Febble: a report and an apology

[Febble]

A while back, USCV issued a paper

http://electionarchive.org/ucvAnalysis/US/Exit_Polls_2004_Mitofsky-Edison.pdf

that inferred from exit poll data contained in the E-M report, that “Bush stronghold had more vote count corruption”. At first I thought their argument was persuasive. I then became aware that the “within-precinct error” values (“WPE”) were themselves confounded with the measure of precinct partisanship (vote-count margin), and that this would create a mathematical artefact that could give the appearance of greater WPEs in Bush strongholds. In other words, the inference drawn by USCV might be an artefact of the math.

This, in fact, was primarily a criticism of the E-M report, which used what I now consider to be a poor measure of ”bias” (the signed WPE) in their analyses, not of USCV. I was, indeed, at the time a member of the USCV list, and was even a signatory to early versions of the USCV report. However, our views had diverged, and I therefore produced a body of work, initially in diaries posted on Daily Kos, and amply supported by other bloggers and commenters, in which I argued the nature of the confound, and a method by which it could be removed from the data. It finally appeared as a paper, linked from a front page post at Daily Kos, and Mystery Pollster.

http://www.dailykos.com/story/2005/4/29/161938/921
http://www.mysterypollster.com/main/2005/04/the_liddle_mode.html

I also acknowledged the role of USCV in developing the ideas, even though they disagreed with my conclusions.

In essence, my paper called for E-M to use my formula to remove the confound from the WPE measure (using my, or a similar, formula), and to look again at their data, particularly at the slope between vote-count margin and bias, to see if it was indeed there, once the confound had been removed.

Someone at DU also inked to my paper and I entered the discussion as a new DU member.

Meanwhile, in what seems to me to be yet another tribute to the power of the blogosphere, Warren Mitofsky heard this cyber-conversation. He used the formula, and investigated whether it got rid of the apparent slope. He let me know that the formula did indeed get rid of the slope, and that he would probably be presenting this information at AAPOR.

This put me in a difficult position. SunshineKathy (Kathy Dopp, a USCV author) on DU was promising a “rebuttal” of my paper any time now. I knew that the data itself confirmed that no significant slope existed. I continued to argue my case on its merits. I also even attempted to level the playing field by asking USCV to wait until AAPOR before concluding that USCV was correct in inferring a slope. After all, I was not saying anything other than that it was possible there was no slope. USCV were making the stronger case that there must be a slope. Regardless of the answer, I had the winning ticket.

Then things changed. Just over a week before AAPOR, Mitofsky asked me to do some further analyses. I largely stopped posting at this stage, only dropping in to DU to continue some existing dialogues (some DEMANDING a response), and again, only arguing from logic and publicly available data, except once, ill-advisedly, suggesting people wait until AAPOR before deciding whether or not my case could indeed by rebutted by the new USCV paper . But I should have kept out completely. I apologise. My only defense is that I have found the attack dogs at DU quite upsettingly vicious. I knew that not only was my logic defensible, but that it was supported by data that would shortly be made public.

My final post was to draw people’s attention to the Mitofsky findings as soon as he made them public at AAPOR. Again, I should not have done so, but I was relieved to have these confirmatory findings finally in the public domain.

They can now be interpreted on their merits. They may suggest alternative fraud scenarios to the one originally proposed by USCV. In fact, as a direct result of my involvement in the debate, far more informative data is now in the public domain. There is an excellent thread on DU, with the scatterplots included, that allows anyone to have a look and draw their own conclusions. And as I have said on many occasions, including in my paper, my “bias” index in itself is fraud-neutral. It does not distinguish between bias in the poll and bias in the count.

The DU thread is here:

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=369886&mesg_id=369886

I am also aware that the authors of the current USCV working paper may not think that the absence of the slope refutes their contention. If so, I am frankly unclear as to what they are saying and why they ever took issue with my position. I said there might not be a slope. USCV claimed that I was wrong. If they are not arguing that there must be a slope this whole geeky argument has got completely out of hand.

However, I apologise once again to members of this forum for remaining in the discussion at DU after I found myself in position where full disclosure, which was not then possible, was ethically required. I regret this. I plead both naivete and anger in defence. I am not a “Mitofsky shill”. It seems to me a tribute to Mitofsky that a) he listened to the blogosphere, and b) he asked a known “fraudster” (in the sense, I hasten to add, of one who believed and still believes that the 2004 election was both corruptible and corrupt) to conduct further analyses on the data.

It remains for you to judge what that last sentence says about either of us.

Lizzie

Links for my fraudster credentials and other info:

http://www.dailykos.com/story/2004/12/12/72758/130
http://www.dailykos.com/story/2004/12/1/5429/18931
http://www.dailykos.com/story/2004/12/19/9056/8130
http://www.dailykos.com/story/2004/12/1/5429/18931
http://www.dailykos.com/story/2005/1/3/122338/7617
(This was a diary commenting on a magnificent paper by DKos poster, Georgia10, chronicling the Ohio debacle -Georgia'a diary, with links to her paper are here:http://www.dailykos.com/story/2005/1/2/19512/470130 )

http://www.dailykos.com/story/2005/4/2/174011/1181
http://www.dailykos.com/story/2005/4/2/174011/1181
http://www.dailykos.com/story/2005/4/2/174011/1181


You might also want to check out my papers on the USCV website here:
http://electionarchive.org/index.php?option=com_content&task=category§ionid=4&id=87&Itemid=63

They include work on voting machines in Florida, spoilage in New Mexico, and voter suppression in Franklin County, Ohio.

Regarding other credentials: I do not have a PhD, but hope to submit my dissertation by the end of July. Otherwise google me. I am a musician, writer, and psychologist. And a data analyst geek.

[ibid]

1. Hang in there Feeble - you have a lot of fans at DU :-)

And I hope you find a job you like post-PHD!

Perhaps on data mining and financial derivative analysis - indeed combine those with getting past the Actuarial exams and I assure you that you will make enough to even live in New York City!

:-)

[TruthIsAll]

18. Using what? Monte Carlo, Black Scholes, Genetic algorithms or Neural Nets?

You must be kidding.

I worked as a financial software developer using all those models.

Febble would be kicked out of WS in a minute if she tried to fudge the simulation results in an executive summary like she's doing now, just to satisfy her boss.

I suggest you read the USCV report and download their Excel model.

[Febble]

19. self deleted

[LightningFlash]

21. I wouldn't be, but I do think you should give the data to USCV.....nt

[ibid]

53. "to fudge the simulation results" -???? - it is hard enough to explain

the results! :-)

I guess I did not see "fudge" - just some work that was not yet ready for the final memo. But then I did not see those first principles whose review lead to the formula change - so perhaps we are now at "fudge"! :-(

And you are correct that I am going from the DU summary posts to a conclusion because I have been too lazy to download the USCV report and download their Excel model.

In any case, as jobs, reduction in investment risk is still a highly paid game, despite the recent failures over at Morgan Stanley. But age Feebles age 53 may preclude a job change!

God forbid Feeble gets into polling! :-)

And TIA - you were into Monte Carlo, Black Scholes, Genetic algorithms or Neural Nets - WOW - I saw the pain that Bob Clancy went through in writing his paper back in the late 70's -and in writing his Fortran program, and I know why I hate APL - and why I did my Sharpes in Mathcad (it is still up on the Mathcad site!) - You have done some heavy lifting along the way. :thumpsup:

At this point in my life I can't fully/easily handle what Excel throws out as tools!

I do long for old days like the early 70's when a simple Monte Carlo result was all that one need to get people to be happy with the work product!

In any case I suspect that you and Feeble will end up best buddies - after the dust has settled - but then again, maybe not.

:-)

[TruthIsAll]

95. Here's a 5000 trial Monte Carlo Simulation using win probabilities

based on final pre-election state polls to project Kerry's Electoral Vote:

http://www.geocities.com/electionmodel/

[LightningFlash]

2. "Regardless of the answer, I had the winning ticket."

What do you mean exactly by the winning "ticket"?

How come Mitofsky contacted you in specific, but has never contacted any of the professors at USCV or Common Cause who have been using his data?

I have no doubt that you have made a difference here. I just can't help but have pressing questions though, about your apparent involvement with Warren Mitofsky and Mystery Pollster who have seemingly worked together on the press conference.

[Febble]

3. Because I'd hedged my bets.

My paper says there might not be a slope, not that there is not one.

USCV, as I understood them, said that there must be a slope (or the likelihood was that a slope existed).

So, whether I'd known the answer or not, I couldn't lose. If Mitofsky had shown a slope, I could have said, well I never said there couldn't be, just that there needn't be.

USCV were saying, as I understood, the regression line can't be flat. So I was in a win-win situation, they were win-lose. I had the better ticket. They were braver.

I worked closely with Mark Blumenthal on developing the ideas. Also with Rick Brady, and DemfromCT, a poster on DKos. All are acknowledged in the paper. I also corresponded with USCV members, some of whom I believe now support at least some of my conclusions.

As far as I know, Mark Blumenthal met Warren Mitofsky for the first time at AAPOR, but I may be wrong. Rick Brady certainly did.

Mitofsky contacted me because he had read my paper. The joy of the blogosphere is that things get known.

I have never met him, and until a few weeks ago, only knew of him as the director of Mitofsky International. I am not a pollster, nor a member of AAPOR. It is not my field. If I am anything, I am a data analyst. I analyse data. I write geeky papers, children's books, and music. I'm a Brit, living in Britain and I'm 53.

[eomer]

4. Febble, correct me if I am wrong, but I think the problem is semantic.

Mitofsky has not proven that there is no slope in the WPE regression line. He has proven that there is one.

You came up with a transform that, when applied to WPE, creates a new data set called WPE Index.

The WPE Index regression line is flat. The WPE regression line has a slope. So when you and Kathy exchange jabs of "It doesn't have a slope" and "It does so", I believe you are talking about two different data sets.

The curves shown in the latest USCV paper all are labeled as depicting WPE, not WPE Index. I haven't slogged through their appendices but I assume you have. Do their graphs depict WPE or WPE Index? And do you agree that the WPE regression line has a slope?

[Febble]

6. Well, it's not semantic, exactly!

It's the whole point of my paper.

The WPE appeared to have a slope. USCV inferred a fraudulent pattern from the slope. At first I agreed. Then I realised the confound. I therefore devised a formula which Mitofsky calls the WPE_Index, and I called the Bias index. I argued that this would get rid of the artefactual slope given by the WPE. Ron Baiman applied his own version of the transform to the means and medians in the E-M report, and found, as I had done myself, that this seemed to strengthen the case for a slope. However, I kept digging, with the help of both Mark Blumenthal and Rick Brady (who I realise is not exactly Man-of-the-Year at DU), and discovered that applying the formula to the means (or medians) was too late. They had to be applied at the level of the individual precincts, otherwise the artefact remained. That's what the computational model was all about. I generated data, and analysed it both levels, and found only if the transform was applied at precinct level would it get rid of the slope.

So yes, the WPE_index gets rid of the artefactual slope, visible in the WPE plot. This is what I said might happen. USCV, or at least four of the original authors disagreed with me, or at least billed their paper as a rebuttal.

I think Ron Baiman still maintains there is some kind of slope, and you can read a bit of exchange between us on MP now. I really don't know what he is saying, but frankly it looks completely illogical. I expect he thinks the same of me, though.

But as far as I am concerned, the data to me looks as though there was no linear relationship between bias and vote-count margin. There was also one heck of a lot of variance, with some pretty extreme precincts below the line as well as above.

Widespread bias or widespread fraud. In both directions. Take your pick. Maybe the Bushies were just a bit better at it than the Democrats. Or maybe polls are a very blunt instrument.

Noisy, certainly.

[TruthIsAll]

9. Yes, Febble: "Noisy" polls, screaming out that Bush stole it..

A question:
How did Mitofsky "correct" for WPE in the 13660 final National Poll, giving Bush 51-48 when at 13047 it was Kerry by the same split?

If we are to believe you, then all exit poll losers/recorded vote winners from the past are justified in claiming nonresponse bias for their opponent. And the same goes for all future stolen elections.

