Statisticians debate validity of 2004 election results (Freeman/Mitofsky)

[FogerRox]

Statisticians debate validity of 2004 election results
By eric chen
October 21, 2005

Eleven months after the 2004 presidential election, statisticians continue to debate the exit poll results.
Last Friday, researchers Steve Freeman and Warren Mitofsky gave a joint presentation in Logan Hall about the debate.

On Election Day last year, exit polls showed Massachusetts Sen. John Kerry leading President George Bush by 51 percent to 48 percent. However, when officials released the vote count later that night, it was Bush who prevailed by about two and a half percentage points.

The reason for the discrepancy between the official vote count and the exit poll estimate is a point of contention for many statisticians. Freeman blamed the deviation on election fraud, while Mitofsky disagreed.

Freeman, a member of Penn's Center for Organizational Dynamics, said that massive vote-count corruption -- on the scale of 8 to 10 million votes -- was to blame. He said that vote suppression and manipulation as variables tied to fraud

.rest of story-
http://www.dailypennsylvanian.com/vnews/display.v/ART/2005/10/21/43589c6ab3b9f

[smoogatz]

1. K&R

Very important story. Interesting that Mitofsky's out in public, talking about this. Interesting that he still can't think of a reason why Bush voters were so reluctant to respond in swing states only, and why such a thing has never happened in an exit poll before.

[FogerRox]

2. It would be nice if Edwin would consider

how he might be wrong-- just as an exercise in Due Diligence

[OnTheOtherHand]

3. fact check

"he still can't think of a reason why Bush voters were so reluctant to respond"

The observed results could be any combination of reluctant Bush voters, eager Kerry voters, inept interviewers.... I imagine many survey researchers can think of quite a few reasons; what we can't do very well is to test our hypotheses after the fact.

"in swing states only"

Most of the red-shift was not in swing states; the largest discrepancies were Vermont (by percentage) or New York (by vote).

"and why such a thing has never happened in an exit poll before"

1992 presidential (within precinct error about 3/4 the size of 2004)
1992 NH Rep primary (10-point error; "I've never done anything that bad" -- W. Mitofsky)
1989 VA governor (about a 10-point error)
1988 presidential ("President Dukakis" reportedly ran ahead in one exit poll)
1980 presidential (ABC _raw_ results showed Carter beating Reagan by 2.5 -- of course, as I've tried to explain, raw results aren't actually used to make calls, and I don't have weighted figures, so I don't know how large the prediction errors were)
...umm, another FAQ for me to write. Sigh.

[papau]

6. Then again - and more likely - the answer could be fraud - but we dare not say

anything - which of course forces us back to a great combination of reluctant Bush voters, eager Kerry voters, and inept interviewers - non of which has an iota of even gossip to support it.

So we go with desperate tries for past examples to show it was not really that bad (in the 50's and 60's exit polls died because of documented cheating - by both Dem and GOP - indeed Illinois never had a clean GOP win in all the years I watched it - and I am talking the "downstate" countries where all the family value GOP lived - so yes exit polls can be very wrong in a given situation - indeed they can be very wrong when there is cheating).

1992 presidential within precinct error about 3/4 the size of 2004 was going to be avoided by sheer numbers, better training, better math - and seemed to have been post-92 avoided - until suddenly it got worse than 92.

And the "1988 presidential ("President Dukakis" reportedly ran ahead in one exit poll)" is a joke - right? What "national exit poll" with more than 10,000 data points showed President Dukakis?

And then you mention the 1980 presidential and ABC _raw_ results showed Carter beating Reagan by 2.5 -- what the hell that has to do with an exit poll error I have no idea. Carter was indeed coming back and in a week if trends continued might have done well - but at the date of the election I recall no exit poll that called it for Carter - and I am talking about exit polls before the standard procedure of a final fitting to the "real" vote.

I am looking forward to your FAQ write up.

:toast:

[Febble]

10. Actually

there is a bit of gossip about all those things.

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

[OnTheOtherHand]

13. what Febble said

about gossip. Moreover, there is the saying that "whatever is, is possible"; there have been plenty of exit poll errors beyond nominal margins of error, and certainly no one has proven that they were all due to election fraud.

