida.

[foo_bar]

64. the presentation seems to be geared towards the internet

None of the graphs have sources, except the (arguably refuted interpretation of the) response rate on slide 31. Some of the tables are sourced, but not the blow-up of the E-M report with MSPaint circles drawn about the "absurd" parts. The accompanying text explains that he's the farthest thing from a statistician, and that's why we should trust his judgment on these matters (TIA made a remarkably similar pronouncement, but it doesn't prove that TIA is Steve Freeman).

But a 50-50 split would mean enough fraud to swing the election -- see my post 44.

I agree. But that isn't what Freeman is arguing:

How can one explain this eight million vote discrepancy between the Election Day exit polls and the official count? Either the exit poll data was wrong or the official count was wrong.

http://www.sevenstories.com/book/index.cfm/GCOI/58322100420010

Not much power there.

Agreed.

2004 is the year that we have the biggest ever red shift. It is also the year that we apparently have the biggest ever opportunity for fraud, in that a good portion of our votes were counted in secret, by convicted felons who just happen to have strong ties to the Republican Party.

And yet the polls were far closer in blackbox Florida than papertrail Vermont and NH. Why not use the Washington Redskins as a predictor of rigged machines? Fraud is fraud, exit polls are exit polls.

Therefore, any lack of evidence for exit poll bias has to be seen as a plus for fraud.

There's no lack of evidence for poll bias (Berinsky, Adam J. 1999. "The Two Faces of Public Opinion." American Journal of Political Science 43(4):1209–30. online; Curtice, John, and Nick Sparrow. 1997. "How Accurate Are Traditional Quota Opinion Polls?" Journal of the Market Research Society 39:433–48 abstract), just lack of Mitofsky's ability to explain it to lay persons convincingly. Perhaps he's tried: http://poq.oxfordjournals.org/cgi/content/citation/62/2/230

What I mostly meant by my comment on this was that it would be nice to know what kind of voter complaints were involved.

Found it:
http://www.philly1.com/CARTER-BAKER-PROB-STATE.xls
http://www.philly1.com/MYVOTE1-STATE-REPORTS.pdf

[Time for change]

66. Thank you for the voter complaint data

Regarding your citations of poll bias, I would say that exit polls are very different than public opinion polls in general. In most public opinion polls, probably the main source of bias is faulty memory, for whatever reason. With exit polls, that type of bias would seem to be of minimal consequence. So, I think it's a whole different ball game, and that's why I think that the best historical references that would be relevant to the likelihood of bias in the 2004 election would be Mitofsky's own previous exit polls.

With regard to your comments about VT and NH having more red shift than FL, it appears that there is something having to do with the way the polls are set up in the far northeastern states that causes them to have a large red shift. Since there would be no particular reason for fraud to be limited to the far northeast, my guess is that something is causing exit poll bias in those states. The association is truly phenomenal (VT, NH, CT, RI, NY, DE). All that is lacking to make it a virtual clean sweep is MA, ME, and NJ, and I think that NJ and MA had pretty hefty red shifts as well.

[kiwi_expat]

74. How was the weather in the North East on Election Day? n/t

[Febble]

81. Well the interesting thing

about that phenomenon, which I posted on DKos about a few months back, is that it runs counter to the claim of higher fraud in Republican strongholds at precinct level (which I think is refuted, but which some people, including TIA still claim). So while redshift appears to be greater in Blue states, the precinct level data are claimed to demonstrate that redshift is greater in Red precincts.

This was claimed to counter the hypothesis that Bush voters were shyer in Kerry territory. But that might still apply at the state level. On the other hand there are a huge number of potential variables that are collinear with state margin, including Geography, and of course Urban-Rural difference.

http://www.dailykos.com/story/2005/4/6/8028/83645

[kiwi_expat]

87. I have never understood "shyer in partisan territory" hypotheses.

Why would anyone think that Kerry or Bush voters would be shyer in partisan precincts? No one knows what answers the respondents give to the questionnaires. The questionnaires are filled out by the respondents, in silence.

[foo_bar]

100. the respondents don't know it's confidential yet

They're still at the stranger-with-a-clipboard stage. When they refuse or walk across the street, the interviewer jots down their perceived age/race/gender to be weighted out later.

If I were an exit poll guru, I'd see what shoes the nonrespondents had on (e.g., Reluctant Birkenstock Wearer).

