Calculatus Eliminatus

I don't think it is in the exit polls.

Fraud, I mean. Let me explain in one post what I have been trying to explain on many.

As most of you will know, I have been an exit poll skeptic for some time. My background, though in all sorts of odd things, is largely in social sciences, so while I was highly suspicious of Bush's apparent win (and devastated) last November, and seriously wondered whether the exit polls were an indication of massive fraud (especially in the light of the Ukraine story that followed close in its heels) I did not share the assumption of many DUers that the discrepancy could not have been due to polling bias. I am not a statistician as such (i.e. it is not my primary speciality) but I have had a rigorous statistical training, and use statistics heavily, daily. I also teach statistics at university level. And I know enough about statistics to know that the "margin of error" refers to "sampling error" - in other words, it is the margin of error that could have occurred simply by chance. If I measure my son's height today, and again tomorrow, I may find he's grown. And he sure is growing. But because he wriggles a bit, I have to allow for random wriggles. A jump beyond the margin of wriggles will mean he's grown. A jump (or a drop) inside the margin of wriggles will just indicate he's wriggling.

OK. So I knew that the discrepancy between poll and count wasn't chance. So what was it? It could have been fraud. It could have been bias in the poll - i.e. the poll might have had a biased sample. And I was perfectly prepared to believe in sampling bias. I meet it every day. You can do what you can to minimize it, but you can't avoid it. So when I read the E-M report in January, and saw that apparently the discrepancy WAS at precinct level (not because of poor selection of precincts) I realised the only alternatives were: fraud; or biased sampling of voters. And although the report was frustrating to read (short on geeky bits like standard errors, F values, probability estimates, degrees of freedom) the apparent findings that "redshift" was associated with precinct characteristics likely to make adherence to strict random sampling protocol difficult, I was convinced that sampling bias probably played a role. However, nothing in the E-M report indicated how much, except the assertion that the evaluation had determined that "non-response bias" (which CAN mean that one group of voters was more likely to agree to participate than the other, but can also mean that one group of voters was more likely to selected than the other) accounted for the discrepancy.

So far, so not very good. Unsatisfactory in fact. In order to establish whether non-response bias was sufficient to account for the exit poll discrepancy a number of analyses need to be done (and though I would like to say I'd thought of all these months ago, I didn't and they've been growing on me slowly - however, the first was obvious).

The first would be a proper test of the polling bias hypothesis. A multiple regression analysis needed to be done, in which all the precinct/interviewer characteristics hypothesised to be contributors to the discrepancy were entered into the same model, ie. not a series of separate analyses where WPE in one kind of precinct is compared with WPE in another. What needs to be done is an analysis that takes into account the fact that in some precincts several of these factors may be present together, and may even interact. If such a model could be shown to account for the discrepancy - or if, to put it differently, after accounting for factors likely to give rise to polling bias, there was still a residual discrepancy, then one might deduce fraud. And actually, even if they did account for the full discrepancy, one might also wonder whether some of the variables were proxies for fraud variables. So that wouldn't be conclusive, but it would be of interest. In my paper, a few months back, I called on this kind of analysis to be done.


But we also neeed fraud hypotheses to test. I am aware of two, both important.

1. This was a test of a hypothesis originally formulated by USCV (aka NEDA) as the "Bush Strongholds have more Vote-Count Corruption" hypothesis. This in itself seemed an odd hypothesis (why, a priori, would we expect more corruption in Bush precinct strongholds?). However, a later formulation expressed by USCV, and suggested to me by Josh Mitteldorf, was that if fraud was responsible for the exit poll discrepancy, you would expect a "bunching" of precincts with highly discrepant poll results at the "high Bush" end of the spectrum. In other words, precincts that ought, if the vote count had been honest, to have been in the centre of the spectrum where the distribution of precincts is fattest, would have been shifted Bush-wards by fraud. As the distribution of precincts by Bush's vote share is roughly bell-shaped (actually there are more moderately Bush precincts than moderately Kerry precincts, but more extreme Kerry precincts than extreme Bush precincts) then if fraud was randomly distributed across the spectrum - a thickish swarm of precincts from the middle of the plot should move to the right (literally and metaphorically) and give rise to a positive correlation between discrepancy in the poll and Bush's share of the vote.

Unfortunately this test was complicated by the fact that the traditional measure of exit poll discrepancy (WPE)does weird things in relation to the way the vote count is distributed. I devised a measure that I think does a better job - and Mitofsky performed the correlation. The hypothesis was not supported - there was no linear tendency for the discrepancy to be greater at the Bush end of the plot.

Plots can be viewed at the links in this DKos diary by HudsonValleyMark (Mark Lindeman):

http://www.dailykos.com/story/2005/5/24/213011/565

We cannot, however, conclude from this null finding that fraud was not responsible for the total discrepancy. Simply that it does not seem to have been randomly distributed through the precincts. Maybe it was concentrated at the Kerry end.

2. So here is a second hypothesis, this time formuated by ESI for Ohio. This hypothesis says: if fraud was responsible for the exit poll discrepancy, as well as for Bush's apparent increase in support in the election (the presumed purpose of fraud) then precincts with greater "redshift" in the poll ought to have a greater shift to Bush in the count. In other words the two effects should be correlated. but how do we measure Bush's gain? One way of doing it is to baseline it from 2000, a year in which the exit polls were relatively accurate (and therefore a year in which fraud, if it occurred, seems to have played relatively minor role in Bush's vote count - after all, Gore won).

ESI performed the correlation for Ohio, and found no association between what Brits call "swing" to Bush and redshift in the poll. But the trouble is that states, in exit poll terms, are small, and therefore do not give you a lot of statistical power. And to demonstrate a null you need A LOT of statistical power. In fact you can never demonstrate a null. What you can do instead is to demonstrate that if there was an "effect" it was, to a given degree of probability, less than a certain size. And the power in the Ohio study would have left fairly wide confidence limits for the "true" association between redshift and swing to Bush. (We have yet to see a proper peer-reviewed report of the ESI study - it is apparently in the pipeline).

So Mitofsky repeated the ESI analysis on the entire dataset, and it was the results of this analysis that he presented at the debate with Steve Freeman in Philadelphia last week. He presented it in the form of two scatter plots, which are posted here, by Mark Lindeman, with some informative text. If you click on the plots, you can examine them more closely.

What they indicate is that there is no discernable association between redshift and Bush's performance relative to 2000 (remember, a year in which the exit poll was fairly accurate, although there was a small net red-shift). Confidence limits are not given, but geeks among you can ballpark it given the precinct N which is 1250. The limits are fairly tight.

So what's with the Cat in the Hat?

I DON'T want to prove that Bush won a fair election. I WANT to prove that he didn't. But I think it is very hard, given that plot, to see view fraud as the explanation for the exit poll discrepancy. But we don't NEED to demonstrate that fraud was the cause of the exit poll discrepancy. We need to demonstrate that fraud was the cause of Bush's victory. Actually, as far as I'm concerned, we don't need to demonstrate even that - what we need to do is to demonstrate that he did not win a fair race. And he didn't. The race was unfair from beginning to end, from the moment he stepped into his Poppy's size nines, to the attack ads, to the lies about WMD, to the felon purges, to the voter suppression tactics, to the rationing of voting machines to Democratic precincts in key states, to the monkeying about with regulations on voter challenges, to the monkeying about with provisional ballot regulations, to the refusal to expedite a recount in Ohio, to the media mockery of those who doubted as "tinfoilers" (a new word to me) - and maybe to the electronic corruption of the actual vote, made absurdly possible by the absurdly insecure software installed on the voting machines.

But if we want to find that certain something - the smoking gun, the evidence that Bush, far from spreading freedom and democracy is the president of a democracy only in name - we need to FIND OUT WHERE IT'S NOT.

And if it's in the exit polls, it's bloody well hidden.


On edit: link added, and I should also say that my name is Elizabeth Liddle, for those who don't know, aka Lizzie.

And if it was in the Machines- and all the other vehicles-- the machine share (%) needn't be a large share-- of the "fraud". SO looking at the exit polls-- the share of fraud would be smaller. Since the fraud was shared- between the machines and suppression- etc.

Its late-- and Im going to crash-- L8tr

Roj

Febble said:
>We need to demonstrate that fraud was the cause of Bush's victory. >Actually, as far as I'm concerned, we don't need to demonstrate >even that - what we need to do is to demonstrate that he did not >win a fair race. And he didn't.

