Machine Theft of the 2004 Election : Real Simple Math

I don't have the numbers handy, but off the top of my head, here goes:

Machines used in the 2004 election = 300,000

Votes needed to swing the election = 1.5 million (one half of the 3 million votes bush won by)

1.5 million votes divided by 300,000 machines = 5 votes average per machine.

5 times 300,000 equals 1.5 million.

It took just 5 votes changed per machine to create a winner.

Each machine could hold as many as a thousand votes.

5 votes changed by each machine would be 1 vote out of every 200.

-------------------------

The preceding announcement, as stated, was done off the top of me poor ol' head, and like the SCOTUS did in 2004, I refuse to be held accountable.

for there to have been as great a discrepancy as there was by fraud. It would have involved too many PEOPLE.

in the voting locations in the first place?

...beauty is in the eye of the beholder.

See, most election officials just LOVE these machines. Heck, they are only human. What human would not LOVE a machine that makes their work ten times easier?

Then too, HAVA paid for the machines.... nearly $4 billion spread far and wide, from sea to shining sea.

HAVA was Tom Delay's baby. He won. His baby beat up our baby.

Then we look at Iowa. Remember Iowa, Stevepol? Remember that Iowa link you left in the middle of a thread?

Remember that 74 votes out of about 200 were changed in just the absentee ballots alone in Pottawatamie County?
http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=432294&mesg_id=432294

We see that the machines, on average, could easily change just 5 votes each, but when you have machines changing over 50% of ballots in some races, it would take far less than 300,000 machines to steal a national election.

Who'd a thunk it? Not the experts, nope.

...programming code. It makes it so easy. One hacker, a couple of minutes, leaving no trace. Or, one insider at the corporations who manufacture and service these voting machines, whose programming no one outside the company is permitted to review. Simple enough to write a distributive program for a large database (say, in the central tabulators of each state--also run on "trade secret" programming)(--and now we're down to a handful of corporate personnel), to randomize the vote stealing, and have it triggered by some circumstance (say, Kerry winning) or by remote or on-site access. Two related corporations "tabulated" 80% of the nation's votes in 2004, under this veil of corporate secrecy, both with very close ties to the Bush regime and far rightwing causes: Diebold and ES&S. You think they didn't use their secret formulae to keep their boys in power? Ha-ha on you!

Ohio was EXTRA, because the desire to oust Bush was so strong, a landslide for Kerry was happening. They couldn't be sure that the secret vote tabulation programs would be enough. So they had to go steal the voting machines in black precincts and hide them, and call black voters on the phone and tell them they'd go to jail if they showed up to vote, and all the other crap they did. But Diebold and ES&S was the foundation, put in place during the 2002-2004 period--with election officials rushed, bullied and bribed to purchase this untested, insecure, hackable, secret vote counting technology.

If they had wanted a transparent election, we would have had one. Transparent elections are not difficult. They had the opportunity to improve election transparency and accuracy, with HA VA. Instead, at every turn, they permitted egregious non-transparency, including not just 'TRADE SECRET,' PROPRIETARY programming code, but no paper trail (let alone a real paper ballot backup), and thus no recounts possible in one third of the country, inadequate audits/recounts everywhere else, secret industry "testing" of the machines, and lavish lobbying (behind the scenes payoffs, deals, revolving door employment).

Means. Motive. Opportunity. History of criminal behavior.

It's a no brainer.

What's the remedy? HAVA did NOT mandate electronic voting; it just poured money into the system to corrupt it. Paper ballots, handcounted at the precinct level (the most transparent and verifiable method of voting) is still legal and doable. The best place to achieve it (the most feasible right now) is at the state/local level. So, go on down to your local county registrar or board of elections, and demand transparent vote counting! Non-transparent elections are not elections. They are tyranny. Ask the citizens of Stalinist Russia!

About absentee ballot voting: YES, it's a good idea! It's not the total answer, by any means (because all ballots are scanned into the central electronic tabulators, and get separated from your ballot), but it IS a tangible record of our vote, to help investigators (as AB ballots did in 2004) and in challenges of suspicious results, and it IS a protest against the machines. If enough people do it (and many are), the machines will be obsolete; then we can work on getting rid of the central tabulators. Caveat: take it to the precinct on election day (if that is allowed); don't trust the USPS if you can help it.

Throw Diebold, ES&S and all election theft machines into 'Boston Harbor' NOW!

:think: :patriot: :argh: :bounce: :argh: :patriot: :think:

at least ehre in california. I watched the election "tally" in LA, and got a tour. the absentees, provisional ballots and torn ballots are all stuck into a separate room, called the snag room. None of them are counted on lection night. Legally they have 3 weeks to count them. they are not amrked by precinct, even if you walk them to your precinct. And they are not included in the 1% manual count. There is really no record of them at all. here they represent 40% of the votes. Yet there is no check of any kind at all.

Not every machine must be tampered with.

A real simple little program inserted that said something like If R vote total is greater than D vote total then R is R and if R is less than D D=R and R=D.
A fe well cousin precincts and you have a win

Instead of 5 votes per machine, the correct number is 2.5 average votes changed per machine.

The reason is that as you move 2.5 votes from one column to another, in reality it shows up in the total as 5 extra votes for the apparent winner. So the real number, instead of 5 votes per 1000, is 2.5 votes per thousand. Or:

1 vote changed for every 400 cast.

Simply put 3,000,000 votes from a total of 120,000,000 is one 40th of the votes cast. So changing just 1 vote changed in every 80 votes, would have swung that election.

That's about right, I think. Heck, I ain't no expert.

Let's stick to round number: about 62 million votes were counted for Bush and 59,000,000 for Kerry.

Which gives Bush 51% and Kerry 49%. Now the exit polls, at close of poll, got those the other way round. So let's say that 62 million votes were actually cast for Kerry, and only 59,000,000 for Bush.

That would mean that 3 million votes were stolen from Kerry and given to Bush - switched, in other words. Now, if there are on average 1000 voters per precinct, that would mean that on average, 25 votes in every precinct were switched to Bush. If only half that number were stolen, then Kerry and Bush would come out about the same.