"1992 presidential within precinct error about 3/4 the size of 2004 was going to be avoided by sheer numbers, better training, better math" -- well, bias can't be avoided by sheer numbers or better math, and it remains to be seen whether better training can do the trick.

1988 -- I wish I knew. Richard Morin claims that it was the ABC poll
http://www.washingtonpost.com/wp-dyn/articles/A64906-2004Nov20.html
(which was huge, by the way), but the dataset doesn't seem to bear that out.

The 1980 ABC exit poll, n = 15,201. Again, for all I know the weights may have cleaned that one up even before official returns came in; I haven't found a discussion of it.

[FogerRox]

7. yes- and the evidence of actual rigging is in the computers-- probably in

Ohio-------

And they have been used for another election so the evidence is gone--as far as any code designed to RIG the election is concerned.

'cept for the paper ballots that were scanned, and the punchcards. Maybe.

So it comes down to not being able to prove election rigging, and not being able to prove the exit polls were fucked up.

It seems that the election of 2004 results are in question, and the proclaimed victor is, I guess the victor.

SO lets impeach and indict all the basturds.

[texpatriot2004]

21. Let's do impeach and indict all the bastards. n.t

[Febble]

4. Mitofsky has been telling

any one who cares to listen exactly why he thinks the exit polls were off for the best part of a year now.

It's your prerogative not to believe him, but there was a seventy-odd page document issued in January about precisely that.

And actually he does give a reason for the apparent slightly greater WPE in swing states (although the biggest difference appears to be between red and blue states) which is that he has noticed that he gets a larger WPE (i.e. more "red-shifted" results") in years in which the interest in the election is high, and he suggests that the same effect might apply in swing states. I presume the argument is that where people are more engaged any underlying differences may be exaggerated. But it is a very small effect, and appears to be outweighed by a much bigger effect of blueness. The biggest WPEs were in the blue states.

As for the claim that it "has never happened in an exit poll before" - well again, the evidence that it has is given quite clearly in the January document. In 1992 the effect was almost as large as in 2004, and it was fairly large in 1988 as well. 2000, the year of the "stolen" election, it was unusually low.

I would also note that "shy Tories" are a well known UK phenomenon, and pollsters routinely assume that Tory voters are under-represented in their poll. I suppose if your political stance is that you want government out of your business, it wouldn't be that surprising that you might want pollsters out of your business too. Not saying that's the reason, but it's certainly not an unprecedented phenomenon.

[papau]

8. When did "shy" not get included in the weighings? Last I looked suppressio

of black votes did not show up in exit polls.

Indeed "shy" was assumed reflected in other variables - aggregate accuracy being more important than having 99 variables spot on and one variable off by 1000%

The we discuss WPE - yet we can not get detail to test by type of machine or by control of the voting apparatus.

Never happened before - ?? - sure it happened before - when there was cheating.

Indeed getting detail might help us pin point the bad guys. It may be a surprise, but not all GOPers are OK with cheating.

[Febble]

9. Suppression of black votes

will not show up in the exit polls if they never get to vote. But if black voters are more likely to vote Democrat, and also more likely to be issued with provisional ballots, and these are subsequently rejected, that will show up in the exit poll. And if the votes of black voters are more likely to be spoiled (and there is good evidence for this, although the reason is unclear), then that will show up in the exit polls as well. In the WPE.

I don't quite understand the other things you are saying about "shy".

I agree that not all Republicans are OK with cheating. It seems to me past time this was a bipartisan issue.

[papau]

11. There is an ongoing discussion of the value of more variables versus the

value of making fewer variables more accurate in the result (the variables may be off individually but errors offset each other in some way that produces a solid answer).

Given the end of the process is always the result - any method we used to get there has taken into account "all variables" - just by definition of getting a "result". The implicit variable approach is always not as clean as the explicit - but again by definition, "shy" was included implicitly.

[Febble]

12. But what are you actually talking about?

Sorry, I don't understand.

[papau]

15. The amount of the change is not made less by saying GOP folks are

shy.