Let's see, while elections are necessary in a democracy, real democracy is defined by the ultimate power being in the people, the government ruling only through the proper consent of the governed. If that consent was not unequivocally achieved, doesn't the government need to go back and get some consent? The government needs to PROVE that it has the consent of the govern, not simply ANNOUNCE (data and methods of counting strictly secret) what the RESULTS are and expect obedience since, if we are in a real democracy, the government is the servant of the people.

So, to correct febble, we don't need to prove fraud, we don't need to demonstrate a lack of a fair race,the Government needs to PROVE that it has the consent of the governed. It fails to do so.

if the people were not the master, but instead the servant or servile subjects of the world's only superpower, then the people have a burden to show NOT fraud, NOT unfairness, but merely to show an incorrect RESULT. Whether it is by accident or by intentional fraud matters little except in criminal proceedings.

SECOND CORRECTION OF A FEBBLE-ISM

Also, it is a logical or at least strategic mistake to go looking for single causation in a complicated world, or to expect one-stop proof shopping with exit polls. I'd be very happy to use exit polls as one piece of evidence in an overall case about the elections. Even if you were right Febble and they don't, themselves, prove fraud, they are still relevant evidence.

I see that scientists and statisticians tend to get all hot and bothered if there's any breath left whatsoever in the "other side" of the debate. They want to prove things to a certainty. Lawyers never have a case in which there isn't another equally talented lawyer on the other side making good arguments based on the inferences available to them. that doesn't mean that the defendant won't be executed, that the judgement won't enter for $500,000 or such. In other words, justice and proof in an evidentiary court never require the elimination of all seemingly plausible hypotheses that go against one's conclusion, but some interpretations of science and statistics militate this way.

"JUDGMENT", the word both for the ultimate operation of the mental faculty as well as the final document in a court case, involves either reconciling or resolving competing inferences in favor of one side over the other. If there were a conspiracy to commit election crimes, it would never be admitted, the only way to prove it would be to look at all of the evidence simultaneously, with the only conclusion then being conspiracy.

The defense in a conspiracy case always takes the micro view. They ask: What in this little fact PROVES conspiracy? The answer is always NOTHING. Because it doesn't prove anything. the scientific approach to life nicely fits with the defense approach on a conspiracy case because it is reductive and wants to focus in on little facts in great detail. applied to the criminal defense context, no one would ever be convicted because it is always plausible that Smutherdude committed the crime (Smutherdude commits most crime in America as a whole). At the end of the day, the alleged fact that there, for example, may have been a reluctant bush responder effect will just be left on the ashheaps of history by the jury rendering a judgment.

It's my Judgment that the exit polls provide good evidence, despite some slender reeds that Mitofsky and Co still cling to.

Demand that a debate between "Mitofsky and Freeman" be televised prime time on cable news? If I where him, I would try anything to restore my credibility with the people I work for (cable news) and the American people.

with Freeman in public. Where does this stop? Now he is at fault for not demanding it was televised?

Do you know that he didn't?

wasn't on Television. But my question is if they are blaming the exit poll discrepancies in 04 on the "Reluctant * Responders" ,how will they make the exit polls in lets say 06,08 more accurate, will the beat it out of the RBR'S

but to make the exit polls more "accurate" I expect they will rely more heavily on the incoming vote counts. But, you will rightly say, that won't make them strictly exit polls.

The point is, they never are. If you really want an exit poll to audit the election (and it is a lousy way to audit an election) then it needs to be designed as such.

The E-M one is largely designed to converge on the count. The exit responses are only part of the data they use. This is not new, nor unique. The UK ones are the same and no-one regards the early projections as anything more than "a bit of fun" (catch phrase of the guy who does them).

To get a better exit poll you'd need to pour far more resources into the actual exit data collection, not rely on other data streams.

A number of things could be done. First would be to have at least two interviewers at all times, more at busy times, working in shifts. That way, coverage throughout the day could be much more even. Second would be to train interviewers properly (i.e. not just a short telephone training), emphasising the importance of true random sampling and warning them that they must not depart from their Nth voter protocol, and to note all refusers, with age, race and gender. Third is to ensure sample adequate sample sizes - aim for at least 100 in each precinct. Fourth is to make sure that the interviewers are not seen to represent any body likely to be perceived as politically biased - so do not station the interviewers near MoveOn workers, do not have them labelled with TV network logos etc. Make them look as professional as possible. And lots more.

But it will still be a lousy auditing tool. You need paper ballots with proper random audits and full recounts if the audits turn up any irregularities. You need secure custody of the ballots between count and recount too. And lots more on that as well. But you know that. We do a good job in the UK if you want a good model (though we still sometimes get a lousy PM).

to own a polling company that has to do exit polls in a country that uses electronic vote counting and tabulating machines. Even when you get the exit poll numbers within the MOE which is probably what Mitofsky did, He has to explain why he didn't. I would flat out retire if I was him.

Mitofsky did not address Freeman's points. I could tell you a lot more about this... other ways he tried not to engage in a pm in you have any interest.
Best,
Melissa

Freeman addressed Mitofsky's either. All reports I've read say they rather talked past each other. I think the format might have been non-ideal. I gather they both presented their own agendas, then talked at each other.

A good debate needs a good format - preferably with seconders, and an opportunity for each to sum up. Also a really clear motion.

But these are not excuses. I'd like to know what each think of the other's points after opportunity for due reflection. The purpose of this thread was to explore the implications of Mitofsky's new information for Freeman's thesis. I've sent him the link!

BTW - I have a lot of respect for Steve, and my impression is that it is returned.

in fact I am saying is that even if the exit polls do not indicate fraud, there is still an excellent case.

And I do not believe they do, in light of the most recent analysis released by Mitofsky. But I specifically do not say that we do not "need" to prove fraud. I say we do not "need" to prove that fraud caused the exit poll discrepancy. Which we can't if it didn't.

Really, Land Shark, I think you have misunderstood my basic point here. You may disagree as to whether the exit polls prove fraud or not. My point is simply that I actually think that at this stage the evidence is pretty overwhelming that fraud was not the cause of the exit poll discrepancy.

THIS IS NOT THE SAME AS SAYING THERE WAS NO FRAUD IN THE ELECTION.

Actually, it is not the same as saying there was not overwhelming fraud in the election. One particular form of fraud would not show up in the exit polls at all. Which is why I bothered to post this. It may ACTUALLY TELL US A MORE LIKELY PLACE TO LOOK FOR FRAUD.

If it isn't in the exit polls. And I believe it is not.

I think if you re-read my post with your excellent lawyer's eye, you will find I am not saying what you think I am saying, and it may even (may) help us find the evidence you need.

What we do know is that the exit-polls show Kerry won the election.

What we do know is that there is ample evidence of fraud, and the exit-polls are a part of that evidence since it shows Kerry won the election.

We do know Febble is paid by E&M.

We know E&M are caught by the short hairs and are squirming to be released.

It was a lucky stroke that the leaked raw data came out before it had been over-weighted by the manipulations E&M applied the day after election day.

What else is there to know except that the election was stolen? The exit-polls can't fix it, can't repair it, and can't undo the stolen elections. But the exit-polls do show it was stolen. While that may be too much responsibility to have laid at the feet of the exit-pollers, whoop, there it is. E&M need to get over it and move on.

No, Febble isn't paid by E&M, at least not on an ongoing basis. She has a full-time job elsewhere. (EDIT: whoops, my bad, I was remembering a different sig.)

No, it wasn't a lucky stroke that the _weighted_ data were _released_ on CNN.com before they had been adjusted to official returns.

And no, in most professionals' opinions, the exit-polls don't "show" Kerry won the election.

If you have another quarter, you can play again.

"I see that scientists and statisticians tend to get all hot and bothered if there's any breath left whatsoever in the 'other side' of the debate. They want to prove things to a certainty."

That is a fundamental misinterpretation both of social scientific method and of the extent to which most analysts of my acquaintance see "no there there" in the exit poll debate. You are entitled to marshal the arguments however you like, but it bears notice IMO that the bulk of expert opinion -- on this particular issue -- is against you. You may find that fact difficult to finesse. (Of course I'm not positing that majority expert opinion is necessarily right.)