So to ensure a Kerry victory (and to square with the exit polls), you really have to steal an average of 25 votes per precinct. But of course you can only steal Kerry votes where there are Kerry votes to steal, instead, let's steal 5% of Kerry's vote - 1 in 20 Kerry votes. After all, if you steal 25 Kerry votes in a precinct where there are only 50 altogether, it is going to look horrible in the exit poll.

The problem is that you can probably only do it in a certain proportion of precincts. Let's say 50%. So you set it up so that in 50% of precincts, 10% of Kerry votes are flipped to Bush.

In heavily Bush precincts, this won't show up much. If there are 100 Kerry voters in a precinct and 900 Bush voters, and you flip 10% of the Kerry votes (i.e. 10) to Bush, Kerry will get 90 votes, and Bush will get 910. So instead of Bush getting 90% of the vote, he will get 91%.

It will look like a modest 1% "swing" to Bush.

However, in heavily Kerry precincts it will make quite a difference. If there are 900 Kerry voters and 100 Bush voters, and 10% of Kerry votes are flipped to Bush, only 810 will be counted for Kerry, and 190 for Bush. So instead of Bush getting 10% of the vote , he will get 19% of the vote and it will look like a substantial 9% "swing" to Bush.

This is actually fine, because it is just what real "swing" looks like. The more Democratic a precinct was in 2000 the more potential there is for a swing to Bush, whereas in a precinct where Bush got 98% of the vote in 2000, the most he can improve is 2%.

The problem for Rove, at this point, is the exit poll. Precincts where a lot of votes were flipped to Bush are going to have a large redshift, whereas precincts where only a few votes were flipped are not. It won't be exact, of course, because polls are only a sample, but, on average, the where the vote flip was substantial, there is going to be a tell-tale redshift in the poll.

So, when some smart person figures out that the redshift in the poll just happens to coincide with precincts where Bush's swing is on the large side, and that polls without redshift, or only a little, or a bit of blueshift just happen to be where Bush's swing is absent or negative, then the game will be up.

OK: back to reality: there is NO TENDENCY for redshift to coincide with larger than average swing.

And that is the problem for the theory that fraud was responsible for the exit poll discrepancy, and, therefore, for the theory that fraud won Bush the popular vote.

But if you can solve it, you may have a case. Actually, you've got a case anyway, because the case for transparent, auditable and secure voting systems does not depend on proving that Kerry won the popular vote. Although it may help Democrats next time if they accept that he probably did not.

...you have said in the past that auditing elections with the exit-polls is not what the polls are designed for.

Yet, you continue to audit the election with exit-polls. Of course, you only look at one side, the rbr side, to the exclusion of any others. Plus, we are not allowed to see the data you work with.

But we have an answer: the raw exit-polls showed not that 1 out of every 80 votes was changed, the exit-polls show that 1 out of every 40 votes was changed.

That $10 million exercise showed Kerry with a 3 million vote margin of victory: 62m to 59m. Then too, there was the obstruction, and turning away of voters from the polls. Add in those written off votes, and it's clear Kerry won going away, but was robbed.

No, the US exit polls are not designed to audit elections, which means if you want to use the exit poll data to figure out whether or not an election might have been stolen you have to think about it carefully.

But the fact is that we all know that there was a substantial gap between what the exit polls were telling us about the result and what the vote count was telling us.

And there are two possible reason for the gap (chance is ruled out, statistically, as everyone agrees).

1. Vote theft
   
2. biased poll
   

So to figure out which it was - or how much of which it was - we have to ask the questions:

1. If the gap was due to vote theft, what patterns would we see?
   
2. If the gap was due to a biased poll, what patterns would we see?

And we look for those patterns. One dead giveaway for vote theft would be if Bush was doing a heck of a lot better in precincts with a big "redshift" gap than in precincts where the exit poll matched the count. Because we would then be able to say: look - Bush is stealing votes - where his vote is inflated, the exit poll discrepancy is bigger, and where his vote is what we'd expect, the exit poll discrepancy is what we would expect.

But when you look at the data (and there are good pictures of it here), it turns out that that pattern isn't there. If we look at precincts with a large discrepancy - where you might expect fraud, if fraud was causing the discrepancy - Bush's vote isn't any more inflated than it is anywhere else, i.e. in precincts where the exit polls match the count much better.

So, to take number 2: bias in surveys is caused when people who for some reason or another get missed by the survey (they avoid the interviewer; they refuse to answer the question) turn out to be different on the characteristic you are interested in (eg. how they voted) as well. And we do know, from experimental studies, that Republican voters are more difficult to persuade to take part in exit polls than Democrats. In fact, if you try to improve response rate by giving away free pens or folders, for example, you get a better response from Democrats than you do from Republicans - in other words you get a more biased sample not a less biased one.

So, given that we know this, if in 2004 Bush voters were less willing to participate in the poll than Democrats, and more likely to avoid the interviewers, the pattern you'd see is that you'd get a bigger redshift in the poll in precincts where the opportunities for escape were greater - when the "interviewing rate" was large, for example, or where the interviewer was more than 25' from the precinct.

And that is exactly the pattern in the poll. Where conditions at the precinct were likely to make it easier for unwilling voters to evade the interviewers, the bias in the poll was greater.

So, although you have to reverse engineer the polls to do so, it is perfectly possible to figure out ways of looking at the poll data that will tell you which is more likely to have caused the discrepancy - fraud or polling bias. And the patterns you'd expect if it was fraud are completely absent, and the patterns you'd expect if it was bias are very strong.

This means that it is highly likely that the discrepancy in the poll was due to a biased poll, and not to massive fraud. And without evidence in the exit poll for massive fraud, we are left with lots of evidence that suggests that it occurred, but no evidence that it occurred on a scale of millions of votes. Size matters.