Folks act like the massive change in weightings is not really that massive because we forgot that GOPers are shy.

And that is bull.

Perhaps the party has suddenly become more shy - another amazing excuse. But "shy" GOPer explains nothing.

Your best effort on the weightings adjusted the m/f and all other weights to get to the expected vote. Within those weightings are implicit assumptions as to everything that is not explicit, with the implicit assumption being the not explicit assumptions will all offset and not affect the outcome.

Saying "shy" in no way explains a damn thing.

[Febble]

17. Well I tend to agree

"shy" is a poor description and I'm not sure where it originated.

Age, race and gender of non-responders are noted in the poll; if the proportions of each among the non-responders is different from the proportion of each in the responder group, then the responder data can be weighted so that the two groups match. And I suppose we could call the non-responders "shy". But the weighting isn't done for "shyness".

And the trouble is, only visible characteristics can be noted. If white male Bush voters are shyer than white male Kerry voters, there is no way the two can be distinguished by eye. (And I don't suppose white male Bush voters are actually shy - but they might be unco-operative about co-operating with a project of the Librul Meeja.)

[The Doctor.]

28. Did Freeman cite the 'correction' of exit polls in the MSM

on election night and ask about precedent?

[SmileMaker]

5. Exit poll experts needed in NJ to give credibility to citizen's exit poll

I'm about to give up on this idea, but I have it all ready to go in Princeton (and pdf sample ballot file for all races in Mercer County) if I had the volunteers and someone who is an expert pollster. Roger - do you know of any exit polling being done in NJ? My pleas for help have been met with silence pretty much. I hate to give up on this :(

After running this by a ton or voting rights activists in NJ and getting nowhere, I appealed to Woodrow Wilson School (they have a new intistute that focuses on NJ that includes voting issues).

Final plea!!  

I have designed a ballot and come up with  a plan to conduct a citizen's
exit poll at one polling place in Princeton, NJ.  The work that I have
done so far could be duplicated if there is  interest in doing exit
polling in more than one polling place  I have one volunteer so far :(   It's been
shocking how hard it has been to convince people of the importance of
auditability; especially in a district that has an assembly candidate
(Gusciora) and a Congressman (Holt) who have promoted or introduced
verifiable voting legislation at the state and federal level.  

The Coalition for Peace Action (peacecoalition.org - they've been involved
in voter verification efforts)  would support this if I had a
person/organization of some prominence with experience in polling and
Republicans involved.  I would like to do this project experimentally at
Community Park School in Princeton as preparation for 2006 where there is
more at stake nationally and to raise awareness about the vulnerability of
our voting system. 

This is very serious and we need to find ways to correct our system before
better laws take effect.  I hope that there might be some in your
organization who would want to get involved in this with me.  If not, I'll
probably not do it because it would be pointless without the support
needed to do it right. 
 
 

Citizen's Audit Exit Poll Flyer
http://www.princeton.edu/~slalbert/citizensaudit.doc

Citizen's Audit Exit Poll Ballot
http://www.princeton.edu/~slalbert/ballot-1.doc

Mercer County Official Sample Ballots.pdf
http://www.princeton.edu/~slalbert/05-General-Sample-Eng.pdf


Thanks to the Parallel Election people in FL and California - I spoke with
Ellen Brodsky (FL) and weaved some of her ideas with wording from the CAPE
project in CA. I decided to use the words exit poll rather than parallel
election.

For more info on Parallel Elections visit -
www.ecotalk.org/ParallelElections.htm
 

[FogerRox]

14. Smile Maker--

do you know Irene Goldman @ CFPA? I do- maybe I could introduce you- ?

see my sig? email me--

And on Dukakis--- he was 17 pts ahead in August- then refused to compete in a hard race-- his lead steadily eroded--- thru the entire race--- I think it was clear that by oct. 1st Mike was going to lose.

[BeFree]

16. It's really quite obvious

The $10 million dollar exit polling process manufactured by E&M either was a waste of money because they didn't do the job properly and are now trying to cover-up their mistakes so they can continue to get work, or it did the job all too well and found evidence of fraud.