If I'm not mistaken, prosecutors sometimes have found that they are better off not emphasizing -- perhaps not even presenting -- marginal evidence whose refutation casts a shadow on the entire case. Television journalists, too, have found that presenting "killer documents" of dubious provenance can backfire. Please be careful.

Perhaps you realize that ON THE OTHER HAND? There are always arguments on the other side? This includes the most one-sided debates.

Given this, pounding the inferences on the other side is not necessarily constructive.

Now, a good lawyer or law firm will definitely want to anticipate the other side's arguments, and flesh them out before the day in court. But in no case would the other side's arguments be made sarcastically or such... Certainly not publicly.

Yes sometimes ambiguous evidence is avoided in court, even by both sides. Sometimes the juries wonder why.

Sometimes weaker claims or refutable claims are presented purposely expecting to lose, figuring that when the jury comes to compromise they will throw out the claim not really wanted and keep the one desired. Thus the jury's compromise can be the plaintiff's win. Sometimes.

But OTOH you entirely miss the point when you make it personal because my arguments are not based at all on exit polls.

ON EDIT: In court we often have the "battle of experts", so it's really nothing to have experts "against" me. The exit poll debate would be no different. You (OTOH) need to realize that courtroom persuasion and political persuasion and statistical/scientific persuasion are different species of persuasion. (I know science makes claims to superiority in method, I've a degree in science myself) Even assuming the majority of statisticians would be against a position I take in court (again, I'm not arguing exit polls in reality), I'm going to get equal time for my view, and the juries will go for justice and the jury instructions, not for scientific theories (if they are different). Even if you and Febble were correct, if you can't get ME to understand why you're right, what chance do you have with a jury?

you did write: "It's my Judgment that the exit polls provide good evidence, despite some slender reeds that Mitofsky and Co still cling to." Sir, I don't know what basis you have for making that judgment, but it is not shared by most survey professionals of my acquaintance. I don't think that is making it personal, but if it is, so be it.

It may be that your side's arguments are so intuitively appealing that the jury would side with them even against the bulk of expert opinion. But it may not be. My guess is, it would turn out a bit like Inherit the Wind. (And I think it is perilous to assume that the average jury is predisposed to think that polls -- even exit polls -- are intrinsically accurate. I like my chances of persuading them much better than my chances with you -- not that I would want to argue against you in court, least of all on the flaws of paperless DREs.)

Here's the scientific truth: Confessions are often highly unreliable, and people confess to things they didn't do.

Now, try that one in court. Or in public opinion.

Now, I'm not JUST pointing to the chance of a "wrong" result from the scientific point of view. I'm also pointing to forks in the road like the following:

(1) OTOH hypothetically states an objection in the following form: Wait a second TIA, you skated past this step here, but because of the possible inapplicability of assumption X that underlies this step, we can't assume that this step is proven but you've skated past it!

Here, the truth may still be on the side of the person you are criticizing, but you're pointing to a chink in the armor that could possibly be a downfall.

If you want to identify a "class of argument" like you did above, it may faciliate understanding, at least by me.

There are others. But i do gotta go right now.

and I'm still thinking.

If Bush's gain and the exit poll discrepancy had the same cause - fraud - the two effects should be positively correlated.

They aren't.

What you need to provide for your client is a similar sentence to refute it. It's not easy.

P.S. I am the client. Represented by a lawyer and firm far better than I.

P.P.S. I disagree that bush gain and exit polls always have to be positively correlated (see my message below in the form of Q & A).

but it's a bloody good argument too.

I'll look at your Q and A

If so, it's an excellent start, although it seems more like a thread on why exit polls wouldn't show evidence of fraud than why they do.

I predict that people will react to it in at least two spirits. Some will engage it seriously to try to figure out whether and how it could work. Others will sit back, reassured that you somehow have shown that the exit polls prove fraud by explaining why they don't.

And that is OK. People in this life believe all sorts of things based on arguments that experts consider insufficient. Some of the beliefs are more accurate than others, but often we cannot tell which are which -- and even many of the inaccurate beliefs are innocuous or helpful. I don't start threads on "why I think Bush won" because, frankly, I don't even want to win that fight, and I don't think it is very useful, unless we learn something from it that helps us fix the election system (or, failing that, understand the world we live in). Your new thread might do that.

See you over there....

We need to start thinking like election criminals.

Otherwise, we end up with proposals like "the paper trail shall be the ballot of record and shall in all cases trump...." That sounds good, and perhaps is good enough, but we have to consider that by making it trump in all cases, the election criminal need only focus his efforts on ONE place. Professor Douglas Jones has pointed this out generally, mentioning that in one case in LA with such a recount-type law they focused on the recount mechanism only, fixed it, then demanded a recount and VOILA, result reversed.

(or engineers, or entrepreneurs....) Yes, your point is very important.

The myth of the silver bullet dies hard. We need to think about complex systems with lots of people involved. No voting method is a panacea in itself (which of course isn't to say that voting methods don't matter).

EDIT TO ADD: And in thinking like criminals, or ecologists, or whatever, we aren't just thinking about 2004 -- we're thinking about 2006, and 2008, and maybe even 2040 and beyond.

the following:

Remember who, what when why and where. Why did I say in the primer that I had millions of people help me?

Because (among other reasons) tens of millions of people are ready to argue the inferences hard for their favorite political party "team", and on top of that, all of these millions of people know:

WHO they want to benefit: Prez candidate
WHAT they want to benefit with: Votes
WHEN they need the benefit: November 2
WHERE they need the benefit: swing states, esp. OH, FL, NM, NV, PA
WHY they need the benefit: Electoral college and popular vote

So, not counting foreigners, there are 50 million or more biased and highly motivated individuals on each side of the political wars, each NOT NEEDING A SINGLE INSTRUCTION FROM THE "COMMAND STRUCTURE" OF THE RESPECTIVE CAMPAIGN TO KNOW EXACTLY WHAT IS NEEDED AND WHEN.

The only question left is not motive, not target, but HOW. We don't need a conspiracy or an order from on high, just an opportunity and a method.

I've been trying to think like a criminal for months, and look where it's got me! Everyone thinks I'm a criminal....

You finessed it much better. Well done!



(edited for missing word)

I think that's a little unjust. We are bothtrying to be scrupulously fair about this. We are not acting for the defence. We would both, I think, prefer to be acting for the prosecution. Think of us as the poor guy in your office who has to probe the weaknesses in your client's case, not the guy you actually face in court.

If I had to defend the case that the exit polls proved that there was no fraud (bear with me, I do not believe this) the argument is very simple.




Well, you get my drift. I'm not very good at this. But what I am trying to say is that the latest analysis gives your opponent a rather better argument against fraud rather than it gives you an argument for it. Stack that up with copious academic works on the problems of non-response bias and measurement error in survey data, and I think you have just got yourself a proportionally spaced-font memo.

Use it if you want. But to me, it's worth checking out that memo pretty thoroughly and so far, to me, it looks like a fake.

So, to turn to your earlier point, where I think you are implying that these things should not be publicly rehearsed - well, you may be right. But we don't have much option. It is on forums like this that we have an opportunity to probe the case for holes. Several posters have suggested ways in which Mitofsky's alibi for the polls could be busted (and to give myself credit, I've been fairly active in soliciting them and trying to work them through logically - it's a dirty job, but someone's got to do it). But I have to say, I'm not convinced yet, or anything close. Mitofsky's evidence certainly suggests that something OTHER than fraud was responsible for a major part of the exit poll discrepancy. And once that's conceded, the case looks pretty thin.

On the other hand (boy, this is catching...) if we bite the bullet, and say, well, perhaps fraud was NOT the cause of the exit poll discrepancy, THEN perhaps we can develop hypotheses, given that we still have every reason to believe it was a filthy election, that will lead us to evidence that WILL stand up in court.

(Edited to remove snarky comment. Sorry, I'm still a bit cross.)