But, of course, voter suppression wouldn't show up in the poll anyway, and if anything cost Kerry the presidency, it was that - systemic suppression of the votes of largely ethnic minority voters who tend to vote overwhelmingly for the Democratic candidate. This is not only a travesty of democracy but a serious civil rights issue. I'd like to see it take centre stage, instead of what I consider unsupported allegations of digital hyper fraud.

You look for discrepancies - a big "redshift gap" as a sign.

But we see that just 1 vote out of 80 need to have been changed to swing the election.

That means that in a precinct with 800 votes, 10 votes would need to be changed. Here is an example:

Kerry 400 votes
bush 400 votes

Take 5 away from Kerry he now has 395
Give that 5 to bush he now has 405

The precinct level exit poll data is so small that it would not be able to pick up that 1 of 80 change. Overall, nationwide, the redshift is quite evident, and was seen by the exit-polls, and in the results compared to 2000, but at the precinct level, 1 out of 80 is unknowable. Even 1 of 40 would be unseeable at the precinct level, but shows up big time nationally.

and you are right that if every single precinct the same tiny amount of fraud, then it wouldn't show up as a correlation between shift and swing. Eomer made this point a while back and it is a good one.

But I don't think your numbers are quite right. If you want to square the result with the exit poll, you need to transfer 3 million votes from Kerry to Bush, although only 1.5 million if all you want to do is equalise them (in which case you'd have to assume the poll was a bit biased, which is fair enough). If you want to steal 3 million votes, that's 1 in every 20 Kerry votes, and if you want to steal 1.5 votes, that's 1 in every 40 Kerry votes. But let's stick with the second scenario:

In a precinct with:

Kerry 400
Bush 400

Kerry is on 50% and Bush is on 50% - margin is zero.

You need to steal 1 in 40 Kerry votes, i.e. 10 Kerry votes. So we take 10 off Kerry's total and add it to Bush:

Kerry 390
Bush 410

So when the vote is counted, Kerry is at 49% and Bush is at 51% - margin is now 2%, approx. Not a lot, but more than zero. Within the margin of error of the exit poll, but, as you say, overall it would add up to a big redshift. However, it won't show up as a correlation between shift and swing, because the extra shift won't show up as extra swing because they are the same everywhere.

But for this to work it has to happen in every single precinct. In every single precinct in the country, someone has to carry out the fraud - on lever machines, in paper ballot precincts, on punchcards, on Diebold, on Sequoia, on ES&S machines - and in every state. If you can figure out a way that might have happened in 120,000 precincts without anyone noticing, then please tell me.

A more likely scenario is that it happened in, say 50& of precincts. But in that case, you will have to steal 20 Kerry votes from your sample precinct, rather than 10, so you will end up with

Kerry 380
Bush 420

Kerry will have 48.5% of the vote, and Bush 52% of the vote - and instead of a zero margin (the true margin) the margin will be 5%. Quite a difference. Not only that, but the exit polls will tend to be off in these precincts, where the fraud took place, and not where it didn't. And Bush will do better in these precincts than he does elsewhere. So you will tend to get a pattern where Bush does better than expected in precincts with bigger redshift - the tell-tale pattern of fraud.

But certainly, if you can figure out how someone could have stolen just 1 in 40 Kerry votes in every precinct in the US then you have a case that Kerry might have (just) won the popular vote, and it wouldn't show up in the swing-shift correlation. If you want to give him Bush's margin, you have to steal 1 in 20.

Cheers

Lizzie

"So you will tend to get a pattern where Bush does better than expected in precincts with bigger redshift - the tell-tale pattern of fraud.

Yeah, Pasco county is a good example of fraud.
In 2000, bush lost Pasco by 1,000 votes, but in 2004, bush 'won' by 30,000. And Pasco is not alone.

The other evidence points to millions of votes stolen by the old fashioned method. So we can stick to the 1 of 80 to bring things even, and the old fashioned theft is added on to Kerry's margin of victory.

Think about all the reports of machines switching votes right in front of the voters. Easily done, say, 1 of 100. Now you take some precincts where no one is looking, no polls, no paper, no records. There one could steal 100's of votes without anything more than a question. It happened where I vote.

The thing is, 1 out of 80 is easy to steal. 1 out of 40, even. The precinct polling wouldn't pick up on that, and one could surmise that the exit-polls were NOT set to count in places where big machine fraud would take place.

There are more than two ways to skin a cat. The theives were good, they set it up far in advance. Stealing hundreds of votes in one precinct after another, without running into the few precincts with exit-pollers hanging around, would be a priority for them, and a piece of cake.

But the general exit-polls told the story. Kerry 52%. Good work.

They are the smallest unit of analysis we have for the exit poll.

But you are still missing the point. If the theft was uneven it would show up in the swing-shift correlation, and it doesn't. And if they avoided the exit poll, it wouldn't show up as a discrepancy at precinct level, but it would show up as a discrepancy between the NEP precincts and the rest (and there wasn't one).

So there is no evidence that it happened on that scale and actually evidence that it didn't.

And you have to steal 1 in 40 Kerry votes (not 1 in 80) to bring it even. Or, alternatively, destroy 1 in 20. There is certainly evidence that Kerry votes went uncounted, as opposed to stolen.

I think, BeFree, all we are arguing about here is numbers. I quite agree with you that Kerry votes were switched to Bush, or went uncounted, or were never cast because of voter suppression. And I agree these things are terrible. I agree they must be stopped.

The only thing we disagree about is the numbers. I don't see any evidence for millions. I'm not sure you do either.

But we both certainly wish that 3 million more votes had got themselves into Kerry's column on November 3rd 2004. Or just 580 more in Gore's column in 2000.

Peace.

...is the early, raw numbers of the exit-poll that showed Kerry ahead by three points. There is your discrepancy. There being no evidence otherwise that points to anything as much as machine fraud, that evidence stands.

Your 'evidence' is not proof because it has not been proved by anyone else. No one else has been allowed to look at the data you used to make your 'evidence'.

On the other hand, the other evidence we have assembled has been proved. It has proved to be factual because it has been fully examined.

1 vote in 80 stands.

• it is the raw numbers I am talking about.
  