You want to give them a free ride.

I know.
You think polling sucks.
It can never prove fraud.
Febble, it doesn't have to.
But it points us in the direction of Truth:
Massive fraud.

You're logic sucks.

2-4-6-8; EIGHTY-SIX OF EIGHTY-EIGHT!
Any comments on this calculation?
http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=369374&mesg_id=369374
Would you not call it fKr: Frustrated Kerry Responders?

In fact, the rVr theory has been floated only a few times by Mitofsky for as long as he has been exit polling.
And he did so at least TWICE in favor of Bush.

Remember the 2000 primary: Buchanan vs. Bush?
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x369999

The FL 2000 exit polls said Gore won, by the way. Until John Ellis called it for his cousin GW on Fox. The reversal was based on the fact that 16022 votes for Gore were mysteriously dropped from the Diebold database in Volusia county. And 10,000 of them went to the Socialist candidate.

Dan Rather went along with the computer "glitch" explanation. Oh sure, they fixed the error later - long after the damage was done.

Kind of reminds me of Mitofsky's mainframes going down around 11pm.
And the VNS "system problems" on election day 2002 which prevented them from releasing exit poll results.

In 2000 FL, 185,000 spoiled punched cards were 70% Gore voters.
Can we agree it was rGr: Real Gore Removals?

But those are only historical FACTS, Febble.
You prefer to deal in outlandish hypotheticals.
Just like your boss, Mitofsky.

And Febble, here's a question for you:
Why must we assume 100% rVr were rBr and not rKr?
Why not assume rBr/rKr at 50%/50% and be done with it?

Febble, your reputation has been permanently tarnished. You can't put the toothpaste back in the tube. The simple truth is worth a lot more than your Fancy Febble Function.

No mea culpas, are necessary.
Don't claim ignorance.
That's just like Bush.
That's Bushit.
That's Febble Fiction.

I give you credit for trying your best to continue the obfuscation.
But who are we to believe: Your scatter plot logic or our lying eyes?

You are being quite naive to think that we will be fooled by that pretty picture and your false interpretation of it.

[LightningFlash]

14. Makes a good amount of sense..

It is obvious the original point has been refuted, being that consistent mean bias caused this result and is the reason for the analog exit poll discrepancy.

Seemingly Febble basically admits this now.

But the real question is, since this simulation has been used in past elections before and treated with the same secrecy by Mitofsky, what is it going to take to get all of the real data released from him?

All of the questions and WPE percentages in their official, complete format. WHY has it not been given to the scientists, WHAT is the reason for stonewalling and surely as anything they want to prove Bush won don't they?

Eerily, I feel they are deliberately hiding the data to save face. Ironically the precinct results do prove there was calculated and complete fraud, as well as the usual response bias.

Otherwise in any other circumstances I'd see the exit poll company would be aggressive in releasing their information for inspection immediately.

[Febble]

16. "Constant Mean Bias"

is what I have been arguing may exist, whether in poll or count.

I also argue that it is supported by the flat relationship between my bias index and vote-count margin.

It is what is disputed by Baiman, Dopp et al at present.

I'm not quite sure what I am supposed to have admitted.

[eomer]

11. Okay, but...

You made a big deal in the above posts about the supposed fact that you won and USCV lost (not too good at apologizing I guess). Whatever Ron Baiman is saying now is not what you were talking about in that win-win, win-lose stuff. You were talking about the previous jostling with USCV. If you are going to insist on claiming that you won and they lost, then please quote exactly the statement of theirs that you dispute. That way we can judge what they really said. Or better yet, drop the focus on whether you won or they won on some particular detail. Especially when you are apologizing.

[Febble]

12. It was a metaphor.

Meant ironically.

(sigh).

It is so difficult to be understood around here.

I played safe. My position was that either could be true. USCV went out on a limb. As I said (I think) - their's was the more courageous position. Foolhardy maybe, but certainly courageous.

I'm not saying "I won" in anything other then the sense that I won the lottery. I had no idea whether there would be a slope or not. I would have far rather that it had not been an argument.

I can only say what my position is. Kathy and Ron have repeatedly said they disagree with it. I THOUGHT they were saying that the data couldn't have constant mean bias. I think they are still saying this. All I ever said was that it could. And it does.

[autorank]

94. Self delete

[OnTheOtherHand]

13. allow me...

Here's the first sentence from the abstract of the USCV working paper as of yesterday:

"New evidence from mathematical simulations conclusively shows that any constant mean exit poll response bias hypothesis such as the 'reluctant Bush responder' (rBr) hypothesis is not consistent with the pattern shown by the Edison/Mitofsky exit polling data."

Well, not so much.

There are a lot of arguments to slog through, but most of them consistently have come down to the assertion that the error rates in the high-Bush precincts were Just Too High. USCV has recently added the argument that moreover, the error rates in the high-Kerry precincts were Just Too Low. The WPE Index scatterplot really makes both these arguments hard to buy, because if they were true, it too should show a positive slope.

[LightningFlash]

15. And your WPE index scatterplot shows what exactly?

On what data are you even basing these assumptions on?

Accordingly the WPE does not go through on both sides, but actually rises disproportionately to the original Mitofsky reported data.

If it conflicts with its own data sources, I see no conceivable way that could show the mean has produced a constant bias.

The bias tilt being argued here is impossible according to its own calculation. Why hasn't Mitofsky actually released all their ready data for the real scientists to study? Surely there's nothing to fear.

[Febble]

17. Check the plots

on the DU link in my post.

The WPE has a slope, the bias index does not. Therefore the slope is an artefact of the WPE.

The bias index, which measures the ratio between the two theoretical response rates (Kerry voters and Bush voters) and is not confounded by the margin itself (the WPE is a difference betwen two margins) is simply measure of bias, and does not distinguish between bias in the poll or bias in the count. It is the one that is constant across all degrees of vote margin.

[LightningFlash]

20. If it is a measure of bias only...

Then it doesn't explain anything, because both lines do not match up with the original Mitofsky data.

They in fact spike while going up the page using the real revised simulation models and the polling methodology.

http://uscountvotes.org/ucvAnalysis/US/exit-polls/USCV_exit_poll_simulations.pdf

Further, the percentages switch. This can not at all be explained by bias (another large hole poked in the theory) but must be re-configured in order to show an entire precinct and its results for one single election.

If this "data" for this scatterplot is to be studied, it should absolutely be released by those with-holding it. It should summarily be handed over to the team at USCountVotes and other so forensic analysis can begin.

There is absolutely no reason to be with-holding that data.

[OnTheOtherHand]

25. huh?

The lines are based on the actual results, same as the tables that USCV have reported, which they took from E/M's January report. If you think that E/M somehow rigged the scatterplot, then they might as well have rigged the tables -- and they might as well rig the raw data before releasing it. They can't win.

I'm not sure what you mean by "real revised simulation models," but the actual results are more real than the simulations, no?

The reason to withhold at least some of the original data is that to release it all would compromise American citizens' privacy and secret ballot. Every DUer should take that issue seriously.

[Febble]

26. It IS the original Mitofsky data.

That's why he presented it.

It's not simulated data, it's the data.

[LightningFlash]

28. Then you should have no trouble proving it.

Provide the data for others to inspect and see, and allow the simulation to be conducted.

In the recorded simulation, your data still ends up contradicting by going over the actual mean limit in both graphs on the USCV report which was just released.

You can only seem to benefit by releasing all of the data into others hands.

[glitch]

47. Peer review demands it.

It looks to me like E/M is trying to establish peer review credibility without actually going through the process.

That is not going to help the reputations of those involved.

[LightningFlash]

48. One sided peer review, where has it been seen before...

Iraq WMD. Pity we don't have much of a media as peer review then would have saved lives...

[glitch]

63. Absolutely - since there is no such thing as a one-sided peer review.

If it's one-sided, it can't be called a peer review. But I am encouraged that E/M is attempting the peer review illusion though, more evidence they are worried about whatever it is they are hiding being discovered, whether it's incompetence or complicity.

And a belated welcome to DU to you! :hi:

[Febble]

22. OK, I get it

"constant mean bias" was never my term, but I'll buy it.

USCV postulated that bias was greater in Bush strongholds. My position was that it was probably randomly scattered everywhere, i.e. not systematically greater at the Bush end of the plot.

That's why "constant MEAN bias" is the term - across the whole plot, the average doesnt' vary.

It varies enormously at any given point in the plot, with many precincts having large degrees of bias in both directions.

In other words there was lots of different stuff going on, all over the place.

[Febble]

8. Continued...

Sorry, didn't address your last comment.

Their paper changes frequently, and as I am NOT the author, obviously, wouldn't like to say. A WPE that produces a slope, but which is generated by a constant mean bias, should have a flat line when bias is derived from it - obviously, because the whole point of the bias index was to retrieve the underlying bias, and if it is constant, it will be constant.

The last time I looked they had a plot of ln(alpha) (aka known as WPE_Index or bias) and it looked a bit quadratic. This is either an artefact of the model, or the result of one particular run. I have run their simulator a few times, and mostly the line comes out flat. If it is doing anything else, there must be something odd about the model, because if you put flat bias in, you should get flat bias out.

It works with mine.

[eomer]

76. USCV is not using your linear approach.

If you fit a line to the transformed WPE (WPE Index) it does come out to a roughly horizontal line. The problem with this approach is that you can have something happening at each end of the line and miss it because it was averaged out or overwhelmed by whatever is happening in the middle.

From what I understand of USCV's latest paper, they are not forcing the data to be a line. They are looking at the shape of the data based on 20% segments. When looked at this way it apparently does not resemble a line.

Once they established the shape of the actual data then they tried various simulations to see which would create a similar shape. They found that a combination of vote shifting and a response bias that is different from rBr is at least one scenario that approximates the shape.

The response bias that they used in this successful simulation was one where Bush voters are more willing to respond in Bush strongholds and Kerry voters are more willing to respond in Kerry strongholds.

The other scenarios that they simulated (including rBr with no vote switching) all created a different shape than that of the actual data.

Do you take issue with this analysis? If so, how?

[LightningFlash]

86. In addressing this...

I have tried to ascertain where the response-bias is coming from and what the actual numbers were, that was posted at the AAPOR.

It seems that they all agree with Mystery Pollster's argument, that the result becomes a null flat line which is untrue. The result clearly spikes, or else it would nullify E-M's data. The spike does not match evenly, but indicates response bias and fraud have basically happened.

In MP's anlaysis using the WPE_Index and apparently new data, they are suggesting the spike does not occur and that it resorts to a straight line. This on its face appears to be an argument to dismiss exit polls, and there is clear flaws in the simulation. I would stake a real test of the entire precinct needs to be performed. Waste no time to create a new one.

[Febble]

88. As a statistician

I have problems with dividing continuous data into segments, especially when the data is normally distributed. Firstly, it means that the extreme categories have much smaller numbers than categories in the centre, resulting in much greater leverage per data point. Secondly the boundaries are arbitrary, and small adjustments to the boundaries, especially between the ectreme and the second most extreme categories, where leverage is high, can make a huge difference to the means. For example, you will notice that there is a very low data point that only just avoids being in the "high Rep" category - if it had been only a little higher, or the boundary had been a little lower, it would have radically pulled down the mean in that category. Neither mean would have much validity.

Thirdly, doing parametric statistics assumes normally distributed data. If you have sliced up a normal distribution, you no longer have normally distributed data within each category. There is therefore a limited amount we can infer from the category means.

However, I agree with Ron, that a coherent non-linear fit might be run through the data. The more complex the curve, though, the more degrees of freedom you need, and the probabilities become meaningless. In other words, you can look at the data, but you can't do inferential, probabalistic stats.

See my conversation with Ron on Mystery Pollster if you really want to be in on this geeky discussion.

http://www.mysterypollster.com/main/2005/05/aapor_exit_poll.html#comments

I also agree with Ron that a multiple linear regression would be informative.

Edit - sorry clicked post instead of preview....