Q: If a person is going to engage in election stealing, they will want to make it look like a real election, right?
A: Right.
Q: And a real election has both ups and downs for each candidate, does it not?
A: What do you mean?
Q: I mean precincts in which the candidate beats expectations and those in which she does not, some by a hair, others by a more significant margin.
A: Ok, yes.
Q: So, if a candidate simply won across the board in every precinct all through a state including areas controlled by each party, in the context of a tight race that would look suspicious wouldn't it?
A: Yes, it would.
Q: So the fraudster refrains from "taking too much" then?
A: Yes
Q: But the fraudster simply MUST, no doubt about it, give his candidate a NET benefit.
A: Yes, of course.
Q: Would the fraudster want to make the discrepancies, if detected, look like real voting behavior?
A: Sure, or like accidents.
Q: Correct. Assuming one sees the discrepancies, if the fraudster has any ability or power whatsoever to do so, he or she will make it look like the fraud is something else, so they don't get caught, don't you agree?
A; I suppose if they could, they would.
Q: Naturally. So that would include letting the opponent win a few he seemingly shouldn't win, right?
A: It could.
Q: In fact wouldn't that very control give independent experts, independent statisticians, and the loyal troops of the candidate the fraudster wishes to benefit (and the political operatives of the same) ALL KINDS OF AMMO with which to pound those who would expose the fraud?
A; OK, I see your point, but you'd have to show a net benefit for a given candidate.
Q: Yes, you would, but the very secrecy of the ballot means not being able to connect ballots to voters. So, it becomes dicey to show what the "real" election showed, so "net benefit" becomes a somewhat obfuscated issue as well.

One key is therefore in distinguishing, if possible, deliberate obfuscation as an expected fraud tactic with the normal fog of data.

On edit. So, I expect a certain amount of data suggesting "He went that-away" that is misleading. Can you statistically detect these kinds of things? It is certainly the most common thing in the world for criminals to cover their tracks, destroy evidence or plant false evidence to keep them from getting caught.

is the one about net benefit.

What is plotted in Mitofsky's first graph is Bush's counted vote share in 2000 against his counted vote share in 2004. We have no a priori reason to think the 2000 vote share was particularly inaccurate as the mean WPEs (PLDs) in that year were close to zero. But we don't know whether the counted vote share in 2004 is fraudulent or not. We know Bush increased his average share of the vote in 2004 relative to 2000, but there is variance - sometimes it increases a lot, sometimes a little.

OK But we can get an average. The regression line through the middle of the swarm of precincts tells you the average swing. If you find Bush's 2000 vote share on the horizontal axis and go up till you meet the regression line then turn left until you hit the vertical axis, you can read his "predicted" vote for 2004. In other words what his vote would be in 2004 if his increase was at the "average". Now, for any given precinct, this may not be accurate. For some precincts vote in 2004 is way lower than "predicted" by the regression line. In others it is way higher. The distance of each data point from the regression line is called the "residual" - it's the amount by which Bush's vote-share change for that precinct is different from his average vote share change.

Now, your lawyer is suggesting that to mimic real behaviour, fraud should be jittered a bit. There will be natural variance in "swing" (useful British term for change in vote share), some natural variance in the error rate in the poll, but also some designed variance in the amount of fraud. Some precincts will have none; some will have some; some will have a lot. But it's spread about a bit.

Here is where the statistics comes in. Some of the variance in swing will be due to sheer swing. Some of the variance in PLD will be due to sheer variance in PLD (error in the poll, of various kinds). But, if fraud responsible for some of both (and the more fraud there is the greater the swing will be relative to where it would have been without fraud - that's what fraud is for), some of the variance between swing and PLD will be shared. The amount that is shared will tell you the amount that is due to fraud. And there is a statistic (the "R squared") actually tells you the proportion of variance in PLD that is shared with swing. And the answer is .009%

Now that is a somewhat meaningless figure. We do not know the true answer to that many decimal places. But we can calculate its confidence limits. I've done it elsewhere but I can't recall the thread. I've got to put my son to bed, so I won't repeat it now. But the upper confidence limit is still very, very small.

Now there may be ways round it. I've done my best to list the loopholes, and there may be ones I haven't found. Maybe we can work them a bit larger. But that very low upper confidence limit on the percentage of shared variance is going to be key evidence for the defence.

But the R-squared figure you say is .009% you say we can't know it to "that many decimal places". But it has only ONE significant digit, so it would seem that the one significant digit would count, the zeroes are just placeholders.

And doesn't .009%, according to what you say above, representing the shared swing and PLD, constitute an initial calculation of the "fake evidence" I referred to in the "net benefit" post above?

Perhaps if we start from a list of things that MUST be true of a fraudulent election we can get somewhere faster:

1. The method MUST provide a net benefit to the favored candidate, at least in the targeted Electoral states, and is EXTREMELY LIKELY to provide a net benefit nationwide.

2. If the method involves directly distorting/changing the actual reported results as opposed to suppression of voters prior to actually voting, the method must use actual votes as an input or else pollbooks will be way too far away from actual counts. (can't have 1000 votes in 500 voter precinct, but can have 503 voters in a 500 vote district since officials will blithely conclude that 3 voters "forgot" to sign)

3. If computerized the method would BE EXTREMELY LIKELY TO use a randomizer of some sort, in order to avoid detection by well known mathematical tests for randomness.

4. IT WOULD BE VERY LIKELY to mix it up a little, providing some benefits for the opponent as well as the favored candidate, giving something for other supporters of the favored candidate to point at and scream bloody murder.

5. Maybe this would be good as its own thread. Really, I would love to see a "living" document as a lead document that could be edited to incorporate suggestions over time as it gets improved.

I put it in percentage terms to save using as many zero.

But worse - the actual R (not the R squared) is NEGATIVE. "Insignificantly" negative of course, but negative. In other words, the relationship represented by that R is actually in the wrong direction - Bush did worse in precincts with greater discrepancy.

But the point of the "signficance" level is this: it means that the true relationship is so small that we cannot tell whether it is positive or negative. A lot of rot is talked about p values, and about null hypotheses but that is what this one means, simply.

What you do next with a null finding (a finding in which your conclusion is so small that you cannot tell which way it goes) is you compute confidence limits. What we are interested in is the upper confidence limit of the relationship. This is not "proving a null" - it is saying, if the effect we are postulating exists (that swing and redshift arise from the same cause and are therefore positively correlated) then there is a 1 in 20 (or a 1 in 100) that the effect is of this size. We can then calculate an upper limit on the effect size that is likely, for a given level of likelihood.

I like your list. We need a new brainstorm list. Why not start it as its own thread. Seriously, for months I have been wanting to get past the "the exit polls prove it was stolen!" "no they don't!" argument so that we can actually figure out what is plausible within the constraints of what we know. And now is a good time, as that plot of Mitofsky's is really very informative.

Your four is interesting. I need to think about that. Randomizer (3) is not a problem - random is easy to detect. Non-random would be more of a problem if it was the right non-random. Agree with 1. 2 is interesting too: BillBored and Yowzayowzayowza have argued for tiny amounts of plausibly deniable fraud everywhere as a way to escape detection.

Cheers

Lizzie

because the exit polls are reading from a confluence of many diverse types of voting systems, some of which cancel each other out or mask the others' impacts.

Ask yourself the question Febble: What combination of voting systems and fraud could produce the data actually observed? Luckily, we know the voting systems, so that part is solid.

But the most solid part of this whole analysis (the voting systems and their properties) can be a statistician's weak spot that keeps them in the dark as to the full universe of possibilities, but political campaigns definitely know the systems as to absentees, the type of technology etc because they are working the absentees and training their pollwatchers for problems. I'm unaware whether this ignorance of voting systems is true of anyone here on this board, just making this point as to its statistical relevance in terms of generating theories about what can and can not happen.

Fraud is not a constant? Therefore trying to use fraud as a measurement or a number in the calculations would never work?

That would explain why the masters at numbers would never grok the idea that fraud wouldn't show up in their numb world?

but if the field has 50 different species in it, or at least fifty sections containing say 30 or so different species and subspecies, many different diseases will be necessary to wipe the whole field out.

But this diversity of species, like diverse election voting systems, creates a lot of noise and interference when we look at "the effect of disease X on all 50 parcels". On the other hand, if we were looking only at the one affected parcel of land among the 50, the data would be clearer.

So, you can have the nastiest bug lay waste to one or more sections, but if you are using a NUMBerical approach, then the killing of that section will be masked by ALL KINDA "NOISE" from other things happening, or even the SAME DISEASE having a somewhat different effect on a different strain, variety, subpecies or species in that area. (Perhaps the disease even affects plants more in the shade than in the sun, creating an effect within a section.