Either way, E&M is negligent. Negligent because they screwed up, or negligent because they are attempting to cover-up what was uncovered.

Cover-up of fraud gets my vote. I can hardly believe they screwed up the polling. They are professionals. Too, given they are hiding some data that could clear up the controversy, and given that they have stated there was no fraud, points to E&M engaging in a cover-up.

Had E&M said it could have been fraud that screwed things up, I would be praising them today, instead of damning them.

[Febble]

18. Oh, for goodness sake.

They did what they were commissioned to do, in exactly the manner they said they would do it, on their website, in great detail, including a clear statement in the FAQ (that vote count data would be included in the projections).

If you want an exit poll to audit the election, than that is what needs to be commissioned. The poll wasn't designed to audit the election. It was designed to allow the networks to call the states for each candidate in advance of the final count. It did so. The networks were happy - a lot happier than they were in 2000 when they messed up Florida - that, you might legitimately argue, was negligent. Frankly, I think it probably cost Gore Florida.

This year, Edison-Mitofsky were determined not to have any state called erroneously (by which I mean, strictly, "calling" a state differently from the result of the final count). And they didn't.

However, they needed vote-counts from their precincts and counties in order to this. Possibly because there was fraud. Possibly because one of their data sources was unusually biased. Which of those is the case is what we are debating.

[BeFree]

19. Yes, for goodness sake

E&M should get over it and move on.

Release the relevant data so that we can get to the bottom of this controversy they manufactured. For goodness sake, come clean. Quit saying there was no fraud. They can't prove there was no fraud, so quit saying it.

Come clean and say simply this: It was skewed, we think it may be our fault or it may have been fraud. In the interest of goodness sake, we shall allow the relevant data to be examined - with safegaurds.

There, I wrote yall's press release for you.

[Febble]

20. Fair enough

And in fact this was done, for the Ohio dataset. But the "safeguards", which I agree are important, involve "blurring" the data, a technique developed by Fritz Scheuren, president of the American Statistical Association, I think originally for confidential tax documents.

The problem is that the blurring is a fairly labour intensive process, as it is important that it does not obscure real relationships in the data, or introduce artefactual ones. Scheuren's paper on blurring is here:

http://www.amstat.org/sections/srms/Proceedings/papers/1986_070.pdf

(not that I recommend it as light reading)

I would be in favour of datasets being prepared and released for other states, particularly other key states such as Florida and New Mexico.

[Peace Patriot]

22. 80% of the 2004 votes were tabulated by two rightwing Bushite electronic

voting corporations, Diebold and ES&S, using SECRET, PROPRIETARY programming code. That's really all you need to know about the 2004 election. It was egregiously non-transparent. But there is so much more.

This new electronic voting and tabulation system--which many experts warned was extremely insecure, unreliable and hackable--was put into place in the 2001-2004 period, under the false guise of "election reform," with $4 billion in funding to the states that acted, effectively, as a bribe, for the states to quickly convert to these untested electronic systems, and pour money into the pockets of the major Bush donors who manufacture and control them. In providing the funding, Congress failed to require any kind of "paper trail" to audit the accuracy of electronic vote counts. Tom Delay--recently arrested for campaign money laundering--bottled that provision up in committee.

They also failed to put any controls on "trade secret" vote tabulation or lavish lobbying of state officials. The Bushite companies (major Bush donors and campaign chairs; billionaire supporters of far rightwing causes) have spent millions of dollars on lavish lobbying, including, for instance, a week of fun, sun and high end shopping for election officials from around the country, at the Beverly Hilton, this August, sponsored by Diebold, ES&S and Sequoia:
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x380340


These Bushite corporations lobbied against even a meager "paper trail" and got their wish in one third of the country, and, with miserable auditing/recount provisions everywhere else--even where some kind of "paper trail" exists--the election was unverified, and unverifiable.

This was a new election system being tested out for the first time nationwide. Given these facts about the election system--its insecurity and hackability, and who owns and controls it--the situation CRIED OUT for independent verification tools in 2004, such as exit polls, which are used worldwide to verify elections and check for fraud. The war profiteering corporate news monopolies, acting in concert, instead hired Edison/Mitofsky to CONFORM their exit poll results TO FIT the results of Diebold's and ES&S's secret formulae.