I am not saying that you can't infer stuff looking at the data - those outlying points all over the plot clearly had weird stuff going on. But statistics cease to become a useful tool if you are doing this kind of observational work. The General Linear Model is one of the most powerful statistical tools we have, and it doesn't deal well with non-linear hypotheses on the whole (but we have ways...)

[TruthIsAll]

129. Great point. They want to assume a HORIZONTAL linear fit. Why?

WITH ZERO SLOPE, TO TOP IT OFF.

Yet, it's clearly non-linear.

IS IT THAT THEY WANT TO CHANGE THE FORM OF THE FUNCTION TO JUSTIFY UNIFORM NON-RESPONSE?

Is the function
Parabolic?
Exponential?
Quadratic?
Cubic?

WHY NOT FIT A SPLINE CURVE TO IT? A SERIES OF TANGENTIAL CUBICS.

EOMER, YOU ARE GETTING VERY CLOSE.

STRAIGHT-LINE THE BEST FIT?
NO WAY, JOSE.

ONLY THEIR ARGUMENT REGRESSES LINEARLY:

Y = - X

[LightningFlash]

5. Well that's very interesting...

That's a very interesting and well corroborated answer. Does this mean that Mitofsky will gladly hand over his real precinct level data if Stephen Freeman says he knows what the mean slope for divergance is?

Could someone possibly get him to release all the data by employing these tactic of guess, therefore receive option?

It is my understanding of his past history that, he only lets his data be reviewed by certain individuals he carefully alligns with. So if this was the case, someone would need to be in the know with him to be able to look over the data.

That way, fraud in the exit poll numbers could be conclusively proven (or disproven, which is unlikely since the vote is corrupted) and if Mitofsky was at any point complicit in covering up this uncomfortable truth, he could be convicted in court.

[Fly by night]

7. Believe me, you're not the only person attacked by a DUer ...

... who confuses sunshine with a huge natural gas flame. Don't be discouraged; you are part of a growing crowd of otherwise well-meaning election fraud researchers and activists to be burned by getting too close to that gas flame. Pretty soon, at the rate that the gas is spewing, that flame will burn out.

[Febble]

10. Thanks

And then maybe we can focus on all the African American Democrats in Cuyahoga County, Ohio who had their votes spoiled or who had to use provisional ballots, and those in Franklin County who were disenfranchised by the blatantly inequitable provision of voting machines.

And on Ken Blackwell's illegal obstruction of the recount.

[pgh_dem]

62. Hi! Sorry, this doesn't make sense to me

If a significant number of voters had their votes for Kerry spoiled, or voted for Kerry on a provisional ballot, which was disqualified, then who needs this rBr theory?

Maybe the * vote isn't underestimated due to non-response so much as Kerry's was honestly overestimated due to responses that just weren't true.

Maybe the variance of the exit polls from the actual vote count can be accounted for by the simple statement:

"The results varied because more people *thought* they voted for Kerry, and they told exit pollers that they had, but their votes were thrown out for technical reasons. Therefore, the introduction into the sample of people who unwittingly provided the pollers with false information caused an error which is multiplied by whatever weighting factor they applied to voters in said demographic."

This seems like something that could actually be investigated without so much 'mathifyin', at least to the point of asking "were voters polled in precincts with high spoilage/discounted provisional ballots?" and "if so, what were the responses?"

Then, all the analysts could pull out their calculators to see what effect, based on the percentage of spoilage/disqualification, has on overstating the Kerry support. Call them tivKr, for thought-i-voted-Kerry-responders. Test it against some of those precincts which had 4 hours lines but miraculously <40% 'turnout'.

It probably has been looked at, but without the actual raw data (not a scattershot USAToday chart), it looks impossible to know why the variance really happened, as opposed to hypothetical ways it could have happened, based on 'crumbs from the table' evidence from E/M.

I'm not sure how you mathematically could account for the possibility of releasing falsified data, call it MSDMF for "Mitofsky is a scumsucking dishonest motherfucker"...maybe someone has an idea. Either way, E/M has a responsibility to release the data for peer review, and everything else is conjecture. Next conference he shows up at, I hope someone's got a pie for him.

[Febble]

65. A lot of problems

as you say, can't be solved by math.

But if there was vote spoilage in a precinct, and it affected one demographic more than the other, it would certainly show up in the exit polls.

But vote spoilage doesn't need exit polls to be investigated. If you checkout the USCV link to the studies I did, you'll find a study I did with Josh Mitteldorf on spoilage in New Mexico. That was done with mathifyin'. It's what I do.

[pgh_dem]

89. Gotcha, but...

What I was asking was if this actually has been accounted for already?

I would figure that the baseline assumption for spoilage effects would be that it was random, affecting all equally, and therefore disproportionately affecting the majority demographic in a precinct. As such, it should already have shown up in the exit polls as a deviation favoring Kerry, since the majority of the spoilage was in heavy Dem precincts.

Does the data bear that out?

[OnTheOtherHand]

128. hmm...

Seems to me that if spoilage actually affected all equally, then it couldn't alter the percentage error in a precinct. The spoilage would reduce Kerry's absolute margin but not his percentage margin. (And if spoilage were really the same across the board, then Bush's absolute margins would be reduced, too.)

That said, I would expect the spoilage rates to vary a lot across machine types, locations, and -- yes -- demographics. I would expect that even if there isn't intentional fraud (and I'm not saying there isn't). But I don't have any hard facts to back that up right now (basically I'm trying to remember analyses I have read, I think including Febble's). I'm just trying to keep the ball rolling a bit!

[pgh_dem]

130. sorry, I guess I wasn't clear

My assumption was that spoilage could be assumed to be constant within precincts where spoilage was high, not across the board (I assume by 'board' you meant across the state).

For example, a hypothetical Kerry stronghold with round numbers, out of 1000 votes in a precinct, with 100 people polled, where a known number of votes (provisional or regular) were tossed out (say 50, where roughly 5% spoilage was not uncommon in Dayton, Montgomery County according to RH Phillips' deposition at http://web.northnet.org/minstrel/deposition.htm):

=B (poll)|K (poll)|B (projected)| K (proj)| B(actual) | K (actual)
25 | 75 | 250 | 750 | 238 | 712

50*.75 =37 K votes lost to spoilage (round down)
50*.25 =12 B votes lost to spoilage (round down)

Therefore K polls at 75%, but comes in at 71.2% (outside the 2.5% MOE, right?), and B polls at 25% and gets 23.8% (within MOE). Wouldn't this look like a 'red shift', too?

No real numbers obv, I'm just trying to illustrate my suspicion of what kind of effect spoilage would have on honest polling results from voters who *thought* they voted. Also, imagine the difference if a polled voter thought they voted for Kerry but through total miraculous coincidence the machine recorded the vote for Bush...

I guess what I'm asking is has anyone with access to the precinct-identified raw data actually crosschecked exit poll location with spoilage rate, to see if such an error could have been introduced?

If so, I imagine this error would be far more significant and quantifiable than guesses about how someone who refuses to respond to the poll may actually have voted, and a fraud-null hypothesis could be maintained (if not really logically defensible, a la 'those poor folks tend to vote Dem on crappy machines with high spoilage rates, oh well' reminds me of a Deep Thought that was like "At first I felt sorry for the homeless guy pushing around his grocery cart full of junk, then I thought 'Man, no wonder he's homeless, look at all the stupid stuff he bought'.")

[OnTheOtherHand]

132. gotcha

Almost a side point, but I don't think those results would be reported as 71.2% K, 23.8% B -- ordinarily the presidential returns are only for people whose votes were actually counted.

So (and here is why it isn't a side point), I don't think that this sort of spoilage would affect the WPE. But obviously it would affect the aggregate results if (and I'm thinking this isn't really "if") there were higher spoilage rates in high-Kerry precincts.

"Spoilage" that disproportionately affects the Kerry voters would affect the WPE (as would butterfly-ballot scenarios and touch-screen fraud, inter alia).

I'll ask Febble if she knows of any studies of spoilage in 2004. I vaguely recall some pretty interesting ones from 2000 -- none demonstrating outright fraud (although there may've been some of those, too), but certainly indicating how Florida should have gone. These studies, as far as I remember, didn't have to do with access to exit poll results -- they just looked at correlations between reported results and spoilage rates, sometimes controlling for other variables. And I think there may have been some similar work for 2004.

I'm thinking that part of the January E/M report sheds some light on how large the effects of disproportionate spoilage could be -- if it radically threw off the differential turnout rates, it would give the appearance that either the precincts they chose or the weightings were far off. And they concluded, I think, that that wasn't true, but rather WPE accounted for almost all the trouble. But I never studied that part of the report closely and haven't looked at it in months, so don't take my word for it.

[OnTheOtherHand]

138. a bit more on spoilage

I punted this question to Febble, then frankly sort of forgot about it. She suggested some names to look for: Oliver Dawshed; Liddle and Mitteldorf on the USCV website; Joe Knapp, who has written on DailyKos about Cuyahoga County OH, she thinks as JMKnapp. I think I've read at least three of these briefly, but not closely enough to suggest what you might look at first.

[Febble]

140. Pieces by Joe Knapp

on Daily Kos here:

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

Lots of precinct level stuff.

[autorank]

23. Try a 12-Step Apology

Mitofskyep programs, apologies are delivered to those the alcoholic/drug addict has offended or harmed as a major piece of the addicts recovery. One of the remarkable things about these programs is that the apology is delivered without excuses.

I'm struck by your two statements of apology:

"But I should have kept out completely. I apologise. My only defense is that I have found the attack dogs at DU quite upsettingly vicious. I knew that not only was my logic defensible, but that it was supported by data that would shortly be made public."

(That's not a defense, that's an excuse. The anger stemmed from now justified suspicion.)

"However, I apologise once again to members of this forum for remaining in the discussion at DU after I found myself in position where full disclosure, which was not then possible, was ethically required. I regret this. I plead both naivete and anger in defense."

(That's not a defense, that's an excuse. The anger stemmed from now justified suspicion.)

This is also revealing:

"I am not a “Mitofsky shill” It seems to me a tribute to Mitofsky that a) he listened to the blogosphere..."

You say you're not a shill and in the next sentence you are paying "tribute to Mitofsky," the man who denies the wider research and investigative community access to his data.

You came here and posted under false and misleading pretenses and conflict of interest by you own admission. That speaks volumes for your credibility as a "fraudster" (Kos' pejorative term for those who believe in election fraud).

I believe that the reaction to you on this forum was fully justified. Many could see the conflict and lack of "disclosure" through the content of your messages. Now, after this disclosure, the facts of the case, once based on justifiable inference, are clear.

[autorank]

27. Would you please explain your relationship to The Mystery Pollster

[Fly by night]

91. Remember: attraction, not promotion ....

[Peace Patriot]

24. I'm sorry but I just can't get past Edison-Mitofsky's collusion with the

TV news monopolies--who sold us the war and the slaughter of over 100,000 innocent people as a hunt for Saddam Hussein's WMDs--in doctoring these exit polls on everybody's TV screens on election night, and in withholding the raw data for six months now, meanwhile dripping out newsbites, bits of "analyzed" data and unsupported theories.

It's called shuckin jive on this side of the pond.

------

"It seems to me a tribute to Mitofsky that, a) he listened to the blogosphere, and b) he asked a known "fraudster" (in the sense, I hasten to add, of one who believed and still believes that the 2004 was both corruptible and corrupt) to conduct further analysis on the data." --Febble

Is this a "tribute" to Mitofsky, or is it something else?

He did this while denying that data to the Ph.D.'s at USCV who had asked for it, and to Congressional investigator John Conyers who had asked for it, and to the American people, most of whom don't yet know it exists and don't yet know who won the exit polls.

And what it looks like to ME is that he chose someone out of country and with no Ph.D., vetting the blogosphere for a "fraudster" he could pick off and compromise.

--to "conduct further analysis on the data" in a privileged position, behind closed doors.

--to divide and conquer; sow confusion and dissent; and play right into Karl Rove's hands.

That's what it looks like to me, Febble. That you have been had.