I think our number crunchers might in some ways be expecting the exit polls to sing with a strong voice, when what we are listening for is a signal of extraterrestrial "life" that SHOULDN'T be there amongst all the background "noise"

BOTTOM LINE: In a diverse election systems' environment, a dviersity of approaches are necessary if more than one state is in play. These diverse approaches at the national level create noise that interferes with seeing or hearing the evidence clearly, because it's mixed in with so much irrelevant and interfering junk.

On edit: don't hold me to numbers, but we should expect something like 25% of precincts are affected. The rest is masking noise, whether deliberate or accidental. It would be more helpful to take a state by state approach.

Depending upon which host the disease attaches itself to, the disease mutates itself into the desired form.

In nature numbers can be used by scientists to describe the possibilities of a disease. But when a mutant is injected into the system all bets are off.

Election systems are not natural anymore. Depending on which machine is used, where it is located and who is the programmer, the natural order can be diseased with impunity. There is no natural order to the election system.

What now controls the possibilities in our election system is, in essence, the Mad Scientist. Witness the way the numbers were altered in the final exit-polls. The uncorrupted numbers before the final showed Kerry as the winner but the mutant injection by the Mad Scientist, E&M, altered the final output.

We can trace backwards from that final output and find where the mutant was introduced - as has been done. But there is no way the natural order can be used to describe the final outcome of the entire election. It will have to be done at the minutest level. Unfortunately, those numbers are in the sole possession of the Mad Scientist.

let me get us straight about what "number crunchers" might expect, if I count as a number cruncher which I guess I do.

I would certainly not expect the exit polls to sing with a strong voice. I never did. Some number crunchers did - the ones who thought the song was fraud - Freeman, Simon and Baiman, TIA; I was the boring number cruncher who said they wouldn't. That there was loads of NOISE in the exit polls (quite innocent noise, including bias) that could be mistaken for the song.

Now, perhaps you think I am trying to have my cake and eat it; having said that the exit polls needn't be telling us there was fraud, I am now saying that the exit polls are telling us there wasn't. Well, I plead half guilty. What I think is, as I've said above, is that the exit polls put fairly serious constraints on the kind and extent of various sorts of fraud, just as you have put constraints on other kinds (adding too many votes to a precinct because the poll books wouldn't match up, etc).

So to get to your main point:

I agree, it's going to be diverse. I never believed the lone hacker story, which means I never thought the lone hacker story was a good argument against the "well how could so many people have been involved and no-one let on" argument.

I think it's got to be diverse. Most robust things are (it's almost the definition of robust, ecologically). And if it's diverse it's got to be done on a "need to know" basis - no-one person must know about the whole plan. Everyone just thinks they are doing a bit of local fiddling. Maybe they are even told that what they are doing is legit - a check on voter fraud maybe. I dunno about that part. I'll stick to numbers.

You are right in that the noisier it is, the more hidden it will be. But I think there is a misunderstanding here about the nature of noise. Statistics is all about signal to noise ratio. If fraud is the signal, it needs to be hidden in the noise. But we are postulating a very large signal (if we are postulating theft of the popular vote - I can think of plenty ways that theft of a single state could be hidden). And while the amount of noise in the exit poll itself is conveniently large, so is the number of precincts. This means that signal to noise ratio is going to be good in a national sample - there is a lot of statistical power to detect the signal.

And we can compute the power, and it is large enough to detect, on my ball park estimate, a signal of maybe a couple of hundred thousand vote switches, maybe even less. Certainly millions. Now it's possible that your fraudster got lucky, and the signal was masked by noise that just happened to jump in the lucky direction and cancel it out. But I frankly refuse to believe that any God would be that crass, although I could imagine that your master fraudster did not anticipate those scatterplots.

However, if the fraudster DID anticipate the scatterplots, he (assuming it was Rove) would have realised that the noise would not be enough to cover the massive fraud he planned. So he actually needs to do intelligent masking, like those noise cancellation gadgets you get in fancy cars that take every sonic frequency band and copy it exactly out of phase. I certainly haven't figured out how that could be done, yet, but I'm working on it (shame Rove won't be hiring in the near future).

OK, finally, back to diversity: Given that the exit polls have a lot of power to detect the fraudulent signal, I'd minimise actual vote switching, which will show up in the polls as a correlation between shift and swing. I'd do a bit, because it's efficient, and doesn't show up in the poll books. To maximise the amount I could get away with, I'd concentrate it in precincts where Bush wasn't doing too well. But I'd be pretty sparing with it. It's going to show.

I'd roll out as much voter suppression as I could, because that's not detectable. I'd also make sure more Democrats got issued with provisional ballots, then I'd reject them. That would show up in the exit polls as red shift, but not necessarily as a correlate of swing if I'd been shafting the same voters in the same precincts by similar scams for years. I'd also be tempted to tamper with the new electronic machines to make them randomly fail to record a vote for president - I'd be worried that where the old levers and punchcards, which for years had been a good free source of extra margin (because I'd made sure the crappiest machines went to the poorest precincts for years) had been replaced by new machines, Bush would suffer. So I'd program machines in predominantly Democratic precincts to randomly miss votes, regardless of who they were cast for. This wouldn't show up in the poll at all, and if it was spotted, could be blamed on stupid Dem voters. Obviously I'd be deleting Bush votes as well as Kerry votes, but provided I only triggered the program the machines in predominantly Kerry precincts, I'd still be ahead. In fact, thinking about it, that would be a pretty good scam. I'd just have to hope that no-one was checking up on the undervotes.

I'd also let my imagination rip in Ohio, concentrating my efforts on particular counties (so as not to affect the exit polls too much) and make damn sure any recount was conducted according to my own rules. In a final, beautiful touch, I'd bribe Edison-Mitofsky to screw up the exit polls on purpose, so that everyone went looking for fraud in the wrong place. I might even set this up by getting Tom Feeney to commission a vote-switching program from Clint Curtis.

knew which precincts were going to be exit polled (or even some of them) that would be enough to hide the true breadth of the fraud. And that goal would be, I would think, highly achievable for any committed and well funded party.

On edit:
check out <http://www.votersunite.org/info/SnohomishElectionFraudInvestigation.pdf>
specifically pages 13-14
note the noisy bell curve for paper ballots for both parties and the twin peaked smoother curve for electronic ballots, in the same race, same precincts.

The twin peak is some evidence of a "phase shift" What harmonic vibrations or cancellations come to the mathematical mind?

On the other hand, I could make sure that my exit-polling would pick precincts that would cause mutations to the poll that I could later use to destroy arguments that would use the exit-poll as evidence. Or make sure my exit-polling was done in such a way to accomplish the same ends.

'Twould be especially successful were I to be the only holder of all the evidence.

for a while (to general hilarity) that it might be the unpolled precincts that had the fraud.

However, there is still a bit of a problem, because not only do we now that the redshift was at precinct level, we also know that the selected precinct vote counts were fairly good predictors of each state as a whole. The discrepancy between the vote count in the selected precincts and the vote count in the whole state was small, and, if anything, slighlty blue-shifted. i.e. Bush did, in general, slightly better in the polled precincts than the unpolled, although of course this varied from state to state.

I'll take a look at your link. Fourier analysis is something I do (and I was a musician in what seems like a former life....)

in 55 you say the fraud could be in the unpolled precincts (the fraud could be in the exit polls too, in whole or in part). That would be a strong additional reason to resist disclosure of everything (sense of a skeleton in the closet oops better not go there)

On the other hand, the other things you say above as being required would be nowhere near impossible if some of the exit poll precincts were known to the fraudsters, or if the exit poll was subject to some type of non-sampling error.

All inferences and possibilities should remain open until the evidence definitively closes them.

I think my current sense is that your diversity idea has the strongest legs.

It is possible that some grand statistician-fooling algorithm was used to hide a big heist, but seeing as it's taken quite a few quite good minds to figure out how to start to look at the exit poll evidence, I'm not that convinced (although it is certainly possible that Rove's hackers are better statisticians than any of the statisticians that have looked at the data so far).