• there are two possible causes for the discrepancy
  
• I have tried to explain how you would distinguish between the two causes
  
• When you do this, it points to biased poll, not fraud
  
• you need to switch at least 1 in 40 (not 1 in 80) Kerry votes to Bush to give Kerry a popular vote win, and 1 in 20 to give him the margin that Bush ended up with.

Bush finished with 3 million more votes that Kerry. If you think those 3 million votes were originally cast for Kerry then switched to Bush, you need to account for 3 million stolen Kerry votes. If Kerry started off with 62 million, and ended up with 59 million, then 3 million divided by 62 million will give you the percent of Kerry votes were stolen, which is about 4%, or one in 23. If you think that both Bush and Kerry for real got 60.5 million votes each, and Bush stole 1.5 million, pulling Kerry down to 59 million and himself up to 62 million, then 61.5 million Kerry real votes divided by 1.5 million stolen votes is 2%, or one in 40 Kerry votes stolen.

1 in 80 stolen Kerry votes won't swing the popular vote to Kerry, and you need to steal 1 in 20 to make the exit poll discrepancy go away. If you can't do it in every precinct, you will have steal more than 1 in 20 where you do. If you do it in half the precincts, you will have to steal 1 in 10 - 10% of the Kerry vote. That will markedly inflate Bush's vote above pre-election expectations, and the inflation will tend to coincide with a redshift in the exit polls, so you will get a correlation between the two. There isn't a correlation, so the probability that 10% of Kerry votes were stolen in 50% of precincts is vanishingly small.

But keep a look out for eomer's posts - if anyone is going to find a loophole in my logic, eomer will.

Febble, I think it is from you that I heard that the E-M redshift numbers are based on the official count at the precinct (polling place) whenever possible. When it's not possible to get an official count that was done at the precinct then the official count from the centralized tabulation is used instead.

Do I have that right? If so, then in the {E-M precinct-count} precincts any machine fraud that occurred after the count was transmitted from the precinct to the county would not show up as redshift and would not create correlation between redshift and swing.

On the other hand, for those precincts where E-M used post-precinct (let's assume final official) counts then machine fraud anywhere in the chain up to the final number would show up as redshift and potentially as swing correlation.

In DRE counties, post-precinct machine fraud would not include DRE fraud of the type that is most often talked about (real-time vote switches and such) but there could still be post-precinct fraud in DRE counties. One example would be tampering with memory cards after they leave the precinct (which if I remember correctly is the favorite flavor in Volusia County, FL to name one place). In op-scan counties you can have similar memory card tampering and in any county you can have central tabulator fraud. So there are plenty of opportunities for machine fraud that would be post-precinct count.

So my point is that your conclusion needs to be qualified to clarify that it covers only certain types of machine fraud, similar to the way (as you point out frequently) that vote suppression would not be caught by the swing correlation analysis.

Do you have, off the top of your head, an idea what percentage of precincts in the E-M redshift calcs were based on precinct-based counts and what percentage were based on post-precinct counts?

off the top of my head it's about 60% counted at the precinct and about 40% from the county. I assume that older technologies are more likely to be precinct counts.

So there could certainly be some machine fraud that does not show up as redshift (if it is done at county level) - so you are right (again), although my point has been that there is no exit poll evidence for such fraud.

And of course, that adds variance to the distribution of fraud, increasing the chance of a swing shift correlation

BTW I set up a model to try to answer your question about the amount of fraud you might be able to sneak in with a following wind. Unless I made it virtually uniform, I couldn't ever get the regression line to drop below zero with anything like an popular-vote winning degree of fraud, and as soon as I decreased the proportion of corrupt precincts to below about 70%, or introduced variance into the amount of fraud, the regression line never dropped below zero on many hundreds of iterations.

It was tricky to do, though, as each time I changed a fraud parameter, I had to change other parameters in order to keep the variance in shift and swing constant. I didn't get as far as automating that, so I just had to put in discrete parameters and test those, rather than systematically working my way through a continuous set. Also, life intervened, and I didn't seem to be winning the election for Kerry.

The other possibility of course, is a fraud algorithm that adjusts Bush's vote count to keep it off the floor, but without rising more than a set amount. I just can't see anyway of implementing that over the proportion of precincts necessary though. Feel free to suggest a mechanism!

I should also say, however, that while postulating uniform fraud means that only the intercept (or the mean value) of the discrepancy need by explained by fraud, not the variance, a very important finding from my work for Mitofsky is that the intercept itself is well accounted for by precinct level variables. Essentially, the reason I consider fraud to be an extraordinarily unlikely cause of anything more than tiny proportion of the exit poll discrepancy is twofold: the finding that the discrepancy is well-accounted for by variables that indicate selection/non-response bias, leaving nothing to be accounted for by anything else (but room, nonetheless for other minor effects that could include fraud to a small degree); and the finding that there is not a sniff of a positive correlation between redshift and swing.

I did find that a significant portion of the redshift could be attributed to precincts in urban areas, particularly those with largely African American or Hispanic populations, using older technology which may reflect greater differential residual vote rates in those precincts, and as there is a long and bitter history of this (cf Gore, Florida, 2000), you wouldn't expect that to show up in the swing shift correlation because it is likely to have been a feature of those rpecincts in 2000 as well. It probably showed up more in my analysis than in the original E-M analysis because the measure I proposed (and used), unlike WPE, doesn't downsize the discrepancy in extreme precincts, as black urban precincts tend to be. But it would also have figured as a larger effect on the total than it actually was because black and hispanic precincts are deliberately over-sampled (and then proportionately downweighted in the projections) in order to give adequate sample sizes of ethnic minority voters in the crosstabulations. But that remains my only finding of evidence to support miscounted votes as a contributor to the discrepancy, and it, ironically, finds that digital voting systems appear to have been more accurate in this regard.

So this is why, although I think you are absolutely right to distrust digital voting systems, I don't think there is any reason to think that they had anything more than a minor effect on the result in 2004. And regarding corruption at county tabulation - the state level projections (at close of poll) were actually more accurate than the WPE, which does seem to nail the discrepancy down to precinct level. In general, the selection of precincts within each state seem to have been good predictors for the state result.