I am sitting here feeling very bad, indeed. For my country. For the slaughter of those innocents. For the slaughter of the truth. And for the pain and suffering of the living, who get hacked up by this dreadful thing that our government and its lapdog news monopolies has become: a monster that divides, confuses, brainwashes, smears, corrupts, twists, spins, stomps on, vomits on, blackmails, bribes, bullies, beats to a pulp, humiliates, befouls, ridicules, tortures and assassinates everything worth living for. Our highest ideals. Our love for the truth. Our desire to be generous and fair.

Whether I'm right or not--and I really don't know, except by implication of what you've told us, and from my experience of similar situations--you, and all of us, really need to look at this situation with clearer eyes. We are dealing with a brutal regime that has not the slightest regard for the truth, that exercises enormous control over the illusion called "the news" in the U.S., and that eats subcontractors like Edison-Mitofsky for lunch.

A regime suspected of killing journalists in Iraq, deliberately.
A regime suspected of anthraxing journalists and Democratic Congressmen.
A regime suspected of crashing a Democratic Senator's plane.
A regime suspected of planting a memo to discredit CBS, in retaliation for the broadcast of torture photos, and as a preemptive strike on a true story of Bush shirking his Air Force duty.
A regime that has "black ops" and secret prisons all over the world.
A regime that conspired with Tony Blair to cook up an unjustified war, inventing a 100% pack of lies to hoodwink the public with.
A regime that has squashed every investigation of its crimes.

In short, a regime that appears to be very good at shutting people up, crushing dissent and covering its tracks.

When a little flurry of experts disputing experts occurs--replete with jargon and charts and partially hidden data--on a matter of such profound importance as the Bush regime's legitimate election, we MUST ask, what is really going on? We would be idiots not to.

You should have asked this question when Mitofsky approached you. You should be asking it now--instead of presuming that it is a "tribute" to Mitofsky.

And I ask again: How do you get past E/M's withholding this data from other experts, with only you privy to it, and from a Congressional investigation, and from the American people who have the most right to it of anyone? How do you get past their damned lie to the American people on election night about who won the exit polls?

[LightningFlash]

29. Agreed to that. I want to know why she was the only one to get the data?

[helderheid]

30. BRAVO

[OnTheOtherHand]

31. if I point it out enough, maybe someone will catch on...

USCV's critiques have been based on information released by E/M.

If you want to believe that E/M is in the business of covering up a stolen election, then they must be pretty damn stupid.

And if you want to believe that E/M is in the business of covering up a stolen election, then you won't trust the raw data if they do release it.

So what on earth are we arguing about?

[Name removed]

32. Deleted message

Message removed by moderator. Click here to review the message board rules.

[Febble]

33. Of course he hasn't given me the data.

It belongs to his clients.

[LightningFlash]

34. Except you just admitted that....

You do have the data, or at the very least have seen full evidence of it during the press conference. For that reason alone, it would make sense to release exactly what you have seen and/or taken pictures or have evidence of.

The whole scientific community is waiting for that. In the news it appears you were the only one given access to this exclusive data along with Mystery Pollster, and that alone seems to warrant something does it not?

Share with the world, since that's the prudent thing to do.

[Febble]

36. As I have said

I have now have a working relationship with Mitofsky.

The data belongs to his clients. Only they can decide what is publicly released. I am not a member of the public in this instance. That is what this post is telling you.

[LightningFlash]

38. "Clients?"

"Only they can decide what gets released."

And why exactly do they have the final say Febble, given that you just admitted to viewing the data for yourself?

Why exactly can't *you* do the great public some good....By passing on that knowledge so that others can do the study?

"That is what this post is telling you. Only their clients can decide-"

You mean Voter News Service? Is that their clients? And why are they so bent up on not letting anyone else but them see the information for themselves?

I'm afraid this is sounding more and more increasingly like a planted story by Mitofsky who just happened to get a foreign person involved that will do his bidding. It fits in nicely, because in another COUNTRY, uniform response bias is well documented, proven, and expected..

[Febble]

41. His clients are the NEP

That is all I know.

[LightningFlash]

45. Then he most certainly lied again.

He owns Voter News Service, that seems to be his main protoje and his "Clients" according to his original speech in November, are the roper group.

Roper institute has all of that data but yet, none of it is available. Wonders never cease.

[Febble]

50. The Roper data is all available.

The link is on Mystery Pollster.

Anyone can download it.

For the 2004 election the E-M clients were NEP.

[LightningFlash]

52. Roper only includes the following items.

Precinct percentages, means, measurements, etc. the things already found on exitpollz.org basically.

I was referring to all of the data, including questions. Mitofsky motioned it was handed over to Roper institute, however even though they paid him it either never was or is locked up on the server.

[Febble]

57. I know that's all there is

I thought you might not even know that was available.

[OnTheOtherHand]

35. whoa, I'm getting dizzy here

I thought the party line was that exit polls have always been reliable.

But this is an excellent development. Let's say that E/M has been in the business of covering up Bush election theft for years.

Why on earth would they have released results that indicated that Kerry was ahead?

Why on earth would they have released tables that USCV could use to argue that Kerry won the election?

And, again, why on earth would we believe the raw data?

Folks, if you think E/M has been in the tank for years, then I don't see how you can also think that the exit polls prove fraud. I mean, you can if you want, but yeesh.

[LightningFlash]

37. For the same reason that...

"Why on earth would they have released results that indicated that Kerry was ahead?"

1) Because they released the results showing Buchanon ahead also, and deemed them as wrong.

2) Because they released the results showing state Georgia senator ahead also, and deemed them as wrong.

3) Because they have a subsequent and well followed pattern of documenting the same actual Discrepancy and because they have a history.

Right there tells the whole world, that these "results" at 12 midnight showing Kerry ahead never should have been released at all unless it was leaked. And it seems it was leaked, since the results are the true exit polls which would enable Voter News Service to apparently cook up one more accident to refute them.

[OnTheOtherHand]

39. midnight leaking wasn't the issue

I don't understand your "because"s at all -- I can only say again that once we assume that E/M has been making stuff up for years, the sky is the limit.

But I watched cnn.com releasing exit poll results state by state as the polls closed. How the public release of all those pro-Kerry results constituted or furthered a conspiracy to cover up election theft is way, way beyond the bounds of my weak imagination. Believe what you want.

[LightningFlash]

44. I'm sorry, but your argument disagrees with itself.

How the public who released all those Kerry results

) The public didn't release those Kerry results. The earlier 5PM results were released to erroneous applause and much scrutiny.

) The networks themselves are who broadcasted those Kerry results, on FINAL exit polls, after 78,000? people have been sampled.

Those are documented, proven, and analyzed. http://exitpollz.org

I don't see how in the world this was the "public's" fault whatsoever if this was such an error. Seems more to me to be the actual networks owned by Voter News Network such as CNN, ABC, NBC and so on.

Seems their poll was the result of magic or, magically the vote counts are all wrong. The latter stands up to occum's razor.

It seems much more apparent that Mitofsky has used Febble/Elizabeth to put innocent spin on this without her knowledge, and also force his same argument down everyone's throught so everybody stops even talking about exit polls.

In addition, it appears conclusively Mitofsky has been working with Mystery Pollster and his friend the entire time on the anti-fraud argument to make this thing drop.

[OnTheOtherHand]

54. you might want to slow down a bit

"public release" = "release to the public" not "release by the public" -- you've actually misquoted me.

smartvoter of course is right that exit poll results were leaked before the polls even closed, but that wasn't my point.

I think it is pretty clear that Mitofsky has <i>not</i> been working the entire time to make this thing drop -- if he had, USCV wouldn't have been able to draw on the January report for supposed evidence of fraud.

[LightningFlash]

55. So let me get this clear.

"if he had, USCV wouldn't have been able to draw on the January report for supposed evidence of fraud."

You basically take the position then that something clearly wrong happened here, and yet you disagree and say that the exit poll results were leaked?

I don't see how you can stand in two positions.

Allow me to clarify. Clearly Mitofsky has been working to have it drop, since the year 2000 was when Voter News Service also mistakenly showed Al Gore was the winner. Reluctant Bush Responder across the entire state, country, everywhere.

Of course, after it was proven 16,000 votes were deleted and there were more Al Gore votes anyway the stolen election was reality. Mitofsky never said another word about his polls and we went to business as usual.

What's being maintained here is the CORPORATE MEDIA NETWORKS "leaked" the exit polls, not the public, there was no "public" release.

The release can be seen here

http://exitpollz.org

And later at 12 AM, was the ACTUAL exit polls displayed by the networks. Mitofsky has never confirmed/denied if they were leaked. He has only substantiated that these results were wrong--and after the re-weighting occurs, those results are RIGHT.

That is the hypothesis.

[OnTheOtherHand]

64. ??

You basically take the position then that something clearly wrong happened here, and yet you disagree and say that the exit poll results were leaked?

I don't see how you can stand in two positions.

I'm not sure what the two positions are, I'm not sure whether I hold either of them, and I'm not sure they are contradictory. Other than that, I'm right with you. (grin)

Once again, I never said that the public leaked the polls. I'm not sure why I would have to explain that even once, much less twice.

Some exit poll results were leaked well before the polls closed; as far as I know, cnn.com released exit poll results for each state not long after each state's polls closed, showing Kerry ahead in Ohio and other states; I assume that other networks did something with these results, too; at some point, the reported exit poll results were reweighted to match the official results. I'm not sure what you mean about "ACTUAL exit polls." It's possible right now (or at least last time I checked) to download the raw exit poll data and verify that more exit poll ballots were marked for Kerry than Bush. I don't know whether Roper ever got around to posting them -- I think perhaps you need to be a Roper Center subscriber -- but I think I downloaded them from ICPSR right at home.

My point, once more, was that if E/M was trying to cover up, they evidently did a lousy job.

[LightningFlash]

69. And again you dodge the question.

So you admit the media networks are who displayed the exit poll results.

But then you go on to say Mitofsky covered it up by doing a lousy job of it. I don't think that jives with the idea of uniform reluctant bias theory.

By the way, Mitofsky has purposefully done a lousy job in covering up election irregularities in the past, because apparently the technique makes more sense by those who hire him.

In other words, history records it makes more sense for the exit polls to be shoved out that are basically the real exit polls, than adjusted later for unfriendly vote counts.

If he did it the other way and actually released the final data point blank, it would be much harder to prove that the polls were a biased fraud. Because that would be adjusted exit poll results which in every country, including the UK in fact, are accepted as fact.

So by making sure the "earlier" polls are the ones adjusted to meet the real votes later, it is much harder to see any basis for vote corruption. On contrary--your hypothesis which flies on his dodgy results, presents a much more prime case for having fraud, not dismissing it.

[OnTheOtherHand]

74. ???

So you admit the media networks are who displayed the exit poll results.

I don't have to "admit" that -- everyone knows who displayed the exit poll results.

If you place a question mark next to your question, perhaps I will understand how I am dodging it.

But then you go on to say Mitofsky covered it up by doing a lousy job of it.

I said what? He covered what up by doing a lousy job of what? Things might go better if you quoted actual words I actually wrote. Including, in this case, the word "if."

In other words, history records it makes more sense for the exit polls to be shoved out that are basically the real exit polls, than adjusted later for unfriendly vote counts.

Are you saying that Mitofsky schemed to conceal that Kerry won the election, by releasing data that showed that Kerry won the election -- only later they were changed to show that Bush did? And history shows that this works better than, say, releasing data that indicated that Bush won the election all along? (And history shows this how, exactly?) And it's obvious to all of us that if the corporate media networks had released exit polls that showed Bush ahead from the beginning, then we would be more suspicious of the results?

Hmm.

[LightningFlash]

80. The difference is I'm not saying it.

Because the bottom line is the facts are saying it.

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=369999&mesg_id=369999

This pattern has been consistent in repeated elections, where complicity is very documented. However the contention seems to be "Mitofsky" scheming, when that is not what is being said.

Mitofsky who owns Voter News Service, has no reason to do that. But the media networks under VNS certainly have. And they are who released the results, and released them repeatedly. It was not leaked by bloggers or the public as is argued.

It was directly showed on national televion and the web, by the media itself. Who is paid under VNS, and who along with NEP do not just hire the W-M firm, but control exactly who they work for.