But AFAICT, every method will leave a fingerprint. Add too many votes and the poll-books will look suspicious. "Spoil" too many votes (more than usual) and people will start to look (and they are). Shift too many votes and it will show up in the exit polls - which might be OK, as long as it doesn't become apparent that red-shift in the exit polls is correlated with advantage to Bush. And if you avoid the polled precincts, then it will start to show up as poor precinct selection in the exit polls. The trouble with any crime is that if it is to be effective crime it's going to have consequences - and the consequences will tend to show up.

So if I were the master criminal, I'd commission a bit of everything, using different experts for each technique. Some of them wouldn't even know it was fraud. Some, like the BoE of Franklin County might just think they were doing a good job of rationing a scarce resource (voting machines) by allocating them on the basis of past turnout (pah!). And I'd keep everything just under the statistical radar, and ensure plausible deniability.

That way there will never be a "smoking gun" - and, with luck, no big prosecutions. Everything will be separate - if one scam is exposed, it won't lead to another.

Such a scheme would make truly massive theft much harder (I'll leave it to other to judge how much harder), but it could certainly swing key states (I think myself, that the popular vote was stolen in the usual way - by telling lies). But it does mean that there probably isn't a single piece of evidence that is going to send Rove to jail.

But I'll keep working on that algorithm.

actually second and third... given how noisy the paper curve is, I'm not sure what to make of the twin peak under electronic. With something like, umm, 650? precincts being analyzed, individual percentage-point bins are pretty susceptible to random noise, I would think. But if you or your brother has the source data, it would save some trouble importing it from the 40-page PDF. I do understand the thinking behind the twin peaks idea.

this was not planned in Washington state, someone with a fucking sense of media humor and a little cultural TV literacy planned this. (The cafe featured in the Twin Peaks tv show is actually in King County, --one can go get a bite to eat there-- not Snohomish county where the DREs were.)

Not brother, but brother-in-law Dr. J. H.

I'll see what i can do about a little data, but I'm on the road right now.

I would just like clarify some ways in which your explanation above does not rule out or even cast doubt upon fraud committed in the following ways (which I'm sure you wouldn't disagree with):

1. Electronic deletion of votes from Democratic precincts, in proportion to the pre-existing vote count (such as perhaps in Cleveland).

2. Electronic addition of votes to Republican precincts, in proportion to the pre-existing vote count (such as, perhaps in Miama and Warren Counties)

3. Voter registration fraud.

4. Fraud via voter suppression, such as insufficient machine allocation, intimidation of voters, and failure to process absentee ballot applications of Democratic voters.

5. Vote switching fraud targeted at counties or precincts in which Bush was doing poorly relative to 2000 (e.g., Cleveland).


I agree that your work on this has been and should be used to target our efforts. It was certainly a reason for my focusing of efforts on Cleveland, where I believe the key to the loss of Ohio and therefore the whole election lies, as indicated in this thread:
http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=104&topic_id=4998616


It should be noted that item #s 1 and 2 above, while not affecting PLD, would affect the overall red shifts in state and national exit polls. Item #s 3 and 4 would affect neither the PLD nor the state exit polls. And item #5 would affect both the PLD and the state and national exit polls. Therefore, the one significant issue where I disagree with you is when you appear to suggest that the exit polls provide no supporting evidence for election fraud.

but I agree with your five points, and I'd add that any kind of "deletion" of votes from largely Dem precincts would help Bush and not show up in the poll, including mechanical failure - hanging chads and all.

Not sure how 1 and 2 would affect overall red-shift. Doesn't matter though. The point is that it would help Bush win.

Five would affect PLD, yes, but would not show up in that correlation. Bit thin though. I'd "consistent with" rather than "support for".

Need to think about that.

But we are on the same page I think. And my ballpark confidence limits give you a bit of wiggle room.

You may not like this but I actually think they rule out most kinds of erro r that would show up in the exit polls. This leaves the forms of fraud that TfC has listed (and I am grateful to TfC for pointing these out.

I think there is also room for a bit of selective vote spoilage, which WOULD show up in the exit polls, but WOULDN'T tend to show up in that correlation if it had happened in the same precincts in 2000. In fact anything that happened in 2000 wouldn't show up in 2004 in the correlation - but the problem is that the exit polls were pretty good in 2000 which sets a rather low baseline.

There are some fancier loopholes, all of which seem to me to be a stretch, but maybe could have legs (scuse mixed metaphor):

Absolutely uniform fraud in all precincts. All variance in exit polls is then due to polling error, but the shift of the mean is due to fraud.

Fraud only in unpolled precincts (but this is limited by the finding that the NEP precincts on the whole were good predictors of the state count).

Fraud targetted in precincts where Bush was doing badly, or some algorithm that stopped Bush's vote share dropping below a certain level, relative to 2000. This sets up a testable hypothesis, that unfortunately I suspect would not be supported: if this was the case, then the distribution of "swing" (Bush's change in vote-share since 2000) should be positively skewed - more precincts where he did better than precincts where he did worse, roughly speaking. Unfortunately, eyeballing the plot it looks the other way - if anything there were a more sensible places where he did dramatically worse than his average than places where he did dramatically better.

Here's the latest, possibly from anaxarchos, except I hesitate to attribute it to him in case I've got it wrong, and the poor guy is battling Wilma right now:

That there was massive fraud in 2000 but in some exit-poll invisible form (e.g. crappy machines in Dem precincts), that was replaced in 2004 by exit-poll visible fraud (e.g. vote switching) - I'm still trying to figure out what this does the the swing, but it certainly would raise the 2000 baseline in precincts where we are postulating fraud in 2004. The good news is that evidence should be available in looking at rejected votes at precinct level in both years. But I have to say, this is very recent, and I need to do a bit of modelling before I'd even buy it as a mathematical possibility, let alone a practical one. But anax is a smart guy.

Any takers? Any more suggestions?

The goal is to produce fraud in 2004 that is reflected in the exit poll discrepancy, but not in a way that producees a positive correlation between Bush's gain and the magnitude of the discrepancy.

backtracking on that in this post -- if I understand you correctly.

I don't see why targeting of fraud to areas where Bush was doing worse would require that there would then be more precincts where he did better than worse than 2000. It seems like this would be affected by so much random error (not to mention randomly or not so randomly distributed fraud) that it would be awfully hard to assess anyhow.

And I don't think that you meant to say that the exit polls rule out certain types of fraud. I think that you were referring to Mitofsky's new study and the ESI study, when assessed in conjunction with the exit polls, that rules out certain types of fraud.

Yes, it is the ESI study and the new analysis I am talking about. I do think they put strong contraints (I'm not saying rule out) on the magnitude of certain types of fraud.

You may be right about the distribution of swing. It's just that I saw the glimmer of a nice testable hypothesis, as skew is something you can measure and get a significance value for (you test against the null hypothesis that the distribution is normal). But it looks as though it goes the wrong way.

since your first sentence addressed it.

Sorry, trying to deal with too many threads!

Drop me an email

Lizzie

but no PLD.

Scenarios 1 and 2 would both change the official vote count in Bush's direction, while not affecting the exit polls in any given precinct. That would cause a red shift for the statewide exit poll.

on the way the state projection was made. If turnout was factored in, you may be right. But it wouldn't account for the redshift in at precinct level, and we know that the redshift was greatest when computed at precinct level. We also know that the sampled precincts on the whole were representative of the state - in fact there was a slight blue-shift.

But maybe someone else can chime in here. My expertise is at precinct level!

I thought that would be too sophisticated of a manipulation for the exit polls to deal with.

And you are probably right.

I just can't see how the state projection could be affected but not PLD, if the precinct selection was good.

Here is the ESI hypothesis as I intepret it:

Say you have an Ohio county in which the 2004 exit polls showed Kerry ahead by 4 points. The 2004 official results, alas, show Bush winning the same county by 7 points, a huge redshift from the exit result to the official result.

Therefore, goes the ESI hypothesis, if the 2004 discrepancy is due to fraud, Bush's percentage of the official 2004 vote should be significantly greater than his percentage of the official 2000 vote. For the love of God, why? This is nonsense. If there was an organized effort to rig the election in Bush's favor and get away with doing so, the first thing the crooks would look at would be Bush's 2000 percentages, and take care keep the rigged 2004 percentages within the ballpark of 2000. The perpetrators would have anticipated intense scrutiny of the election, and would know that big shifts in Bush's overall share of the vote would be the first anomaly to arouse suspicion.