Kerry won the exit polls, but the exit polls were ALTERED--in ways that defy all probability--late on election day, to FIT the results of an unverifiable and partisan-controlled new election system.

The American public was thus deprived of major evidence of election fraud, which helped to squelch election protests and investigations. With the Democrats having blown the Republicans away in new voter registration in 2004, nearly 60/40, and Kerry winning both the state and national exit polls all day long (until the doctoring of the data), voters were angry and mystified by the apparent late in the day reversal of Kerry's fortunes. UNKNOWN to them, the exit polls had in fact confirmed their conviction that Kerry won. But, WITHOUT THIS KNOWLEDGE, they had little ground to stand on, except for the massive, overt Republican suppression of black, poor and other Democratic voters in Ohio. The fiddling of the exit polls materially changed the political landscape--and sent the likely winning voters into resignation and despair.

Only a few alert bloggers knew what was going on--had captured screen shots of the real exit poll results--and began desperately to analyze what happened and to get the word out to the general public, against the overwhelming indifference of the war profiteering corporate news monopolies.

Edison/Mitofsky, which has refused to release most of their raw data to independent researches, has been trying to cover their asses ever since--including their lame and unsupported assertion that their own polls were biased in favor of Kerry--and they have furthermore promised that it will never happen again that independent minds view their early results.

All future elections will now be locked in a perfect loop of privatization: secret, proprietary vote tabulation; secret, proprietary exit polls--both controlled by warmongers, fascists and the super-rich.

Febble defends this inexcusable behavior by saying, that, well, E/M disclosed on their web site that they were going to fiddle the numbers, and that E/M was doing what its clients had asked them to do.

That is horse manure, to put it politely. We are looking at a war conspiracy--at a manufactured endorsement of Bush's war for the benefit of war profiteers.

If the Bushite Congress had wanted a transparent election in 2004, we would have had one. It is not difficult. Instead, they did everything possible to produce a NON-transparent election, and placed every possible obstacle in the way of transparency. Why? Can any adult person living in the world today believe that, having CREATED the capability of switching millions of votes behind closed doors, in nanoseconds, at the touch of a keyboard, leaving no trace, that Karl Rove & Co. didn't USE it to keep their little dictator in office?

And the news media lapdogs, who propagandized an unjustified war, in which tens of thousands of innocent people were slaughtered, CONSPIRED to confirm that manufactured result.

That is the truth of it. And the only question remaining is, how many Democratic Party leaders who sold away our right to vote, for lavish lobbying junkets, or beneficial tax cuts, or war profiteering, or to please their military/corporate sponsors, need to be purged from the Party so that the people of this country CAN GAIN PROPER REPRESENTATION IN WASHINGTON DC BY AT LEAST ONE POLITICAL PARTY!

Yes, I am livid with anger at the triviality of the discussion occurring on this thread.

NOTHING can excuse what Edison/Mitofksy and their corporate paymasters did. Nothing! To hide evidence of another Bush fraud! To change data! To withhold data! To drag the discussion off into P.R. "talking points."

WE ARE TALKING ABOUT THE LOSS OF OUR COUNTRY AND OUR DEMOCRACY!!!!!!

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

See, for a project for statistical monitoring of the '06 and '08 elections, to help with election challenges--a project that needs our DONATIONS:
http://www.UScountvotes.org

See, for an easy to read pamphlet on the perils of electronic voting,and a project to educate election officials:
http://www.votersunite.org

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

We need...

1. Paper ballots hand-counted at the precinct level (--Canada does it in one day, although speed should not even be a consideration, just accuracy and verifiability)

or, at the least...

2. Paper ballot (not "paper trail") backup of all electronic voting, a 10% automatic recount, very strict security, and NO SECRET, PROPRIETARY programming code! (...jeez!).