Their firm is hardly independant. I must admit that response bias did occur, and absolutely had an affect. It just did in no shape or form have a uniform affect which would distort the exit polls in sway to reverse the exit poll percentages.

Otherwise the swing would be grossly all over the map. Indeed, for there to be a real precinct calculation done someone will need all of the questions and precinct sampling which W-M seems to easily give out to the Mystery Pollster.

One might question exactly why, none of that precinct level data is presently available to run an entire simulation on using the final means.

[OnTheOtherHand]

100. hmm...

Probably at this point it seems like I am deliberately misunderstanding you. I am starting to understand parts, but other stuff still confuses me.

No one says that the public released exit poll results. (I said there was a public release. We went over that, right?) As to whether bloggers leaked exit poll results, well, I guess we basically agree about that one, too (at least one of the basic facts): no matter what anyone thinks about where the afternoon results came from (and weren't they in fact posted on Drudge?), CNN and others certainly released results favoring Kerry after the polls closed, and before the wee hours of the morning. I saw a bunch of those results with my own eyes, but it's nice to see them "immortalized" on the web site you've mentioned.

I think Voter News Service no longer exists, but I guess that isn't the point. I'm not sure what you see as the specific motives of the media in all this.

One problem with releasing all the precinct-level data is, as sure as I stand, it would provide enough information to identify some individual voters -- and that's a bad thing to do after the fact. Especially because I think, if we're honest, we'll admit that nothing in the exit polls can prove all that much one way or another. But that doesn't mean we should settle for two scatterplots for the rest of eternity, either.

[smartvoter]

46. No. If memory serves, Mitofsky slammed the networks for early releases.nt

[smartvoter]

42. I see it more as preserving a big fat contract than long-term collusion.nt

[Peace Patriot]

85. It's kind of like Bush and Cheney testifying together about 9/11, in ...

...secret and not under oath. Highly suspicious circumstance. But wouldn't you want to hear the tapes yourself, or, there being no tapes (that we know of), read the transcripts yourself, or, there being no transcripts (that we know of), read the notes of those present (that they were not permitted to take from the room), rather than relying, say, on Bush or Cheney's account of the meeting?

Say there WERE transcripts. Wouldn't you rather read them for yourself than rely on someone else's summary? EVEN if you thought there might have been some funny business about their "chain of custory," or EVEN if six months had past.

You want to have facts from as close a source to their origin as possible. But even with tainted evidence, you might be able to SPOT inconsistencies, from which tampering or fraud could be inferred or proven.

This is with regard to this particular circumstance--the E/M data--which they withheld for six months, are still withholding, and have selectively leaked bits and pieces of, to friendly parties.

That is a highly suspicious circumstance. But I'm sure any statistician--or American voter--would like to see it anyway.

What would be their reason for doing this? For putting this data under a cloud in this way?

As with Febble being contacted by Mitofsky on the side, and pulled out of the USCV group, we cannot--with such an important matter at stake as a fraudulent election, and such dirty players as the Bush regime at work--presume innocence with such behavior.

Is it a Karl Rove "confusion to my enemies" game? Or what? Withholding the data for six months, fooling the American people on election day about who won the exit polls, and now selectively allowing little peeks at the data, to make this or that self-serving point, positively inspires suspicion. Why do this? And it makes us want the data even more, after all these shenanigans.

We DON'T KNOW what occurred--in this whole narrative of the exit polls and the election. We don't know the whole story, and are trying to figure it out. Why WOULD they put up polls all day showing a Kerry win, and then suddenly change them in impossible ways? Maybe we need phone logs, as well as the raw data. The whole thing smells to high heaven--and cries out for full disclosure and all data.

[smartvoter]

40. Febble, if I may, let me suggest that there is a very simple

solution to all of this. The data needs to be released.

The fact that it remains locked up is troubling, especially since we are not talking about analysis of what makes one person buy Brand X over Brand Y, or which SW platforms perform best at peak loads.

It's data on a public election.

We (read: collectively) can argue back and forth in thread after thread until cows drop from the sky and none of this will be cleared up as long as the data remains locked up. It does not invite suspicion, it creates it.

[garybeck]

43. a problem with all this "slope" talk...

Can I just speak for many of the common folk and say that all this techno-talk about slopes and non-slopes is going way over the heads of most people. It's nice to see that it's being discussed, but we need to reach out to the common people. Most people will read a report by looking at the abstract, conclusions, who signed it and their credentials), and maybe scan through some of the graphs and charts to see if they make any sense.

When you reach your conclusions, it would be really helpful if it could be summarized into a couple sentences or a paragraph. For example, "my research concludes that the E-M report is flawed and there are unexplained anomalities in the exit poll results" or the opposite, if that's what is concluded.

We don't need the whole history about how the research bounced around. We just want a summary of the conclusions and what your credentials are. It's fine to include all the data and explain how you arrived at your conclusions but I think most people will only glance at that and probably not understand it.

Thanks

[Febble]

49. You are absolutely right

and the whole thing is pretty irrelevant to the fraud debate anyway.

But FWIW, here it is, as I see it:

USCV thought they saw a pattern, from some figures given in the Edison-Mitofsky report, of the "bias" (greater discrepancy between the polls and the count)being greater in Bush strongholds than in Kerry strongholds. So there was a "slope".

Bias could occur either because fewer Bush voters where responding, or because fewer Kerry votes were being counted (fraud). As it seemed unlikely that Bush voters would be more "reluctant" in their own strongholds, a more plausible alternative might be fraud. This was supported by the observation that response rates were actually slightly higher in the Bush strongholds.

However, the way "bias" was measured had a problem - it tended to exaggerate bias in the middle, and at the Bush end of the graph. The question then was, if my formula was applied, would the apparently greater bias at the Bush end disappear?

I said it might. I think USCV said it couldn't. They certainly said I was wrong.

The evidence seems support the idea that the "slope" - bias being higher at the Bush end than the Kerry end - was indeed an artefact of the way bias was measured. Using "my" index, the line is flat. No slope.

Doesn't mean there was no fraud. Doesn't mean there wasn't widespread fraud. Simply means it wasn't greater in Bush strongholds.

And those response rates? Well, it's still a puzzle, but they were so variable, it's difficult to conclude anything from them.

Hope this helps.

[LightningFlash]

51. Except strangely in repub strongholds votes were switched from K to Bush..

[ibid]

56. "the way "bias" was measured had a problem" - huh?? - I did not

see from first principles the problem you refer to.

Having made a few formula changes myself, I wonder how "results based" were these changes.

And you say the folks at USCV now agree with you - except Ron Baiman does not - and you say his reasoning is illogical. So the slope in the WPE regression line is to be ignored. Because with a transform - applied at a specific data level(namely precinct) - there is no linear relationship between "bias" and vote-count margin. huh? - OK..

Perhaps the "lot of variance" ties to other reasons - but we will never know now until the data is released.

Widespread fraud in both directions? - now that there is a conclusion I would not have gotten to as the answer as to why an exit poll failed to reproduce the "results"

[Febble]

60. Have you seen the DU thread

also my paper?

No I am not saying the "USCV folks now agree" with me. I am saying I believe some may. They are individuals. Four authors are on the latest USCV report. Presumably they do not.

The WPE problem is described on the Dkos front page piece, and also the Mystery Pollster piece. The problem is also described in the USCV first paper. Although their formula is different (because the were solving a slightly different algebraic problem) the essence is the same. This was done by Ron Baiman and is perfectly logical.

The problem, essentially, is that where the vote margin is very large, any difference between the predicted margin and the actual margin is going to be very small, even if there is massive bias in either count or poll. And if there is an overall bias, the WPE is going to underestimate the bias more at one end than the other. So to equal it out, instead of simply subtracting the predicted margin from the vote count margin, you derive a measure we (me, then USCV) have called "alpha" - which is the ratio of Democratic reponse rates to Repubican response rates. The log of this ratio gives a symmetrical measure of bias that is independent of vote count margin. So it can be plotted against vote count margin without getting a spurious relationship. Mitofsky's statistician called it WPE_index, I called it bias index, USCV tends to call it ln(alpha).

This is not actually in question. I am still not clear what Ron Baiman is saying. What I am saying is that this plot looks as though it has a slope of zero - that bias is not greater at the Bush end than the Kerry end. If Ron Baiman now agrees with this then there is no problem.

And it certainly does not mean there was no fraud.

And if you look at the plots on the DU thread linked in my post you will see that the "errors" are both in Bush's and Kerry's favour, and that many in both directions are very large - much larger than the average error, which is the value at the centre of the regression line. But more of the "errors" are above the line, which is why the polls predicted more votes for Kerry. But it doesn't say what went wrong - whether it was response bias or vote corruption.

However, it would seem a bit far fetched to say the errors on one side of zero were all polling errors, and the ones on the other side were all fraud.

There is also the possibility that much of the variability was due to random factors - it certainly looks like noise. But that would not account the fact that the average bias was not zero.

That is what needs to be explained.

[ibid]

72. Thanks for the explanation :-)

:-)

[Chi]

58. Can you provide a formula to reduce the noise around that apology.

:popcorn:

I was having trouble differentiating it from all the, 'I won' talk.

A laugh a day makes...makes life seem less sucky.

[ibid]

59. LOL - :-)

:popcorn:

[helderheid]

61. You made me laugh out loud

needed that! :thumbsup: :rofl:

[autorank]

66. More disclosure re: Mystery Pollster -- ?

From Mystery Pollster:

"Also consider that the previous problems with the UK exit polls did not produce election fraud conspiracy theories. Why not? One reason, as our friend Elizabeth Liddle points out, is that the count in the UK is "utterly transparent." Paper ballots are sorted and counted in public at a centrally located place, open to all who wish to observe. People have faith in the result because of this transparency. Another reason, as Liddle puts it, "why auditable elections are so important."

MYSTERY POLLSTER
http://www.mysterypollster.com/main/2005/05/uk_polls_an...

[Febble]

67. And your problem is?

You don't agree that auditable elections are important?

[autorank]

68. Well since you're disclosing relatinsihps, I am just asking.

What is it?

[Febble]

70. It's not a question of disclosing

because there's nothing to "disclose". He's a blogger. His name is Mark Blumenthal. I have posted comments on his blog.

Mark linked to one of my diaries at DKos, because it was relevant to a post he was making. We corresponded by email, and still do.

We corresponded during the preparation of my paper, and his contribution is acknowledged on that paper, as are the contributions of other bloggers including DemfromCT at Daily Kos, and Rick Brady.

I've never met him in my life. I don't even know where he lives. But I would now count him as a friend, and clearly he counts me as one. If you follow the Mystery Pollster link in my post you will find the post he did about my work.

[LightningFlash]

71. Seems a bit more complicated then that...

Apparently Rick Brady and the other Pollster have already run two more samples they claim to derive from data of Mitofsky. Clearly you do recognize that they have more available data from Mitofsky.

It doesn't seem feasible that they would have access to everything without help from you, given that you came up with the mean-response rate. Also no one else has commented on this, including Ron Baiman who authored the latest paper. It's also demonstrating the mean-line is wrong, and needs to actually be re-constructed in a real professional model.

[OnTheOtherHand]

77. "two more samples"

If you mean the scatterplots that Mystery Pollster has posted, he explains that those are two scatterplots that Mitofsky presented at the AAPOR meeting on Saturday. It doesn't mean that MP has access to data.

[LightningFlash]

81. Question.

"It doesn't mean that MP has access to data."

How exactly did he get these calculations? Slides sent via email? Personal correspondence? Other?

[Febble]

82. Rick and Mark were both at AAPOR

I know that, because, as I have said, they both collaborated with me on my paper, and we are in contact.

They both attended Mitofsky's talk.

It was a public talk. There were handouts of the slides. I expect because Rick and Mark are bloggers they wanted slides, rather than handouts so they could post them on their blogs.

I expect if you'd been there, you could have had them too.

[LightningFlash]

83. AAPOR

"I expect if you'd been there, you could have had them too."

Question..Did anyone else besides Rick or Mark receive these slides, and has it since been reported?