Have you or E-M compared the "at the precinct" counts to the final official counts for the 60% precincts where it can be done?

BTW, thanks for the info (and of course for all the time and trouble) about the uniform fraud question. I don't have anything to add right now but just wanted you to know I've read it carefully and will be digesting for a while.

I don't have that data, and I would be surprised if it had been done by E-M.

It strikes me as being something that needs to be done at every precinct i.e. not just NEP precincts, as a check on election fraud, but by the same token, given that it would be so easy to check, county tabulation fraud in jurisdictions where precinct vote totals are available seems like a heck of a risk for any election stealer to take. But perhaps I underestimate the chutzpah of the election stealers.

And it's all complicated further by the myriad ways in which absentee totals tabulated, and one thing I do agree with Kathy Dopp about is that BoEs should give a breakdown of where each subtotal of votes comes from.

Let me know when you've done some thinking - PM me if I'm not around!

I agree it should be done everywhere in the future but was thinking that the E-M data may be the only data we have for that purpose with regard to 2004 (not sure though).

wouldn't it tend to show up as a red shift in the precinct sample?

That is, consider the pure case where the exit poll is unbiased, the precinct counts are correct, E/M records them all at precinct level, and then the results are hacked centrally. We would expect essentially zero mean Within Precinct Error -- the exit poll should (under these assumptions) match the precinct-level returns within sampling error. But the estimate would be way off, which would be manifested as a discrepancy between Kerry's (precinct-level, pre-hacked) vote share in the exit poll precinct sample and his overall (hacked) vote share.

Ever since the E/M evaluation report came out, AFAIK all parties have conceded that there is no sign of general red shift in the precinct sample. In fact, the average shift is slightly blue: Kerry's performance is a bit better in the overall official returns than in E/M's figures for the exit poll samples.

I think this is a problem for massive-hacking scenarios unless we assume that the hackers had access to a list of all exit poll precincts (perhaps even all the precincts included in E/M's quick-count model) and avoided those precincts. So, instead of seeing the exit polls as evidence of multi-million-vote fraud, we are then rather frantically trying to explain why they aren't decisive evidence against multi-million-vote fraud.

EDIT to clarify subject header

I'm assuming that the hackers didn't have access to the list of exit poll precincts and that they hacked a lot of precincts without regard to whether they were exit poll precincts or not.

In that case the precinct sampling of the exit poll will appear to be good sampling because the official count of the sample will tend to be fraudulent to roughly the same extent that the official count of the full population is fraudulent. Or am I missing something?

On edit: Hey OTOH, didn't notice that was you and not Febble. You should wear glasses or something so I can tell the two of you apart.

(in each precinct) -- either a quick count obtained at precinct level (which may not match the final official count from that precinct, if the official count is hacked), or an official count at the county BoE or equivalent. So any particular centralized hack favoring Bush should show up as red shift either at the precinct level (if E/M uses the hacked BoE result), or in the precinct sample (if E/M uses the precinct-level quick count, which by assumption is not fraudulent).

Or am I missing something? Perfectly possible, especially since I woke up at 4 am and couldn't get back to sleep. E.g. I may be rebutting an argument you aren't even making. Bear with me.

And if I can stick with the more pure case for a minute, it seems to come out to these scenarios:

1. Bushhackers were indiscriminate and hacked all precincts without regard to whether they were exit poll precincts -- should produce no redshift or blueshift in the precinct sampling.
   
2. Bushhackers avoided exit poll precincts -- should produce blueshift in the precinct sampling.
   
3. Bushhackers perversely hacked only exit poll precincts -- should produce redshift in the precinct sampling.

Now to the slightly-less-pure case where E-M captured official counts from the precinct in 60% of cases and from the county in 40%. Assuming Scenario #1 above, then for the 40% segment we should get redshift and for the 60% segment we should get noshift.

If I've got that right then the way of detecting fraud in the slightly-less-pure case is that there should be correlation between the count-level attribute of a precinct (precinct data vs. county data) and redshift.

Edit: that is, if our case is still relatively pure and the only kind of fraud is county-level. I have to think about what happens when you introduce precinct-level fraud.

If E/M is using unhacked precinct-level numbers, and the Bushhackers are hacking the results at the BoE level, then E/M's numbers should point to a much better Kerry showing than the official returns, i.e., red shift in the precinct sample.

You seem to be having it both ways, with E/M using unhacked quick counts to calculate "no red shift in the WPE," and then using hacked official counts to calculate "no red shift in the precinct sample." No?

let's define some new terms:
{E-M redshift} is the difference between the exit poll estimate and the {E-M official count}.
{actual redshift} is the difference between the exit poll estimate and the {actual official count}.

In the pure case there is no {E-M redshift}. There is {actual redshift}.

who was missing something (I think).

When considering whether their precinct sampling was good they would have been measuring against final official counts at some level (is it state level?). That's the point I was missing (I'm hurrying to get in the car to go to work so don't think your the only one with an excuse).

as long as we are inventing new terms, we are gonna have to do something about "exit poll estimate," because I can only guess what you mean by that.

"{E-M redshift} is the difference between the exit poll estimate and the {E-M official count}."

You are saying, I think, that in the pure case, the exit poll proportions should (on average) match E-M's quick counts, zero mean WPE.* I agree.

* EDIT TO ADD: Or actually maybe not quite zero mean WPE; at least the expected median should be zero.

"{actual redshift} is the difference between the exit poll estimate and the {actual official count}.... There is {actual redshift}."

Right, and that's what would show up as red shift in the precinct sample, based on the discrepancy between the official final margin and the margin estimated from the quick counts in the precinct sample. See pages 29-30 in the evaluation report. No?

In our pure case, the exit poll sample would be representative of the true count. The official count used by E-M would be equal to the true count. Thus there would be no red or blue shift on average in the WPE. The final official count would shift to the red.