-----------

Restoring transparency to our election system has to be done at the state/local level, where the control over election systems still resides, and where ordinary people still have some influence. Bush's "pod people" in Congress are not to be trusted, and WILL make things worse if they meddle any further (for instance, with anti-poor, anti-black "voter IDs", and centralized electronic databases that make it easy to purge unwanted voters, and impossible to get unpurged). The corruption at the state/local level is a formidable obstacle; still, state/local is the only viable venue for change.

We must also find OTHER means to verify election results, such as "parallel elections" and truly independent exit polls, and with massive requests for paper ballots (such as absentee ballots, which may be miscounted in the electronic tabulation, but which nevertheless provide a useful paper record).

We should,in addition, work to overwhelm the fraud with sheer numbers (more new voters; even bigger get-out-the-vote campaigns), which I believe it is possible to do, given the nature of electronic fraud, which may be limited to pre-programmed percentages and formulae, and is also constrained by the fraudsters' desire not to tip their hand with outrageous results.

And maybe the people in this country will finally just get fed up, and throw Diebold and ES&S election theft machines into 'Boston Harbor'! It has happened before!

[Peace Patriot]

23. P.S. See also...

TruthIsAll: The 2004 Election Collection
http://p216.ezboard.com/frigorousintuitionfrm9.showMessage?topicID=47.topic
TruthIsAll.net
or TruthIsAll.com

Contact: [e-mail]

[garthranzz]

24. Beautiful!

:toast:

:kick:

[OnTheOtherHand]

25. I'm hearing that you are angry

but so many of your comments about the exit polls are just plain wrong that there is probably not much point in trying, once again, to explain why. So, enjoy your righteous rage.

Personally, I suspect that some karma is incurred in lumping honest people with "warmongers, fascists and the super-rich," but only time will tell.

[Febble]

26. Well, it is a lovely rant

and I do love a lovely rant.

All I ask is that the rant is directed at the right people. There's quite a bit of collateral damage here, that better targetting would reduce.

You are absolutely right that the extent of the adjustment between the projections made at close of polls (before the vote-count data-stream came into play) and the projections made after the precinct and county-tabulations had been included in the inputs was astronomically beyond the possibilities of chance.

And you and me all were right to be suspicious.

But a few points really do need to be made:

The inclusion of the vote count into the final projections was not secret.
The projections made before the vote-count data was included were not secret, and not intended to be secret.
The inclusion of the vote-count was done in exactly the same way as it is done every year - except that this year, because of the discrepancy, it made a far greater difference to the projection

It is the CAUSE of the discrepancy that needs to be the focus of your fire, not the alleged cover-up. There was no cover-up. Everyone knew, including me in the UK, that the initial exit poll projections were for Kerry, and that the final projections were for Bush. No-one, least of all E-M, has attempted to deny it.

What we want to know is why that adjustment was necessary.


(edited for grammar)

[papau]

27. Well said - but why isn't detail being released so we can perhaps solve

the theft of the Century?

[Febble]

29. Well I'll tell you the reason

I think the detail isn't being released, although you may not like/believe it.

Here it is:

Firstly, that the ethical guidelines on respondent confidentiality for survey data are extremely strict. This is as it should be. It is absolutely essential that survey data are never released in a way that would allow the respondents to be identified.

Secondly, the "raw data" are always released, and and were again this year. These data include the detailed responses from every single respondent, together with demographic details (age, race and gender). However, to protect respondents from being identified, no precinct identifiers are given.

Thirdly: without precinct identifiers, the vote count recorded for that precinct cannot be linked to the precinct. If vote counts were provided for each precinct, the precinct itself could be identified - because the vote count will have a high likelihood of being a unique number.

This means that although a certain amount of analysis can be done on the raw respondent data, responses cannot be compared with vote-counts at precinct level - which is what everyone wants to do.

There is a solution, and it was developed by Fritz Scheuren, who published a paper on it in 1986. It was developed precisely so data (actually tax data) could be analysed statistically without compromising the confidentiality of the data. It is called "blurring". It is a fairly complex process as it is important that real relationships within the data are not obscured, nor spurious ones introduced. It is therefore fairly labour-intensive.