[Febble]

84. I have absolutely no idea. n/t

[OnTheOtherHand]

92. replying here for neatness... but I don't know either...

According to Mystery Pollster's blog post, the two .jpg scatterplots are the same ones Mitofsky presented -- so maybe Mitofsky e-mailed him his PowerPoint, or copied the pictures onto a disk, or whatever. I don't think the mechanics of copying the images matter very much. The point is, the images aren't the data. If you copy and paste from the blog, you can have the images too -- and you won't have the data either. (But it might be worth it to get a better look at the images.)

[Name removed]

73. Deleted message

Message removed by moderator. Click here to review the message board rules.

[autorank]

75. Hey there, Welcome to DU!!!

Here's what PeacePatriot said:

I'm sorry but I just can't get past Edison-Mitofsky's collusion with the

TV news monopolies--who sold us the war and the slaughter of over 100,000 innocent people as a hunt for Saddam Hussein's WMDs--in doctoring these exit polls on everybody's TV screens on election night, and in withholding the raw data for six months now, meanwhile dripping out newsbites, bits of "analyzed" data and unsupported theories.

It's called shuckin jive on this side of the pond.
---------------------------

He did not say that this collusion was proof of fraud, he said it was an indicator that Mitofsky or others who collaborate with * and co. are to be viewed with distrust.

You said "It drives me nuts..." That's no drive, its a short putt (sorry I couldn't resist). You're not nuts but you surely know that qualitative research can lead to quantitative studies. In any case, people here are free to comment as they choose...and they will.

:hi:

[Name removed]

78. Deleted message

Message removed by moderator. Click here to review the message board rules.

[autorank]

79. Well, thanks. I'll be around.

[meganmonkey]

87. Wow - this thread is certainly helping me catch up on

the dynamics of this forum! Geez!!!!

:crazy:

[OnTheOtherHand]

90. welcome!

so if you have any actual questions, some of us will be glad to try to help....

[autorank]

93. And Welcome to DU!!! to you OnTheOtherHand

Have you been lurking here for a while?

[OnTheOtherHand]

97. hiya

Have you been lurking here for a while?

Maybe a few weeks. I know Febble from USCV. Full disclosure: I am a Friend of Febble. I think her paper is outstanding. I think her efforts to convince Kathy were outstanding. I think her apology is outstanding -- and whatever else anyone says about her, I say she's got chops to offer it and stand by it, without backing away from her arguments. I argue with her about stuff, too, but I usually count to 20 first. Even then I generally lose, but as far as I can tell she always calls it straight, which is one of the reasons I like her.

Other than that -- I was raised as a Democrat in Franklin County, Ohio, so I take the 2004 election a bit extra personally in that sense. And I am not now nor have I ever been employed by Warren Mitofsky, although I honestly don't think Febble should have to apologize for that either -- the bottom line is, Mitofsky needed her.

[meganmonkey]

98. An Ohioan - did you attend the Election Reform Teach-In in Columbus

a couple weeks ago? It was great!

[OnTheOtherHand]

99. naw, I'm not there now

but I still have family in Columbus. Was there a write-up of the teach-in? maybe in the Free Press?

(You caught me in a gentle snark upthread, I was just a bit cranky at that moment.)

[meganmonkey]

101. I haven't seen a write-up

unfortunately. Some great speakers were there, lots of talk about the machines, updates on the investigations...very encouraging. I just wish there'd been more people. I went to college near Columbus many moons ago, and I took a little road trip down for the event. It's becoming a habit - I've been down there a couple other times since the election for various events.

[autorank]

106. Hmmm...

You say, "the bottom line is, Mitofsky needed her." You should know?

Your appearance is certainly timely, arguing against the need to apologize just as the apology is delivered. Synchronicity perhaps.

Here is some general information on behavior people find troubling.

It's troubling behavior when:

-- You volunteer for one candidate while on the payroll of the opponent;

-- You represent yourself as one thing, let's say unbiased, when in fact you have a huge bias; or

--You associate with one group while serving the interests of an opposing group.

The list goes on and on. An apology was clearly called for.

Will you be around long or is this a guest appearance?

[OnTheOtherHand]

112. whatever

What, you think I should have argued against the need to apologize before the apology was delivered? Now, that would have been strange.

Seriously, I wandered into DU, saw she had put her name on a thread, read her post, thought it was pretty good and she might use some company. She never told me she had posted it. It's not for me to argue against the need to apologize -- she made that call for herself.

Mitofsky doesn't share his ideas with me. But it kinda makes sense that if you're trying to analyze your data, and someone has just come up with an innovation in data analysis (one, by the way, that USCV has accepted and used in their latest work), you would want to get that person's help.

On the other hand, if Mitofsky is trying to cover something up, then he could just cook the data himself, or buy off some American PhD to cook it for him.

What opponent? Febble is not on Bush's payroll.

What bias? Do you think that Febble got special Republican software to run her analyses?

What opposing group? What groups are you talking about?

You can be as troubled as you want, and you got your apology. Now what?

[kiwi_expat]

113. I wonder if you might be able to answer my questions.

Hi OnTheOtherHand. Welcome to DU !

You suggested on another post that you might be able to answer some questions. Could you possibly answer any of the questions in my post #103, below? Thanks.

Cheers
kiwi

p.s. Would you like to examine (recount) some ballots in Franklin County, under Ohio's Sunshine laws? Liam_laddie is pioneering the way in Hamilton County. It would be fantastic if you could help.

[OnTheOtherHand]

133. thanks for the welcome, and by the way...

I actually answered (or tried to answer) your questions in #103 without ever having seen this post. I could use some newbie help here, because I don't know whether there is an easy way to find all pending replies to my posts. Meanwhile, if I ever leave you hanging, just politely hunt me down.

As I mentioned elsewhere, I'm not in Franklin County any more, so I can't help on that front, sorry.

[kiwi_expat]

135. You could try the "My Posts" function at the top of the first page....

it lists all of your posts for the last 48 hours - and shows how many replies you have had to each one.


Cheers!

[meganmonkey]

96. Thanks

I know who to ask for info, no worries.

And welcome to you, too :hi:

[kster]

102. This thread reminds me of

the old western scene,where everyone has guns drawn on each other.

If we want to get the answers, the question should be why Mitofsky and/or the media will not release the data.

Forget about all the rest it clouds the issue, WHY WON'T THEY RELEASE THE DATA?

No matter which side you are on, you guys are the smartest,most talented group of people out there.

Because of that I know we should all be on the same page and not except answers like (because they own it) to question's as serious as this one is.

WHY WON'T THEY RELEASE THE DATA?

[tommcintyre]

104. You need to ask one other question

If the data ever IS released, will it be uncooked?

[autorank]

105. So you smell what Mitofsky is cooking!

This is a great question. How would we know it was "uncooked?"

This is why their needs to be a total investigation of election fraud with clear incentives to be truthful.

Great question.

[tommcintyre]

107. The "stink" has been in my nose for too long now.

From the nearly forty years (he "invented" exit polling in 1967) of "bad brewing", you can pretty much count on the fact that, if the numbers were ever released, they would most certainly be cooked, if there was ANY chance they could indicate a Kerry win.

They would likely be manipulated to indicate a minor pattern of fraud on both sides, possibly more by the repubs, but CERTAINLY not enough to change to outcome of the election, right? <From some of the posts I've seen her lately, I get a feeling that's where they may be leading us - if they feel they really must "throw us a bone" to get us to stop gnashing our teeth. What they don't realize, is we are WAY mo' hungry that that - so it will never satisfy us! ;) >

In other words, the chances of them releasing data indicating that Kerry won is about as great as he odds that Bush actually won the election.

[autorank]

108. ...or the odds that pigs live in trees.

Great message:yourock:

[kster]

110. Well you explained why

this "Febble: a report and an apology" just didn't feel right to me all day today. Thanks

[autorank]

111. Good times, good times kster!

[kiwi_expat]

131. How about bringing out your peace-pipe again, autorank?

Please.

[kster]

109. Mitofsky is setting the stage,Gotcha

Autorank,Tomcyntyre you guys are on a roll keep up the great work.

[kiwi_expat]

103. I have three questions related to upcoming Ohio manual recounts.

Hi Febble

This seems as good a thread as any to ask if you could please answer these three questions:

(1) Do WPE and WPE-index involve just the PRECINCTS' figures or do they involve the figures for the entire POLLING PLACE where the NEP precincts were located?

So far, we have been able to match precinct names with 15 Ohio NEP precincts. One of them, Cincinnati 4-M (NEP# 52), has a 28% absolute variance when comparing the NEP raw data with the precinct data. But it has a 22% variance when comparing the NEP-raw data with the POLLING PLACE data.
http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=329749#363703

The NEP interviewer at Cincinnati 4-M attempted to select his respondents randomly from ALL of the voters at the polling place - which I infer from the E/M Report is normal procedure.


(2) If the "fraud" (as opposed to the "rBr") interpretation of the flat regression line (on the second scatter plot) is correct, how many NEP precincts would have to be recounted in order to have, say, a 95% chance of locating some fraud?


(3) Has anyone suggested that there might actually be a "fraud" regression line that slopes downward from the Kerry strongholds and an "rBr" regression line that slopes upward towards the Bush strongholds - resulting in an overall flat line? Does a regression line have to be straight? If not, perhaps both lines flatten out shortly before approaching the 100% partisan limits.


Thanks very much for considering these questions.

Cheers
kiwi

[OnTheOtherHand]

114. well, I'm from Ohio...

(1) Ideally, WPE ought to be applied to precincts, not polling places, since it means "Within Precinct Error." But if the exit pollers lumped precincts together, well.... That's all I know about that. Febble or others may know more.

(2) It seems to me that there is more than one fraud interpretation, so you would have to decide what kind of fraud you were looking for -- and how to do the recount! If you are comparing to the exit polls, then it's a problem, because who knows what counts as real proof of fraud? If you can do a physical recount (op-scan precinct?) against reported results, then it wouldn't take many precincts to start to get a sense of a vote-stealing pattern -- how many depends on the pattern, I guess, but "add 6%" or "add 20 Bush votes" would stick out pretty fast. It's hard to say what would provide a 95% test of a hypothesis without knowing exactly what the hypothesis is.

(3) I don't know if anyone has put it quite that way, but yes, it is absolutely possible that fraud and rBr could tend to cancel each other out. The "regression lines" we are talking about are straight by construction, but that doesn't mean that anyone assumes the underlying reality is straight. It's just that USCV had argued that the WPEs were very high in the high-Bush precincts (and very low in the high-Kerry precincts) -- so, if that were the right way to think about the data, definitely one would expect the best-fit straight line through the data to slope up.

Hope this helps somewhat; keep asking!

[eomer]

115. Unless...

"definitely one would expect the best-fit straight line through the data to slope up."

Unless, of course, there is some shape in the middle that cancels out the alleged shape at the ends. USCV does see such a shape in the data when they divide it into 20% segments.

[OnTheOtherHand]

118. two thoughts about that

Well, three -- first of all, of course you are right that a shape in the middle could cancel out the shape at the ends. (If I ever skip the part where I say you're right, just give me another try. It can get really confusing when people don't point out what they agree about.)

Second, if one draws a straight line through the five mean WPEs as reported by E/M and analyzed by USCV, one does get an upward slope. I'm not saying what this proves, but when you say that USCV "does see such a shape," I'm not sure that's right. Am I missing something?

Third, the scatterplots indicate that in the 80%+ high-Bush precincts, the outliers pretty much all favor Bush, but somewhere in the high 70s, there are some outliers that pretty strongly favor Kerry. I'm not sure what this proves, either, except that if E/M had broken up the precincts differently back in January, we probably would have been arguing about them pretty differently since then.

[eomer]

119. Regarding "Second"

What I'm saying USCV sees is a non-linear shape. The first graph on page 7 of their paper shows the WPE when broken into 20% segments. It looks non-linear.

[OnTheOtherHand]

121. thanks

I was gonna say that it might be clearer if we went by page numbers -- and we still could get confused, but are you looking at the May 17th revision (the one currently posted, at least as of when I started typing) and the graph called "Simulated Response Bias vs. Actual E-M Data"? I think so.