So, yes, I think you are right but would like to look at how E-M evaluated the precinct sampling and to think on it some more when I have a chance, just to be totally convinced.

Can you check my understanding of those figures on pages 29-30? They are not just a simple summation of the sample precincts. Rather, they are a projection that uses the sample precinct data as input and then runs them through some model that has been developed empirically over time, with the final product being an estimate of the state level count. The model amounts to applying a weight to each precinct, taking a weighted average and then scaling up to make it statewide. There is no opportunity to do age-race-sex weighting directly because the official count data doesn't include those voter attributes. Instead it attempts to adjust for those factors by way of precinct weighting that has been developed by trial and error over some period of time and effectively the precinct weighting is a proxy for age-race-sex weighting. Perhaps the model is more complex and takes into account registration data for the sample precincts, registration data for the state and some dynamic adjustments based on expected versus actual turnout. I'm sure I have some details wrong if not the whole idea of it -- feel free to set me straight on any of this (as if you needed my permission).

If I am not totally off base on how those estimates are done then it seems that fraud of certain types would not show up as redshift. Specifically, fraud that is chronic and endemic to specific locales would not show up as redshift because the model would have been adjusted over time to correct for it. I don't think I'll venture much farther until you can confirm or correct my basic understanding.

I haven't tried to figure out the specifics of the model -- they aren't all public anyway. But the basic aspect is a comparison with recent previous results in the same precinct. I don't know how subtle they get in modeling changes in turnout.

"Specifically, fraud that is chronic and endemic to specific locales would not show up as redshift because the model would have been adjusted over time to correct for it."

Well, I don't know that there is a whole lot of "adjust(ing) over time" that would apply to "specific locales." If a precinct is 90% Dem, but it was hacked in 2000 to be only 80% Dem -- and in 2004 it yields a quick count that is 90% Dem, but it and other similar precincts are hacked to be 80% Dem -- then it will still contribute to apparent bias in the precinct sample, because the precinct sample (based on quick counts) will be bluer than the official returns. I don't think the modeling will alter that basic result.

I agree with the basic result that the precinct sample (based on quick counts) will be bluer than the official returns of those same precincts. That fact would be revealed if you check the quick count of those precincts against the official returns for those precincts, either one by one or at least just for the subset total. Did they do such a check? I don't see it in the evaluation report.

If, on the other hand, the only check they did was against the statewide totals (as I understand the check on pages 29-30 was) and was therefore a check that came only after the precinct sample was leveraged up to statewide proportions based on a weighting system that uses historical data and some art in addition to science then I don't see how we can be sure the leveraging up won't adjust away chronic fraud. I thought the idea in the model was (at least to some extent) that they hand-picked bellweather precincts and then weighted them in a way that tended to produce a good prediction without necessarily knowing why it produced a good prediction. It seems to me that such a model, because its only justification is that it seems to work from one election to the next, would inherently adjust away the redshift if it is chronic.

First, no, I really doubt that they systematically checked their quick counts against the official returns. It would be a pretty sad commentary if E/M were the best source of precinct quick counts, but I'm not exactly wired into the Kerry campaign, so I wouldn't know.

"I thought the idea in the model was (at least to some extent) that they hand-picked bellweather precincts and then weighted them in a way that tended to produce a good prediction without necessarily knowing why it produced a good prediction."

Well, they don't hand-pick bellwether precincts by any means. They use a particular sampling method that is designed to give a clean 'slice' through the partisanship distribution, stratified by geography and by broad size of precinct. It's not as if they are hand-tuning 'take precinct Auglaize AFZ and subtract 5% -- that was pretty good back in 1992.' If there are any ad hoc adjustments to compensate for past problems, I haven't spotted them. They probably downweight outliers, but I don't think that will wildly change the results, at least in 2004.

It's certainly possible that some technical choice is being inadvertently influenced by a fraud pattern in the data. I just don't see how it could "inherently adjust away" chronic redshift, for the reason I tried to express in my previous post. If the quick counts ride above the official returns, they ride above. The whole point of the model is to determine whether, and how much, interview and/or vote count data are riding above or below past results -- not to obliterate the differences.

Caveat: just because I don't see it doesn't mean it's not there.

If the quick counts historically ride above the official returns, wouldn't the model apply an adjustment factor to them in an attempt to get the best projection of the official returns?

What would E/M do if year after year, its quick counts pointed to one outcome and the official county totals pointed to another outcome (not necessarily a different winner, but substantively different proportions)? and has it done it?

I think it would be hard for smart quants to make a model decision that would have the effect of adjusting quick counts to align with official totals, without noticing that they had done it. That is especially true if we are talking about a model that treats the quick counts as official totals. If I understand rightly, E/M has one model that works with precinct-level data (interview data and quick counts), another that works with county-level totals, and an Integrated Model that combines them. The IM could make a practice of goosing up the precinct-level estimates to align with the county-level estimates, but eyebrows would rise. It's hypothetically conceivable that at some point E/M noticed that many of their quick counts were misaligned with official counts from the same precincts, but it seems like a very far-fetched scenario to me, not least because lots of people have access to quick counts (where they are possible).

I'm having a hard time figuring out the scenario I am supposed to be thinking about, so I'm not making much sense even to myself -- but basically, given that E/M basically treats the quick counts as official totals at the precinct level, I just can't see them blithely "apply(ing)" an adjustment factor to the counts.

Having said that, I can't slam the door too hard. There are lots of design decisions in the sampling and the models (including Quality Control), and any of those might be subtly influenced by historical patterns. Hmm, I'm struggling to find a way to say the next thing without being misinterpreted. In the history of the 2004 election debate, some folks have occasionally suggested that pre-election polls and/or exit polls might somehow be calibrated to false assumptions about the accuracy of the count. I think it's an inherently interesting idea -- it's always intriguing to bang on underlying assumptions. But to date it seems to have inspired more vacuous special pleading than anything else. Generally, when I see the first few people dive down a hole and never come out, I don't encourage anyone else to follow them. But Frodo could handle the One Ring (up to a point), and maybe eomer can handle this.