[Febble]

102. And too many Birkenstocks

on your responders might indicate sampling bias.

[kiwi_expat]

105. .... especially if the interviewers wear Birkenstocks too. n/t

[Febble]

101. I don't know.

but it was a hypothesis that USCV considered,and was discussed on Mystery Pollster IIRC. USCV called it something like "rBrMPC" (reluctant Bush responder in mixed political company").

It does appear to be the case that WPE was higher in blue states, and USCV (and I think Freeman) allege that WPE was higher in red precincts. I don't see that either is an indication of fraud in itself. It could be behavioural. It could be mathematical. It could be collinearity with some other variable.

[foo_bar]

99. heavy reading

I found a textbook called "Analyzing Elections", to be published next year. Half of Chapter 10 is about Mitofsky's rise to internet infamy from 1967-2004:

Exit polls were first conducted in 1967 at CBS by Warren Mitofsky. Originally exit polls were used by the statisticians to get a better idea about what to expect in the election, they were seen as guideposts rather than as consequential sources of data.

(37 years later...)

The problems with the exit polls highlighted the difficulty with having one source for exit poll data provided to many clients and the potential for leakage of the results without statistical interpretation. The early results also placed a greater weight on the responses of women, who were more likely to support Kerry. While these weights were corrected later on election day and before the results were used for analysis or projecting outcomes, they increased the popular perception that Kerry was winning the election.

http://www.nyu.edu/gsas/dept/politics/faculty/morton/book/MortonElectChap10.pdf (the manuscript's free until it's in-print)

I guess the professors write the history. But wait, it gets worse:

2. What is the difference between the ratio estimators used by Edison/Mitofsky to project states and the linear estimation procedure used by UNIVAC and Ray Fair? What are the advantages and disadvantages of the two approaches? Explain.

(hint on page 244: "...the engineer in charge of Remington Rand’s new product development division went on camera and apologized to the viewers: “A mistake was made. But the mistake was human. We were wrong and UNIVAC was right. Next time we’ll leave it alone.”")

For a less pedagogic cite, "A Review of Recent Controversies Concerning the 2004 Presidential Election Exit Polls", commissioned by the SSRC (Walter Mebane of butterfly-ballot fame is on their election commission). Compare and contrast:

This information would prove especially important in 2004 when NEP was working with a brand new version of these systems. The data also provide information about estimated point differences between the candidates, and they are often used by the sponsoring organizations to "characterize" Election Day on the early evening news. The news organizations use this NEP information to plan the final allocation of their resources for on-the-air interviews, the last version of the scripting of election night coverage, or the early coverage in newspapers without making any specific calls. But the early data were never intended for general distribution to the public.
...
The data that are deposited are weighted in a complex way that accounts for some nonresponse adjustments, the demographic characteristics of the respondents, and, most importantly, to the actual outcome of the election in a state or to the national outcome. When the outcome of past elections was not as close and the introduction of new voting technology was less common, this was a satisfactory procedure. But in the context of the leaks in 2004, the competing theories about why and how this happened, and the quality of the exit poll results, data weighted to the actual outcome of the votes was no longer a satisfactory dataset for many consumers of exit poll results.

Eleven weeks after the election, NEP and it sponsors released a 77-page report on what happened. This is an unusual analysis in that the same people who were responsible for the exit polls and the projection apparatus also did the evaluation. Furthermore, it is complicated in a way that many post-survey evaluations are by the fact that some information is essentially unknowable. This is especially true when one of the concerns is nonresponse, and there is no information from the nonrespondents to analyze. As a result, there are some sections of the report in which there is an extremely detailed level of disclosure about what the exit poll data show, but in other parts of the report there are only hypotheses about what might have been the cause for a particular observation. These hypotheses can guide future experiments in exit polling methodology or even direct changes in the methods, but they cannot explain in a strict causal sense what happened in the 2004 data collection.

The report actually discusses three issues, two of which received relatively little attention prior to its release. The bulk of the report is devoted to an analysis of how the exit poll interviews were used to estimate the actual support for George W. Bush and John Kerry , and it acknowledges that there was a Democratic bias in the poll. It also contains a discussion of a variety of survey weighting issues, and a statement about technical problems with the computer system that caused two disruptions to operations during the course of the evening. There is an unprecedented amount of information available in the report, much more than has ever been released before. This has prompted considerable discussion of the NEP report on the Web, but it has not silenced the critics of the NEP and their data.

http://elections.ssrc.org/research/ExitPollReport031005.pdf

[OnTheOtherHand]

43. quick note on paper ballots

E/M reports 40 paper-ballot precincts with mean WPE -2.2 and mean abs WPE 11.2. Just fiddling with normal distributions of WPE, I find that to get a large enough mean abs WPE, the standard error of the mean should be around 2.2. For the 35 rural/small-town precincts (mean -1.6, mean abs 10.5), maybe closer to 2.0.