Both the E-M data shown there in orange, and the response bias model in blue, are indeed non-linear. Even so, if you run a straight regression line through either one of those, it will slope up.

That's one reason why a lot of people expected a straight regression line to slope up even after the "bias index" was applied.

[eomer]

122. Yes, I agree it will slope up.

And it will understate the complexity of the underlying data, at least if you accept the shape shown in the graph as being a fair representation of the shape of the data.

[OnTheOtherHand]

124. it definitely understates the complexity of the underlying data, although

if you're asking whether I think the orange line in the first graph on p. 7 of the May 17 USCV report is a fair representation, well, I would say that it is fair as far as it goes, but the second scatterplot reproduced in ribofunk's thread is a better representation. But the entire scatterplot, not just the regression line. The orange line does give more information than just the regression line, although the regression line also gives important information that the orange line doesn't. The orange line may seem to contradict the regression line, but I don't think it does.

The most potentially deceptive thing about the orange line is that it gives the visual impression that all five categories have equal weight. But actually there were only 40 precincts in the high-Bush category (at left in that graph) reported by E-M.

[eomer]

125. The most potentially deceptive thing about the orange line

is also the most useful thing about it.

You can't see what is happening out at the ends of the range unless you decouple them from the middle.

I agree one should be careful not to interpret the 5 bands as having equal weights. But given that caution, the 5 bands do seem to shed some light that can't be had with just a regression line.

Regarding the scatterplot, sure that is the most accurate representation because it is just a plot of all the data. But if you are looking for a pattern of fraud, spoilage, response bias, etc. then you can't just stare endlessly at the pretty plot. You have to start decomposing it somehow.

[OnTheOtherHand]

126. yes, I agree with all of that

Hooray! I love agreeing with people, despite my handle.

The only thing I would add, for emphasis, is that if you are staring at the scatterplot looking for a pattern of fraud, your eyes will probably go different places than if you are staring at the January E-M tables looking for a pattern of fraud. I'm not saying that there isn't any fraud in the high-Bush precincts, just that you get, well, a really different picture.

[eomer]

127. And I agree with all of that

I second the sentiment. Hooray!

On the other hand...

...no, just kidding, I don't have an other hand right now.

[kiwi_expat]

116. Thanks! That is very helpful.

With regard to my question(#2) as to how many NEP precincts would have to be recounted to locate "fraud":

I was referring to Febble's statement that if the FLAT regression line is of fraud, then fraud is everywhere.

I just wondered how "everywhere" that would be. Would there be any way to quantify the probability of fraud being at a particular NEP precinct, if we assume that the entire bias is due to fraud?

(Note that I'm not trying to figure out how to identify the fraud, or what its magnitude might be.)

Thanks again!

[OnTheOtherHand]

117. "how 'everywhere' that would be"

I'm sorry, I'm having trouble wrapping my head around your entire question here. It really does make a difference whether we have a relatively reliable way to reconstruct the actual vote, or not.

Hypothetically, it might be that there was about a 6% vote shift in every precinct (easy to detect if we have a reliable count). Or there might be, say, no vote shift in 2/3 of precincts and 18% vote shift in 1/3 of precincts -- we might be able to detect that even without a reliable count. There could be a lot more hypotheses!

Have you looked at the scatterplots, for instance posted in ribofunk's thread? There is more information there than just the lines, and it gives you some idea of what Febble means by "everywhere."

[kiwi_expat]

120. So-called "fraud" would include uncounted votes and accidental errors.

With the hypothesis that the bias is due to "fraud" (as opposed to "rBr"), so-called "fraud" would include everything which could cause the vote to differ from voter-intent: uncounted under/over votes, rejected provisional ballots, accidental errors, and all manner of actual fraud.

I suppose that most precincts would have some-such "fraud". And we should be able to detect most of it in the manual recount of punched card precincts.

Thanks for helping me think this through. :-)


Cheers!

[OnTheOtherHand]

123. yeah, more or less

I'll put it just a bit differently: discrepancies between the exit polls and the reported results could be attributed to any combination of response rate bias (including rBr), exit poll misreporting (interviewer error and/or respondents flat-out lying), or all the things you mention that could cause the vote to differ from voter intent. So, you're right, it's not just "rBr vs. fraud."

Also, many effective forms of vote suppression won't show up in exit polls because potential voters will never make it to that stage.

[TruthIsAll]

137. Do you know who those voters are? We do. First letter hint: K

As in ALL elections, the democrats get the shit end of the stick when it comes to disenfranchisement.Now there's a a straight-line function for you.

Does it prove bias? Or will you just tell us there is no such evidence? If so, I would be happy to supply you with a few thousand links.

Simple Linear algebra:
Final Bush vote = Actual Bush votes + BBv Switched Dem Kerry Votes + switched Repub Kerry votes + Phantom Votes - Spoiled Bush votes (a few) + Kerry tabulated votes

Final Kerry Vote = Actual Kerry Votes - BBV Switched Dem Kerry votes s - switched Repub Kerry votes - spoiled Kerry votes (a lot) - Kerry tabulated votes.

And as you have correctly pointed out, Disenfranchised Voters are not even counted. Any idea how many?

[OnTheOtherHand]

139. "no such evidence"? heck, no...

I recently read a long paper about felon disenfranchisement, so that part of the picture I have pretty well down. But you probably also mean all the other kinds of disenfranchisement (of course including all the sneaky last-minute kinds).

I'm not sure what the exit polls prove, but I am miles from arguing that there isn't all sorts of bias in the electoral system.

[autorank]

141. We agree on this "all sorts of bias in the electoral system."

What's your most compelling case for the bias and how did it influence the 2004 elections?

[OnTheOtherHand]

142. honestly at this point

I'm not sure what I think, about the 2004 election. I think Bush probably won the popular vote (which leaves room for voter suppression, as we know happened). But I'm not arguing it, because there's a lot I don't know. I can explain the reasons why I'm not convinced that the exit polls prove massive fraud, but that doesn't mean that I am ruling out massive fraud. And since massive fraud is possible, we have a big problem whether or not it actually happened in 2004, so in a way I don't even care that much whether it did.

I bet someone is gonna misread that and jump down my throat. But you asked a fair question, and that's my best brief shot at answering it right now.

[kiwi_expat]

150. When talking about the "fraud" hypothesis for the level regression line...

...in the second scatter plot, we are of course talking about a pro-Bush (net) "fraud". (This is the alternative to the "rBr" hypothesis.)

With this hypothesis, we need to account for all of the "bias" by using the difference between the intended vote and the actual vote. And it needs to be fairly uniform across all partisanship precincts (to get a flat line).

A working assumption that is used for UNcounted votes is that they have the same Bush-Kerry distribution as the counted votes for an individual precinct. We also know that uncounted votes tend to occur more in heavily Kerry precincts (because of "institutional bias", if for no other reason). Thus, overall, the uncounted votes would bias the counted vote toward Bush.

The same sort of pattern would be expected to occur for genuine unintentional errors.

So if we were plotting the bias of just uncounted votes and accidental errors, the line would be expected to have a downward slope from the Kerry strongholds, with the slope being steeper on the Kerry partisanship half of the graph.

So now, I figure that if we are hypothesizing the entire FLAT regression line to be "fraud" (i.e., NET intended vote minus actual vote), we have to propose something that will slope upward toward the Bush strongholds, to balance out the uncountedvotes-accidentalerror downward line. And I figure THAT is the intentional errors line. (We'll make that line entirely above the "x" axis. :-) )

Thus the best place to look for *actual* fraud is in the Bush partisan precincts.



What I can't figure out how to account for is the effect, discovered by Febble, that the bias has to approach zero at the 100% partisanship limits. The actual vote, itself, could be very off in those heavily partisanship precincts. For example, there could be very large numbers of uncounted Kerry votes in the Kerry strongholds, even though the "bias" for those precincts is near zero.

[Febble]

152. Zero bias at extremes

What I can't figure out how to account for is the effect, discovered by Febble, that the bias has to approach zero at the 100% partisanship limits. The actual vote, itself, could be very off in those heavily partisanship precincts. For example, there could be very large numbers of uncounted Kerry votes in the Kerry strongholds, even though the "bias" for those precincts is near zero.

There's 3D plot in my paper that might help, but you are absolutely right - the WPE exaggerates the bias in the even precincts, and minimises it at the extremes - and will minimise it more at the Kerry end, although of course both will eventually go to zero - it simply plummets to zero more precipitously at the high Bush end.

Here's a link to the paper if you want to see the plot:

http://www.geocities.com/lizzielid/WPEpaper.pdf

It's on page 11 (but it doesn't show the 100% - you'll have to imagine the surface going to zero at the edges!)

What my bias function was supposed to do was to enable bias to be measured properly at the extremes, as well as to prevent it being exaggerated in the middle. And you can see this effect if you compare the two plots. Essentially it lets us see where the problems are, without the mathematical confound with the vote-count margin itself.

[Time for change]

134. You say that it says something about Mitofski that he

asked you to conduct an analysis of his data. Presumably that is meant as a compliment to him, in that he was taking a risk on being shown to be wrong.

So what does it say about him that:

1) He released his original report not merely touting the shy Bush voter hypothesis without any evidence to support that hypothesis, but actually asserting it as a fact: “Kerry voters were more likely to participate in the exit polls than Bush voters.”?

AND

2) While continuing to maintain that position he refuses to release the raw data, which would allow qualified persons the possibility of either supporting or refuting his hypothesis on the best available evidence?

[autorank]

136. You will be put on notice soon to "stop making sense" Great Questions

It's time they're answered.

[Febble]

144. Two answers:

So what does it say about him that:

1) He released his original report not merely touting the shy Bush voter hypothesis without any evidence to support that hypothesis, but actually asserting it as a fact: “Kerry voters were more likely to participate in the exit polls than Bush voters.”?

First: the E-M actually does provide evidence, but it is circumstantial, and without statistical details it cannot be weighed independently. This is a problem.

Second: non-response bias is a major problem in any survey, and the subject of many textbooks on survey methodology.

http://writing.colostate.edu/references/research/survey/com2d4a.cfm

has a paragraph on the subject.

Also, the E-M polls have shown significant bias in every year since 1988 according to the E-M report. It should therefore not be surprising that a survey methodologist would consider non-response bias the most likely cause of the problem. It is also, incidentally, not true that "exit polls are always right". In the UK they are frequently "wrong", and usually over-state the Labour vote. In the UK we have verifiable, auditable, paper ballots and an transparent, independent scrutineering system. If exit polls are wrong in Britain, then we can be virtually sure the problem is in the poll, not the count. So there is plenty of precedent for non-reponse bias in exit polls.

That's what it probably says about him - that he doesn't find non-response bias implausible because he knows it happens. It doesn't mean that fraud was not the reason for the discrepancy this time.

2) While continuing to maintain that position he refuses to release the raw data, which would allow qualified persons the possibility of either supporting or refuting his hypothesis on the best available evidence?"

To obtain ethics approval from any university ethics committee, the researcher has to go to enormous lengths to demonstrate that a) the raw data cannot be traced to any respondent, even using indirect inferences and b) that any report released does not allow respondents to be identified. I assume these ethical standards also apply to commercial polling. They certainly should.

[OnTheOtherHand]

146. on the second point

The AAPOR Code says, "Unless the respondent waives confidentiality for specified uses, we shall hold as privileged and confidential all information that might identify a respondent with his or her responses."
http://www.aapor.org/codeofethics.html

I guarantee that if E-M released all the data with precinct identifiers, some individual voters would be identifiable by their demographic responses. As a Democrat, I like being on the side of protecting privacy. I know some people figure that knowing the precinct identifiers would give us some places to focus the investigations, but don't we already have places to investigate?

There's a lot more to be said on this topic. I just think it's important for people to know that Mitofsky isn't acting like a crook, he's acting like a survey researcher. I don't know for sure whether he is a crook, but I haven't encountered any survey researchers who think so.