However, Scheuren commissioned a data set for Ohio, "blurred" to his specification, and it was analysed by ESI. From the ESI website, I gather the report is currently undergoing peer-review. The fact that this dataset was able to be prepared for Ohio suggests that there is no bar, in principle, to similar datasets being prepared for other states. However, given the concerns about Ohio, and given the complexity of preparing the data, clearly it was sensible to start with Ohio.

My own view is that if people really want to have more data available, rather than demand "release" of the data - as though it is sitting there like a laboratory animal, pining for freedom - it would be more useful to put together a specific proposal with a specific set of testable hypotheses, and raise funds for a decent study, including a suitably prepared dataset.

I'm not meaning to be sarcastic here; I am aware that to non-statisticians (I dunno, maybe you ARE a statistician) it seems like the data should be released and it will reveal all. Statistics isn't like that (or shouldn't be). Getting the data and looking for patterns and shouting when we've found one isn't what we do. Or if we do, we suffer such a enormous reduction in our statistical power that we might as well not bother. In order to maximise statistical power to find effects we need to have a coherent theory of what patterns we might expect to see - and then find out whether they are there. So far coherent theories (see my Calculatus Eliminatus thread) have been a bit thin on the ground. But if there is a coherent theory out there, my expectation is that it would be perfectly possible either to commission a dataset to test it on, or to ask E-M to test it.

You might not believe the answer you got - but if you didn't believe the answer, you might not believe that the data you were asking for were the real McCoy either.

[papau]

30. Thanks for the nice post - and I am- or have been in the past - a

statistician, if that word means making a living from knowledge of this area of math. A better word that covers 95% of my career is "actuary".

As to not compromising the confidentiality of the data - why does that go beyond a confidentiality agreement. I've used such data -always under such an agreement - and no one ever got hurt - meaning their confidentiality was maintained outside the analysis. Did the questionnaire not state that it would be analyzed? What the hell is the big deal in an exit poll that makes it different from any other stat work that uses "confidential data"?

As long as there is no public release of precinct data - meaning the resulting report is "cleaned " so that specific precinct data is not commented on - why is not everyone well within whatever confidentiality agreement is in place. Granted Tax data has a special law on confidentiality problem - so blurring is needed here - but why blurring on exit polls? SORRY - but it sounds like an unneeded demand for extra money.


If I am reading your "coherent theory is needed" comment correctly, you are saying that hypothesis testing is the only way to go - Ok - but why not start by just getting all the correlations developed and eyeballing them? A cross-tabs needs expected values and out. Can we get someone to check for the statistical significance in the relationship between the categories of two or more variables - as in throw in precinct data and machine used to record data and see what pops up?

The result will not be "meaningful" in terms of strength of association. But all I am hearing are demands for money - which to me sound like a way to slow down - or stop - anyone getting the evidence to put some GOPers behind bars.

In any case I am glad Ohio is getting done - but the demand for more money before doing a "blurred" Florida (etc.), and before doing a by type of vote counting method, and before doing a by party in control of vote counting in precinct - all on a national scale - is hard to understand.

You say you have a legal problem on confidentiality - are you sure your lawyers can not think of another way to solve that problem beyond more money for "blurring"?

[Febble]

32. Well, blurring would certainly

cost money. But that would seem to be the best solution. There may be legal solutions instead but I'm no lawyer.

[Febble]

33. Either I didn't fully read

your post before replying or there was something more added on edit.

The reason confidentiality is important is that an essential aspect of a democracy is the secrecy of the ballot. If you can identify a voter, you can identify their vote (or their reported vote).

OK. So you are suggesting that rather than "blurring" the data, it should be released to a finite group of researchers.

Maybe. I doubt that would satisfy most of the demands for release. Many of the demands (perhaps not yours) have been for public release of the data.

Maybe there is a solution. But any solution, as I see it, whether blurring of data or supplying data to a specific group, would require the articulation of a hypothesis. It is something I hoped we could thrash out on DU, and maybe we are. It is certainly why I post here!

Cheers.

[papau]

34. hypothesis - Electronic voting plus GOP control of local gov = bias

Please test -

and in test break down by type of voting machine and by manufacturer.