So, for the univariate comparison, the confidence interval surely overlaps with everything but maybe touch screen. For the comparison within rural/small town, it will overlap everything -- there just isn't enough statistical power to say much.

[Time for change]

44. Ok, not enough power to give statistical significance to paper ballots

Still, it's suggestive I think.

Though I also feel that in trying to decide whether fraud or exit poll bias account more for the PLD, I'm impressed more with the evidence FOR fraud than the evidence AGAINST exit poll bias. So any evidence against exit poll bias is icing on the cake, so to speak IMO.

AND it's also important to realize that if there were equal amounts of both, that means that there was enough fraud to swing the election.

PLUS, it's also important to consider the fact that there are a number of kinds of fraud that wouldn't even show up in PLD. Consequently, exit poll bias could have accounted for three quarters of the PLD, and yet there still could be enough fraud to swing the election -- and that's with regard to the popular vote, let alone Ohio.

But I must apologize for getting way off track from the subject of the post that I'm answering.

[Melissa G]

51. Yes! this is what I wa s trying to say about C in Febble and Eomer's

Discussion..Thanks TFC!!!

[OnTheOtherHand]

55. yes, all these things (pretty much)

I think the high PLD in some touch screen precincts (as evinced for instance in the rural+small town split by the mean of -8.0 vis-a-vis the median of -4.8) is suspicious in part because we are already justifiably suspicious of touch screens. I think there is quite a good chance that any apparent "effect" in the exit poll is spurious, but I don't see much point in debating how good the chance is!

Even if DREs were basically clean in 2004 when they didn't actually malfunction (and, of course, "malfunctions" could be a manifestation of yowsa³'s "plausible deniability"), I would still distrust them. (Not to limit the discourse to touch screens; in some ways they stand out in the E/M table, but in others not -- and I don't unconditionally trust any voting technology, including hand-counted paper ballots.)

[autorank]

82. ***179,000*** SPOILED BALLOTS --- Why understate our case?

Greg Pallast has been so on target with election issues it is amazing. He is the authority I cite. Plus Anaxarchos above is making key points as well that MUST be listened too.

What is the point in understating our case? It's a devastating one. Florida WAS STOLEN. There is no question about that.

Look at the snip, then read the article:


http://www.gregpalast.com/detail.cfm?artid=327&row=1

Florida's electorate is 11 percent African-American. Florida refused to count 179,855 spoiled ballots. A little junior high school algebra applied to commission numbers indicates that 54 percent, or 97,000, of the votes "spoiled" were cast by black folk, of whom more than 90 percent chose Gore. The nonblack vote divided about evenly between Gore and Bush. Therefore, had Harris allowed the counting of these ballots, Al Gore would have racked up a plurality of about 87,000 votes in Florida--162 times Bush's official margin of victory.

179,000 SPOILED BALLOTS IN FLORIDA

[kiwi_expat]

88. Consortium recount showed (net) 46,000 spoiled ballots favoured Gore.

(see my posts #85 and #86)

Cheers!

[autorank]

90. along with the 50,000 Floridians pulled from the voting rolls by "felon

purge" software created by a little sompany headed by former M.C. Vin Weber. I'm shocked, just shocked.

Let's do the math

49,000 spoiled ballots in the recount
179,000 spoiled ballots not counted
50,000 removed from the rolls by felon purge

Wow, we're talking about some real numbers here.

Remember this, 1-2% a year lost in Federal elections due to spoiled ballots in mostly minority districts!

How are the elections down there?

[kiwi_expat]

91. The 46,000 are a sub-set of the 179,000.

All the 179,000 spoiled ballots (under and over votes) were counted by the newspaper consortium in 2001. The under votes netted slightly more for Gore. But "an examination of the 113,000 overvotes shows that more than 75,000 chose Gore and a minor candidate and just 29,000 chose Bush and a minor candidate". - Gainesville Sun, November 24, 2001 (see post #92 for the entire article)