[Time for change]

149. I find that very difficult to believe

Please tell us what data E-M could release (for the purpose of being able to conduct a transparent analysis of the data that would allow U.S. citizens to better understand what went wrong with the 2004 election) that you are concerned about that would be capable of identifying an individual respondent.

[OnTheOtherHand]

154. what data could identify an individual respondent

Sometime back in the 1980s, I had a problem. A survey research group had conducted before-and-after surveys with about 200 people in each of 5 different cities, but there were no ID numbers, so there was no way to match "befores" with "afters." Or no direct way. However, the participants had answered something like 9 pretty standard demographic questions -- sex, age, income, race, party affiliation, I don't remember what-all else. Those demographic fingerprints let me reliably (at least to my own satisfaction) match well over half the questionnaires.

In this data-mining era, I bet someone has access to all that information about me right now. It's no particular secret that I voted for Kerry. But revealing that should be my call, not Mitofsky's (well, OK, I wasn't exit-polled anyway).

But I don't know if I'm really answering your question. We might be able to get E-M to release enough data to conduct a "fairly" transparent analysis without compromising anyone's privacy. I'm just saying that if they release everything, I am pretty sure that some individual votes will be revealed.

(Edited to add the following content)

Bear in mind that E/M has already released raw data with their internal precinct identifiers. If they do another release that allows us to match internal precinct identifiers with actual precincts (even indirectly), then it doesn't matter if they strip the individual demographics or not -- the information is already out there.

I'm not sure exactly how blurring works (in the context of what Fritz Scheuren and the Election Science Institute did in Ohio). More information about that will be coming out. Apparently the idea is that E/M incorporated the official vote totals, but fudged them a bit so that precincts couldn't be reliably matched. That was good enough for what ESI wanted to do in Ohio.

[Time for change]

156. Ok, I think then that we agree on the bottom line

You say:
"We might be able to get E-M to release enough data to conduct a "fairly" transparent analysis without compromising anyone's privacy. I'm just saying that if they release everything, I am pretty sure that some individual votes will be revealed."

I agree with that. They don't need to release everything, but certainly every attempt should be made to release enough data to conduct a fairly transparent analsysis without comprosing anyone's privacy.

This really needs to be done, since the fate of our democracy may depend on it.

[OnTheOtherHand]

157. hurray!

(As I've said elsewhere, I really love the part where we get to agree.)

I haven't figured out the answer about the right balance between transparency and privacy -- I don't know enough about the ESI/Ohio approach to judge it. But I applaud ESI for at least moving the ball and finding something that E-M could say Yes to. I think the Ohio data are supposed to be available on the ESI web site any day now, and then maybe we can figure out what they actually did!

[TruthIsAll]

160. Privacy is a red herring. We don't want names. n/t

.

[OnTheOtherHand]

161. no, but...

As I tried to explain upthread, if we have precinct identifiers and demographic details, we -- no, let's say they -- can probably get the names themselves, or at least some of them.

Whether this bugs you or not, some people get paid to worry about it. That's not my gig, except that I've served on an Institutional Review Board that does basically the same thing: worry about any conceivable way that information could fall into the wrong hands and hurt someone, because some well-intentioned researcher (most often a college senior) hasn't quite thought everything through.

I'm not saying we should meekly accept anything Mitofsky says about privacy, I'm just saying that he isn't making the whole thing up.

[Time for change]

148. With regard to your two answers

I don't find them very convincing, for the following reasons, so I would appreciate if you could clarify your answers:

1. With regard to my first question, you state "E-M actually does provide evidence, but it is circumstantial ..." But you don't say what that evidence is. I read his report and I found NO evidence to support the shy Bush voter hypothesis. Neither did USCV. If you maintain that he did provide such evidence, please state what that evidence is.

With regard to your statement that non-response bias is a major problem in any survey, you are over-stating that point. Non-response bias is a POTENTIAL problem in any survey. It isn't necessarily a problem, and certainly not to the degree that Mitofsky proposes it to be in his 2004 exit polls. True, it always does need to be considered as an hypothesis. But Mitofsky doesn't state it as an hypothesis, he states it as a fact. Tell me that that isn't intellectually dishonest.

And with regard to your statement that E-M polls have shows significant bias (and by "bias" I assume that you are talking about the discrepancy between the exit polls and the official results) every year since 1988, please point out that the discrepancy between the E-M exit polls and the official results were greater in 2004 than in any previous year.

2. In response to my questioning Mitofsky's refusal to release his raw data (which could settle many of the arguments between him and USCV and provide the American people with a more thorough understanding of what went wrong with the 2004 election), you bring up the need to maintain the confidentiality of the respondents. In my line of work (epidemiology) we regularly deal with raw data where subject identifiers has been removed so as not to compromise the confidentiality of their responses. Are you saying that that is not possible to do in this case?

[Febble]

151. OK:

First of all, I don't find the evidence in in the E-M report compelling either. But here is my take on what was presented in the report as evidence.

It is very difficult to research differential non-response, for the obvious reason that the people you haven't sample are the ones you don't know about, so you can't compare the two groups (sampled and unsampled) to see if they are alike. However, if there is an underlying response bias (one group is more likely to avoid the poll than the other) you would expect the error to be greater if circumstances were such that good random sampling would be difficult. These were hypothesised to include precincts in which the interviewers had to stand a long way from the polls, precincts in which the interviewers said they had been poorly trained; precincts in which the sampling rate was high. The report gives crosstabs for the mean WPE for each of these, and other factors, and the WPE tends to be higher where those factors were present. So factors likely to affect true random sampling could be regarded as a proxy variable for the underlying bias it could have exacerbated.

But there is no statistical detail given, so we cannot weigh up the evidence in any quantitative way. When this kind of report is written in an academic journal, as you will know, probability values and effect sizes would be given, and people could judge whether the factors cited are sufficient to account for the effect. But they are not given in the report. Also, to refer to my paper, the WPE is a really poor measure, as it is confounded with partisanship - and partisanship is likely to be associated with the "predictor" variables - the variables used to account, statistically, for the WPE. In effect, you have a term on both sides of the regression equation - i.e. some measure of partisanship appears on both sides. Which means the answers, even if we knew the effect sizes, may not be valid.

As for over-stating my point - yes, it is only a potential problem - it may not happen. But any survey where mean response rates are as low as around 50% runs quite a risk. I'd be unhappy, in my own research, with response rates less than about 85%. And the fact is that we don't know that it wasn't a problem, any more than we can know that it was. But I would agree with you, it isn't a fact.

Re bias: I myself use the word "bias" neutrally - something was biased. The question is, was it the vote-counting or the polling?

Lastly - I believe a set of "blurred" data on Ohio was released and has been analysed independently, so it would appear it is possible.

[Time for change]

158. Thank you for your honest responses

I can't comment on the possibility that Mitofsky may soon release "blurred data" because I don't know much about that.

But your responses raise some very important issues, that I would appreciate your clarification on:

1. You agree with me that additional detail in Mitofsky's report, such as probability values, would be helpful in understanding many of the issues that he raises, and you note:
"When this kind of report is written in an academic journal, as you well know, probability values and effect sizes would be given, and people could judge whether the factors cited are sufficient to account for the effect. But they are not given in the report."
I most certainly agree with that, as I have never submitted a manuscript to an academic journal without including probability values with respect to the main analyses, and you probably haven't either.

I don't know about you, but I have never written a report that comes close to having the national and international significance that Mitofsky's report has. So if these things are important enough to be required in the reports that we submit, why aren't they important enough to be included in Mitofsky's report? And what does the fact that they aren't included say about Mitofsky's purpose in writing and publishing his report?

2. You note correctly that:
"if there is an underlying response bias (one group is more likely to avoid the poll than the other) you would expect the error to be greater if circumstances were such that good random sampling would be difficult. These were hypothesised to include precincts in which the interviewers had to stand a long way from the polls, precincts in which the interviewers said they had been poorly trained; precincts in which the sampling rate was high. The report gives crosstabs for the mean WPE for each of these, and other factors, and the WPE tends to be higher where those factors were present."

So let's look at the data that Mitofsky submitted with respect to your examples that purportedly support his rBr hypothesis (acutally, he stated this as a fact, not a hypothesis):

Mean WPE is -6.5
With regard to distance from polls, the closest distance (inside the polls) is associated with a WPE of -5.3. So, aside from the fact that there is no analysis as to whether this variable shows statistically significant variation, even in the best of circumstances there is still a WPE of -5.3. So how can distance from the polls explain a consistently negative WPE?

With regard to training: again, no statistical analysis, and even the best category ("well trained") is associated with a WPE of -6.4, which is only 0.1 difference from the mean. So, same question there.

And I'll add one more example: Interviewers with an advanced degree (the highest degree of education) were associated with the most negative WPE -- -7.9. Is this a circumstance where you would expect the error rate to be greater because "circumstances were such that good random sampling would be difficult"?

Do you really believe that any of this provides ANY support for rBr?

3. I think you misunderstood my question about "bias" in previous elections (Or maybe you were just so fatigued from responding to this thread that you simply missed it.):

Here's what I meant: You noted that there has been "exit poll bias" in all previous elections. I simply asked you to please acknowledge that the discrepancy between the exit polls and the official result was greater in 2004 than in any previous election for which we have data.

[Febble]

159. I absolutely agree

1. that the E-M report was very disappointing, not to say enfuriating. I was enfuriated. I do not know why these details were not included.
It was not in any sense an academic report. I won't speculate on why it was not, because it would clearly be inappropriate, given my position. But it is certainly why I, as well as USCV, TIA and many others, have spent many good hours that might have been better spend doing something else, poring over spreadsheets trying to figure out what could be inferred from the little we know. It's why I got into this damn mess in the first place.

2. that we cannot know the answer as to whether those factors actually account for the discrepancy without the results of a multiple regression analysis, which was not given in the report. As I said, I don't think what they provide is compelling. But "weak evidence" or "implied evidence without the supporting R squareds" is not the same as "no evidence". Which was all I was implying. What evidence was provided was certainly not given with enough statistical information to be compelling.

3. That the WPE, at least, was larger in 2004 than in any of the 5 elections for which the E-M report gave figures. I assessed each year here:

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

And produced p values. My analysis indicated that on the basis of the WPE, the state mean signed WPEs were a) greater in 2004 than in any of the five four previous elections and b) significantly greater than in 1992. In addition, in every year the mean signed WPE was significantly greater than 0 (i.e. there was significant bias, in either count or poll). Ironically, although 2004 was significantly greater than 1992, 1992 was more significantly greater than zero than 1992, from which, you will be able to infer, means that the between-state variance was less than 1992 than in 2004. Probabilities are a poor proxy for effect sizes, which is why we need both!

Let me know if you have any more questions (or if I have not answered something I should have), by email, if you want to stop adding to this thread :).

Lizzie

[Time for change]

162. Thank you, I think now we're pretty close in the way we look at this

Our only difference is that I feel that Mitofsky provided NO evidence to support his rBr hypothesis and you characterize it as "weak evidence".

[Carolab]

143. This might sound extremely simple-minded but...

A while upthread Febble gives an ultimatum: "Widespread bias or widespread fraud. In both directions. Take your pick."

I'm certainly not a "mathefier" by any stretch, but couldn't this noise be indicative simply of stealing/switching votes from one candidate and giving them to another, in a rather random way but based on some sort of logic that would try to reduce the appearance of fraud?

[OnTheOtherHand]

145. not an either-or

I doubt Febble actually intended that as an "ultimatum." There could be widespread bias, and widespread fraud, and lots of random noise, so we don't have to pick just one.

I think there is lots of random noise. But I also think that if someone could alter lots of vote results centrally, it would be smart to do it in a way that would be hard to detect, as you suggest.

So, the exit poll doesn't prove much either way. We have to work to check -- and shut down -- the specific ways in which fraud could enter the system.

[Febble]

147. It certainly wasn't an utimatum

more an invitation.

And not an either/or - pick and mix is also just as possible.

It's a buffet :)

[Carolab]

153. A "smorgasbord" as it were. n/t

[Febble]

155. Absolutely!

BYOB :)