This would mean that a Kerry blowout would not be stealable, but a close election would-just keep Bush's percentages close to 2000 in most places, and when high turnout in Kerry stronghold threatens that operation, spike the turnout in Bush's strongholds, leaving the percenatages alone (and this seems to have been happened in Miami County, which added 19,000 votes-over 40% of the final total, with the tally changing by less than a point-that's old news and maybe has been debunked, but I did not find anyting), and shave turnout in Kerry strongholds, again leaving the percentages alone. You would not NEED to give Bush big percentage gains vs. 2000, because he won in 2000. Turnout spiking/shaving would take care of the rest.

I raised this issue with Febble on a Kos diary ("Paul Douglas Fan" is mu Kos name) I wrote a few weeks ago , and she replied that she did not think that my interpretation of teh ESI hypothesis was accurate. I believe I get it, and the hypothesis still seems to be terribly flawed, no matter how elegant the testing of it may be.

that if the fraud was focussed on precincts where Bush was doing badly, it would help hide it.

But the ESI hypothesis isn't exactly that fraud would lead to his percentage of the vote being "significantly" greater than 2000. The hypothesis is that his 2000 vote share will be "predictive" of his 2004 vote-share, generally. His average performance may be a bit of an increase on 2000. But in some precincts his performance will be better than his average, and in some precincts worse. ESI's finding was that the precinct-level poll discrepancy wasn't greater in the former, or less in the latter. But it was a relatively small study (because of the small number of polled precincts in Ohio) so there's a bit of slop in the stats for good luck (or deviousness).

However, Mitofsky's new plot does the same (or equivalent) analysis for the national sample of precincts (minus a few with high numbers of absentee ballots). And again, there is absolutely no correlation between the degree of redshift in the poll and the degree of swing.

It is simply hard to see (not impossible, necessarily, but I don't see it yet) how fraud can be responsible for both redshift and swing, and yet not show up as a positive correlation in an analysis with that much statistical power. Because fraud must result in Bush doing better than he would do without fraud. That is what the fraud is for!

But I agree if it was subtly done, it might keep below the statistical radar. Two ways in which this could happen - if it was small-scale and carefully targetted; or if it used some fancy statistician-beating algorithm. I'm not at all happy with any algorithm I've found yet. But I'll keep trying.

Nice to see you on DU!

I think a really important key to understanding ESI's argument is to consider what they were arguing about. Febble makes similar points, but I want to focus on something basic (also a point that Febble has made before).

They headline their report on their home page: "Ohio Exit Polls 'Not a Smoking Gun' for Fraud, Study Says."

You write, "Therefore, goes the ESI hypothesis, if the 2004 discrepancy is due to fraud, Bush's percentage of the official 2004 vote should be significantly greater than his percentage of the official 2000 vote."

And you rightly point out that that may or may not be true. If fraud took the form of dialing up Bush turnout and dialing down Kerry turnout, the precinct percentages wouldn't be affected. The fraud might be detected by other means (I should think probably would if it were HUGE, but it wouldn't necessarily have to be), but not this one. (And the ESI report may have missed this point completely, because the brief appears to treat "vote-shifting" and "systematic fraud or error in vote counting" as synonymous, when they aren't.)

But consider the context. ESI was assessing the argument that the Ohio exit polls prove fraud. And the Exit Polls Prove Fraud arguments of which I'm aware do emphasize precinct-level red shift in the exit polls, and take it as evidence of precinct-level vote shift. ESI concluded that, in Ohio, it probably isn't. You don't seem to contest that conclusion.

You may think that the Exit Polls Prove Fraud argument is a straw man. Unfortunately, it has dominated the approximately scientific debate about 2004. Kathy Dopp has released a paper that asserts, "The Smoking Gun: Ohio Exit Poll Data Show Virtually Irrefutable Evidence of Ohio Vote Miscount," and directs us to the forthcoming "The Gun is Smoking" for the virtually irrefutable details. Hmm.

If I try to sell you a 1995 Mercedes and say it's a 1999 Mercedes (or some other year worth more) that's fraud in the broadest sense. Also fraud would be saying candidate X got 110 million votes but in fact got 99 million, in the broadest sense of misrepresenting a reality that the other is relying on.

On the specific context of elections, because of the public's total vulnerability and inability to independently verify election results, if the totals are off in any material way for any reason, I would say that's "election fraud". Others will assume the old common law restriction that it's only fraud if someone does it on purpose and essentially knows they are tricking another....

Apparent agreement here: OTOH wrote, quoting sean "Therefore, goes the ESI hypothesis, if the 2004 discrepancy is due to fraud, Bush's percentage of the official 2004 vote should be significantly greater than his percentage of the official 2000 vote." And you rightly point out that that may or may not be true."

To put part of the above posts in a thought capsule, would you agree that it would be proper to say that "increasing or decreasing the size of the pie but not changing the Kerry or * percentage is not exit poll detectable but does help change election results because total vote margines (NET BENEFIT) is provided so long as the percentages of Kerry and * are not EQUAL"

Whoever is winning will want to increase the size of the pie, whoever is losing will want to shrink the pie.

Febble has moved the ball farther than I have, I think, in working through some particular scenarios. I can think of some ways that massive pie-shrinking or pie-growing might be detectable in exit polls, but it seems to me they would not be very robust. Generally, it seems like a good way to go from the exit poll standpoint.

I'm a bit at sea with the pie.

Pie shrinking might be hard to detect, but the one datum we do have is the E-M report on how well the selected precincts vote counts predicted the states - apparently well.

I don't know how the differences were computed, but presumably some kind of weighting for total vote count was involved. If total vote counts were shrunk (or swollen) then that might show up as larger error in those tables (the errors are fairly small).

But I'm willing to be convinced otherwise.

Certainly pie-shrinking won't show up in the scatterplot we were looking at; as you say, it might show up in other ways. What magnitude of pie-shrinking would have the appearance of blowing up the precinct sample? I have very little idea. But that table is basically comparing exit poll precincts with other quick-count precincts, yes? That doesn't seem like a very sensitive approach to detecting vote deletion.

Dunno, can't really think about it hard for a little while (unless I'm insomniac, of course, in which case who knows?).

(EDIT: Just to put this in context -- I'm not trying to present this as a fraud panacea, no matter how the precinct-sample issue works out.)

I want to see a pie chart of the exit polls! We're almost there!:party:

... I just don't know how to draw a pie.


Pieces of a precinct's WPE pie:

i random sampling error

ii sampling/respondent bias

iii precinct vs. polling-place count (net error)

iv missed/refusals (net error)

v absentee voters (net error)

vi early voters (net error)

vii under/over votes (net error)

viii uncounted provisional ballots (net error)

ix fraud/accidental vote errors:
adding, deleting, switching, mis-counting (net error)

Febble, you write:

It is simply hard to see (not impossible, necessarily, but I don't see it yet) how fraud can be responsible for both redshift and swing, and yet not show up as a positive correlation in an analysis with that much statistical power

So "redshift"=the difference between 2004 precinct-level official results and 2004 precinct-level exit results?

And "swing"=the difference between Bush's official 2000 percentage and official 2004 percentage?

I'm operating on that assumption.

I do understand the ESI hypothesis better now, but I doubt that it proves or disproves anything. Why would the exit-official difference be greater in areas where Bush did better than 2000, and less in areas where he did worse than 2000? If there was a multistate effort to keep Bush's percentages fairly near to his 2000 percentages, there would be a third category that would confound these correlations: areas where Bush's official 2004 result was unremarkable vs. 2000, but in reality, he did piss-poor vs. 2000. The exit-official difference would be significant in such areas, and would blunt the correlations for areas of better or worse performance vs. 2000.

If there was multistate fraud, I suspect that the effort's first task would be to keep its eye on 2000 patterns and not deviate wildly from them. The fraud would not focus particularly on areas where Bush was doing badly; it would be an across-the board "adjusting" of the percentages to the maximum unsuspicious but favorable deviation from 2000 (which would not be much, given the expected scrutiny of the election, but they'll take votes where they can find them), and then play with the turnout within the percentages required for both victory and plausibility. Then you would see an election in which overall turnout was up, but not as much as many were expecting, and the "real tunout story" was how well Bush turned out his base.

Anyway, I don't see how ESI's hypothesis accounts for the third variable I've suggested, and therefore how it can prove or disprove even the proposition that the exit polls prove fraud.