As my guide, you don't know a back door into Mordor by any chance? No? Then you must not be Gollom. I should check with Febble, I remember her saying she's been there before when she was in the service of Mitof... I mean Sauron. :evilgrin:

Back to reality and the original question, which was whether fraud could exist in that 60% segment of precincts without being detected by the exit poll, let me stipulate for argument's sake that the exit pollers would have the ability to detect it if they are looking for it and that they probably would be looking for it. But would tell us if they found it? They don't make a habit of publishing that kind of detail about the model, I take from your comments.

So, as just a citizen without access to the inner workings, I don't take the exit poll reports to be evidence one way or another on this question and will continue to have questions, although I'll stop probing in this particular direction for now since I'm outside the great gate and can only hear drumbeats, chanting and gnashing vaguely in the distance inside.

Edit to add: in my version of the trilogy, Febble turns out to be the Lady Galadriel _{(Of all the wise, Galadriel is possessed of the greatest foresight and bestowed upon Frodo several gifts to help him in his quest, including the Phial of Light.)} The election turns out to have been stolen, largely through vote suppression but also through low-tech vote flipping, stuffing and disappearing, some degree of machine fraud, but at the electoral college level and we are left never knowing who won the popular vote (movies should give you something to argue about at dinner afterward).

Certainly the swing-shift correlation, if it was done, wasn't reported in the E-M evaluation. It isn't that E-M weren't prepared to entertain the fraud hypotheses, and indeed they reported results from there analysis of WPE by voting method. But they had strong a priori reasons for suspecting non-response bias (they had evidence for it in prior elections; there is experimental evidence for it; they were aware, even before a single vote count was in that their results were looking "bluer" than their pre-election estimates), so I don't suppose anyone else in Mordor was looking as hard as I did.

But the reason I ended up in Mordor was because my paper had actually criticised the E-M evaluation for using a flawed measure, and called for a reanalysis:



And of course that was what I was contracted to do.

So the question then becomes: was I looking for it and would I report it if I found it? Well, I was, and I reported to Mitofsky that I didn't, and he made my finding public.

But was I telling the truth? :evilgrin:

See the edit I apparently crossposted past your response.

The current question I was discussing with OTOH was different than the one you looked into. It was whether the conclusion that the sampling of precincts was good (not sampling within a precinct, but rather the selection of which precincts to poll) was also conclusive that there was no fraud hiding in there. You looked at WPE, this is a question about "not-within-precinct error". You haven't looked into that area for Mitofsky have you?

And I'm thinking about your other question too, although it wasn't part of what I did for Mitofsky (which only concerned within-precinct error).

That's the confusion I see over and over again on this board. Too many people discount straightforward, verifiable factual statements by attributing them to some evildoer, or by asserting that only an evildoer would say such things. Not that that is happening here. (And not that everything is straightforward and verifiable.) The exit poll argument "Ring" doesn't confer invisibility -- although I do seem to hear a great many plaintive complaints that the media can't seem to see it. It does seem to instill, in some, an exaggerated sense of power and a fast-creeping paranoia. And Sauron -- whoever Sauron actually is -- probably cheers when his/her/their opponents put it on.

(BTW, I really enjoy your edit, and I never take offense at your humor. I hope you don't take offense that my response is somewhat serious despite the whimsy. Here's to the Lady Galadriel!)*

I'm not sure, but I think you lowered the bar a lot for yourself in your second paragraph. I don't think it even matters if exit poll analysts are looking for fraud, per se. It makes perfect sense to me at least to ask whether the precinct-level red shift in each of the last five presidential elections owes in part to long-standing corruption (although for several reasons I don't think that story works very well). It makes very little sense to me that analysts are staring at -- and acknowledging -- precinct-level red shift in various elections while somehow obliterating red shift in the precinct samples. All the less because as far as I know, they don't have exclusive access to any quick counts.

I guess my advice would be, don't obsess over secret models. If the quick counts diverge from the official returns, there ought to be some other way of proving it, AFAICS. And do bear in mind, the exit pollsters look at a fraction of one percent of all precincts.

* EDIT TO ADD: By the way, now that I have gotten to meet Febble, I can confirm that she is much more interesting than Galadriel. Tolkien famously seems to have had some trouble imagining and writing credible female characters.

but, being an actuary, I find myself coming back to them during my private fantasies.

At this point I think what I am doing is discounting the secret model as telling me anything at all and going back to my default position that there may be fraud in that 60% segment since I have no way to tell that there isn't. But I do agree with your point about the quick counts being available without regard to the exit poll and, in fact, am thinking I may focus my efforts in November on collecting quick counts locally. I need to check what the rules are in Florida and what is already being done and will go from there.

My "honey-do" list is calling me this Saturday morning -- I'll check back later.

I could quibble over whether it seems reasonable to discount the precinct sample results altogether, but if they are inconclusive, there probably isn't much point in debating how inconclusive they are. ;)

per machine. But that would not require a machine. Machines were not everywhere.. let's say they were in a third of the districts.. so it would take 1.5 dems to not vote per poll-equivalent-to-numbers-a-machine-can handle.

That's even less!

And let's say that Kerry being given the win at 2PM on election day.. would force otherwise non-voting repukes out to vote.. then it would take less than one Kerry voter to stay home.

There is more than one explanation for exit polls.

We don't have the hard evidence yet. So it is all speculation.

Is that why there were long lines at closing time in Dem areas?

Who announced Kerry won at 2pm?

And wouldn't just as many bush voters stay home knowing bush had lost?

All in all, not a well thought out post from you, applegrove.

It looked like a win. Then all sorts of people went out at the last minute. Every wonder why the news organizations no longer do polls or will fund exit polls? They got burned.

It is an open question.

The rumor has gone around that the news organizations won't do exit polls, but they say otherwise. (It's certainly true that they got burned.)

I've seen no evidence that the exit polls had any effect on Kerry turnout -- although they may have influenced GOTV operations somewhere, I don't know.

given any excuse.. will not vote.

to Diebold and ESS.

at the poll. Nice try though.