Thanks

[Febble]

35. Good hypothesis

We should collect these.

[OnTheOtherHand]

36. by the way, I think it is worth underscoring --

You wrote to Febble, "You say you have a legal problem on confidentiality - are you sure your lawyers can not think of another way to solve that problem beyond more money for 'blurring'?"

I'm not sure who "you" is in that sentence, but just so we are all clear -- Febble is not E/M or even an E/M employee (although she has done contract work for E/M). She does not have lawyers at her disposal, at least with regard to the exit poll data. (I am not an E/M employee either.)

As far as I can tell, the privacy-heads at AAPOR and ASA agree that there are real confidentiality issues with the exit poll data. That doesn't mean that there aren't ways of accommodating them. It's just good for everyone to bear in mind that {b]Febble has no way of accommodating them.

There is certainly no compelling privacy rationale for not releasing some T tests on the machine results. I think I have a pretty good idea how those would turn out: inconclusive. (But your interaction hypothesis downthread might provide more leverage.)

[Febble]

37. Thanks, OTOH

failed to spot the "your" in there.

As OTOH says, I have done limited contract work for Mitofsky, and I am not on the E-M payroll. I don't even live in America, and I don't know much about law.

I do know about ethical guidelines regarding confidentiality of data, however, because I spend many hours, as part of my job, applying for ethics approval for studies in which participant confidentiality is considered paramount. I have to state very precisely the steps by which data I collect will be secure from being accessed by anyone in a way that would enable participants to be identified.

It is not difficult for researchers in the social sciences (my field is psychology) to understand why the data people want has not been freely released.

[papau]

38. I don't believe "confidentiality of data" needs are being questioned -

it is just that release to a larger group for analysis via a confidentiality agreement is SOP in the US - so I wonder why the chase for money via "blurring" costs demands.

The Univ of Mich does the Justice Department cross-tabs of "confidential" information under contract with DOJ - why is this not possible with the exit poll data?

Does M/F need a contact name at the U of Mich?

I can not believe that M is not networked into the folks that do this stuff.

:toast:

:-)

[Febble]

40. I don't think there's any chase for money

but the kind of work that blurring entails would presumably cost some. It's not something you could do with the click of a mouse. You might like to check out Scheuren's paper.

[CalmMan]

31. Mitofksy's interpretations failed to satisfy many in the room

This is a very weak article. The author suffers from lack of statistical competence, like so many others. Chen's principal mistake is to repeat the nonsense about the reluctant Republic voter missed by the pollsters, a fiction defeated by his own data.

Chen is unclear when he says, "The analysts' interpretations failed to satisfy many in the room." It was clear that "Mitofksy's interpretations failed to satisfy many in the room,"

When statistician Charles Pan said, "Statisticians need to do a better job of explaining models. We have to put data in a way people can use," he was criticizing Mitofsky's failure to engage Freeman's critique or to properly present a meaningful analysis. That's what drew applause from several audience members.

[papau]

39. Well said :-)

:-)

[Febble]

41. I really must insist

that non-response bias is far from fiction. Google it, there are books and books and articles and articles about it. NOT in the fiction section of any library. And it is not, in fact, defeated by the data - what we know, publicly, of the data, actually supports it.

It may not fully account for the exit poll discrepancy. It may conceivably not account for it at all. But there is absolutely no a priori reason to suppose it couldn't, and some evidence to suggest it did.

[papau]

42. Again - no one says the"shy" voter does not exist or is not more one party

than another.

But that is taken into account when the original assumptions are chosen (implicitly in this case).

There has to be NEW SHYNESS - more shy than in previous elections - before it even makes partial sense as an explanation for why the weightings were wrong.

[Febble]

43. I think you put your finger on it.

And shyness is, inherently, a difficult phenomenon to study. But remember, there was not only new shyness, but new voters. Or at least voters who vote rarely. Rare voters may have different characteristics than regular voters.

Or it could be that voters who changed their choice from 2000 were shyer than those who did not.

I don't know. But I don't think it's inherently implausible, just inherently unmeasurable. All we have are proxies for it.