I've taken to using the term "redshift" because it also indicates direction - it means the extent to which the count was "redder" than the poll. In quite a large number of precincts there was "blue-shift", so if you want, you can think of "blue-shift" as a negative value of "red-shift".

There are various ways of measuring "redshift" - ESI used a difference measure, Mitofsky used a ratio measure - I'd argue for the ratio measure, but it's not critical.

I've used the word "swing" because it's how it's termed in the UK (where I live!) - again you could define it as a difference (as ESI did) or as the residuals when Bush's vote-share in 2000 is regressed on his vote share in 2004, which is what Mitofsky did - but they are both measures of how well Bush did relative to 2000. The ESI measure is an absolute measure - negative means he did worse, positive means he did better. The residual measure is relative to his average. So positive means he did better than his average, negative means he did worse than his average. It's not particularly critical which you use.

So that's the terminology dealt with, I hope!


If there were no fraud, you'd expect Bush to do better in some precincts and worse in others, and you'd also expect polls to err in one direction in some precincts and in the other direction in others (if there was actual BIAS in the polls, you'd get more redshift than blueshift). But you wouldn't expect the two to correlate, because there is no common factor affecting both.

However, if fraud were adding extra "swing" in some precincts (regardless of whether Bush was really doing better or worse than his average) it would also add extra redshift in those same precincts, regardless of whether the real polling error in that precinct was in the red or the blue direction. So you'd expect a positive correlation between redshift and swing - when one goes up, so does the other.

So some of the variance would be shared. Mitofsky's plot gave the proportion of the shared variance, and it was .009%, which is very small (and the correlation in fact was negative). I don't know the figures for the ESI finding.

The figure itself is not the important thing though - what is important is the upper confidence limit of the correlation coefficient. I don't know the upper confidence limit for the ESI study, but it will be higher than in Mitofsky's plot, as the power was less (far fewer precincts). So there's a bit more room for a "true" correlation to be lurking in the Ohio plot than there is in the national plot. In the Mitofsky plot the upper confidence limit is tiny.

Now, it is possible that some fiendishly subtle algorithm was used to disguise the correlation that would naturally be induced if the same factor (fraud) was pushing both the exit poll discrepancy and the swing to Bush. But, although I am trying really quite hard, I haven't got the winning ticket yet!

Hope this makes some sense. Nice to talk to you.

I think we're straight on the terms.

Your post makes sense, but it is only relevant to fraud that would produce swing. The confounding variable I propose is the fraud's being deliberately designed to avoid any consistent swing: as I wrote, to keep the numbers unremarkable vs. Bush's 2000 performance. The nasty folks would have to allow for some areas to swing vs 2000 in or against Bush's favor-that's what the statisticians expect. Then, again as I wrote, the adjustment of the turnout would be crucial to victory. And if our nasties were executing such an m.o., there would be no correlation between redshift and swing.

And such an m.o. would not, I think, require, a fiendish algorithm. Investigate what deviations across the state have looked like in elections where a candidate did a bit better than the last election, and don't let a mosaic emerge that would look terribly odd to people knowledgeable about such matters.

Nice talking to you as well, and thank you for, as always, setting a cordial tone.

I've been working on it, and haven't succeeded yet.

The thing is, you can't alter a vote without altering a vote....

Swing will happen in both directions and to various extents anyway. What you will do, inevitably, by switching votes, is to increase it from whatever its "true" value was in that precinct, even if it was only a little bit - and you will increase redshift by the same amount.

Now as long as it was fairly small scale, I think you might get away with it in Ohio, because for any one state, the plot will be fairly noisy. But for it not to show up in the national plot does present a problem for both the fraud-as-cause-of-exit-poll-discrepancy argument, as well the massive-vote-switching-won-Bush-the-popular-vote argument, though not necessarily to the criminal mastermind. However, lacking a criminal mastermind, I haven't yet figured out how.

If you can be more specific about where you would apply the fraud (what kinds of precincts) and to what extent, I'll add it to my model

I got home from work just a bit ago.

I don't think you address my proposed confounding variable to the ESI and Mitofsky analyses.

So do you admit that that if the fraud was done with a close eye on 2000, with vote-switching to improve a just a few points on his 2000 performance, and being densest in high-turnout areas which did not show high-end deviation from 2000, you would not expect a strong correlation between exit redshift and the degree of difference to 2000 performance?

Interestingly, at the county level, the only areas where Bush gained four percentage points or more on his 2000 performance were counties in which he won 60% or more in 2000. These counties are Auglaize, Belmont, Clinton, Darke, Lawrence, Mercer, Miami, Paulding, Shelby, and Van Wert. (Source: Dave Leip's Election Atlas.) If Bush were losing less ground in reality (or doing about as well in reality) in these counties than in 2000's more competitive counties, fraud here would not create conspicous exit redshift vs. more comptetitive counties where he gained fewer points on 2000.

I understand that for the purposes of the exit poll, we are talking about precincts and not counties, but I think those county examples offer food for thought, big gains made only where Bush was very strong.

And once again, we have other, very powerful prong to a fraud m.o.: turnout spiking and shaving, the "pie size" hypothesis, which I know you have admitted would be invisible to the exits.

I am working on something related to the Ohio results which I began after our discussion several weeks ago at Kos. I have not made as much progress as i'd intended/promised-to say the least. I am working on something which I plan to finish soon (hopefully the middle of next week) and I'll e-mail it to you.

I live in Miami and things are fairly messed up here. Thinking of taking a road trip if we can get some gas.

So I hope you guys won't all come to agreement and wrap this thing up before I can get back. :rofl: (I made myself laugh.)

Cheers,
eomer

Sorry to here about your power! Glad to know you are safe!

Hope you get cleaned up soon. :)

Fraud mainly took place in larger precincts with large increases in Dem registration relative to Republicans. (The Dems got out the vote, but votes were switched to Bush.)

Votes switched from Kerry to Bush, offsetting the new Dem registrations, making Bush's percentages comparable to 2000 and blunting the surge in Democratic votes.

BushVote04% is still correlated with BushVote2K%, but a bit higher.

No correlation between 2004-2000 swing and Red Shift (because in fraudulent precincts, there was not much 2004-2000 swing).

Many precincts were left alone because they were smaller and/or were not needed to affect the outcome. They were more valuable as noise sources.

No correlation of Red Shift vs. partisanship since only moderate Kerry precincts ended up moving toward moderate Bush. (And there ARE more moderate Bush than moderate Kerry precincts.)

NEP precincts match states as a whole because fraud took place in both NEP and non-NEP precincts.

So what do you think?

what I like is the use of registration figures as an input. It's also consistent with my hunch that predominantly Democratic precincts might be the best place to hide fraud. Let me add it to my model.

It should make a testable hypothesis - if you can get increase in Dem regisration figures for each NEP precinct, we could ask Mitofsky to test whether PLD was greater in those precincts than elsewhere.

But let me keep thinking....

If they couldn't dissuade 'em thru some form of voter suppression, they steal thru vote switching as many of 'em as they can w/o drawing attention. Based upon the exploit method, they could prolly predict within a very discernible range the percentage of votes switched.

You said:
"I DON'T want to prove that Bush won a fair election. I WANT to prove that he didn't. But I think it is very hard, given that..."

What is making it extremely difficult is that you continue to use illogical invalid arguments that you state so sweetly and convincingly, to rule out vote fraud as the cause of the exit poll discrepancies, misleading the public who don't know much about logic and math.

American democracy deserves better.

In the future, please check the LOGIC of your own hypothesis to dismiss vote fraud - BEFORE you begin parading it around publicly. I asked you last week to check your logic and told you that it was invalid, and yet you continue here to parade this misleading incorrect logic! Lives all over the planet depend on American democracy.

Anyone, especially you, could come up with dozens of counter-examples to show how logically invalid the hypothesis is that you based your analysis on as you just stated again above.

Please check this out:
http://electionarchive.org/ucvAnalysis/US/exit-polls/ESI/ESI-hypothesis-illogical.pdf

and study the counterexample and create a spreadsheet for yourself that generates counterexamples that prove how invalid your logic (stated above) is.

Please stop obfuscating and misleading the public on this issue!

Elections analysis deserve more care.

I regard your paper as extremely misleading.

You might like to check your own logic.

But I'll post on your thread, not this one.