So now we're talking every machine. Or as many as possible. That's one state after another and requiring widespread communication. More people involved the greater chance the scam is revealed. Many of those states a Democrat will be in charge of elections and therefore control voting methods and machinery. The machines themselves vary from state to state and even within the same state.

Are we trying to fix the popular vote or the electoral college? Maryland has Diebold. Likewise Georgia. Many other examples, in so-called safe states. What good does it do to switch votes in those states? If you don't do it in those states, and there was no evidence of the percentages deviating from expected, that ups the number of switched votes required in the key states.

Plus, you're taking the chance of deviation from relationship to other partisan races. Are all the presidential D votes that are switched also flipped at the senatorial level and governor level?

Plus, how do we know ahead of time how much theft is required? This is lovely and pristine retro adjustment two years later, knowing the 2004 numbers. By definition the scam has to be planned and put in place before the votes are cast. What if the outer WPE margins from Steven Freeman had been the actual will of the people on November 7, 2004? It's quite amusing in this thread that the same posters embracing Freeman and his theories are now conveniently abandoning them, saying only 1.5 million votes needed to be switched. If Freeman's estimate of Kerry winning nationally by 4.6% is true, then the machines had to be adjusted by substantially more than 5 votes per machine.

Besides, and I doubt I've ever mentioned it in this forum, but you guys are wonderfully ignorant of what a fix is. My background in sports and Las Vegas provides a massive boost in this regard. A fixed game doesn't fall close to the number. It's cupcakes. Whenever true fixes have been uncovered months/years later, the results of the games inevitably are coast jobs, not threatening the pointspread. Only a stone moron with a defeatist mentality watches a 30 foot bank shot go in at the buzzer, covering the spread by 1/2 point, and screams fix.

Here the stakes are not a meaningless college basketball game, but control of this country and essentially the planet. The GOP has fix capability on the electronic voting machines, any number they want to create. And they let the election come down to one state and a swing of 60,000 votes? How hard am I allowed to laugh? You want me to believe Rove and Co. fixed the election, and presumably knew they would be doing that for years in advance, yet they're being handed the same early exit polls we were getting, then frantically checking early returns in small Florida counties to estimate the actual vote? Fine. I'm happier in the real world.

The programmers know, and their bosses know, but that's about it.
The technicians who do the updating of the machines don't need to know, the people operating the machines don't need to know.

If we are discussing the actual 2004 election, then where specifically do you think votes were stolen, and how many? For instance, are you alleging that "fraudulent program code" was deployed to miscount punch cards in Ohio?

the US election system is so chaotic and messy. To try to relate it to single precincts is almost impossible because of the huge diversity of voting methods and a dozen other variables. The overall exit polls are showing fraud it seems to me, BUT EVEN IF THERE'S NO EVIDENCE OF FRAUD, THE FACT IS THIS:

When the vote is counted by extreme partisans of one side in total secrecy using trade secret, proprietary software that is either impossible to audit or is almost never audited and the electorate is required to accept the result without any means of verification, THIS IS BY DEFINITION NOT A DEMOCRACY.

Leaving aside the statistics, do we want a democracy or not?

Well said.

Not only should Americans want a democracy, but the rest of the world needs America to have a democracy that works.

And I wish the movement would leave behind the statistics, because they are totally irrelevant to the principle, as you so rightly say.

They also suck, mostly. (Mine less than most, of course).

20 votes on 75,000 machines
40 votes on 37,500 machines
80 votes on 18,750 machines

160 votes on roughly 9,300 machines in LARGER counties and precincts. You don't even have to mess with small precincts in small counties, where someone might be more likely to notice. You can achieve some of these results in these precincts by finding the means to deny some voters the vote, as they are more likely to vote for Kerry in these precincts.

if its not to late.

Why is a fraudulent election allowed to come down to the single state Ohio, and a potential swing of 60,000 votes? No one even took a swing at that in my previous post, one mentioning that fixed sporting events never fall close to the pointspread.

Are our opponents that incompetent, that they control the machines and therefore the outcome, yet can't figure out how to win comfortably? They need to hire me.

Here's an inconvenient fact that's seldom mentioned: all the focus on Ohio is allowed because the 20 electoral votes BARELY put Kerry over the top at 252+20=272. The magic number is 270. Meanwhile, we won New Hampshire's puny four electoral votes by 9300. Why didn't the GOP steal that state, making Ohio irrelevant? TIA never wanted to answer that question, but maybe someone can in 2006.

Seems to me, if I'm embarking on a conspiracy of this magnitute I want the most margin for error possible. Multiple chances, not one. Maybe Rove loves a challenge.

It's obvious they tried in New Hampshire. Freeman wants 57% for Kerry there, via exit polling. Plus I remember the Nader-backed mini recount in New Hampshire, where a female statistician isolated 11 specific urban precincts, using Diebold optical scan machines, that she claimed voted suspiciously compared to the rest of the state, changing dramatically from 2000 to 2004. New Hampshire has 301 precincts; 126 of them used Diebold's AccuVote technology in 2004.

Somehow the recount produced no significant changes, nothing different from a recount of punch cards. There's that competency factor again, not cheating well enough.

Strange, because we have verified instance of the GOP cheating in the same state in 2002, phone jamming the Democratic GOTV operation in New Hampshire on the morning of the vital Shaheen/Sununu senate race. That phone jamming case has produced two guilty pleas, one conviction and a forthcoming trial this fall. Jail sentences for the two guilty pleas and the conviction.

Could it be the GOP has no control over the voting machinery at all, but resorts to low tech suppresion like phone jamming and purging voter rolls as a minor boost, schemes to potentially turn a narrow defeat into victory, let's say by 537 votes? If so, and I realize how outer space and naive my theory is, once in a while you'll come up 9300 votes short, like New Hampshire 2004.

all that, I'm still trying to figure out why my Government allows companies to count our vote in secret. What are the odds? Two competing parties, looks ligit, but yet neither party cares to see how the votes are counted?