MITOFSKY EXIT POLL CAVEAT: "FINAL PERCENTAGES MAY SHIFT SLIGHTLY"

cold dead hands.

em effer

TIA says: “The ‘preliminary’ 2004 National Exit Poll is a randomly selected 13,047 sample with a 1.0% Margin of Error.”

The preliminary 2004 National Exit Poll is a prediction of election results based on weighted, non-randomly selected information collected from 250 precincts around the country. To have a truly random sample, Mitofsky would need an interviewer in every precinct in the country.

A truly random sample of 13,047 would need no weighting to be accurate. TIA please react to Professor Howard Christensen’s statement. I have posted it several times, but you have not responded. Mr. Christensen, chairman of the Dept of Statistics at Brigham Young University and creator of the 20004 Utah College Exit Poll (acknowledged as a very accurate poll by Dr. Freeman) says the following:
“A simple random sample is one selected such that every possible sample of the same size has the same probability. To implement it requires a ‘list’ of the elements in the population; in the case of an exit poll, it would require a list of all of the voters who turnout. Since this list doesn't exist, a simple random sample cannot be conducted in an exit poll setting.”

As for the Margin of Error, Mitofsky has had to adjust has preliminary data much more than 1% in elections for many years. He might mean that the final results have a 1% MOE. I’ll ask them at the NEP web-site. Sometimes they do answer questions.

If I remember right, Christensen was talking about the two tier calculation of MOE in the Utah Exit Polls. I can't really check because you didn't post a link.

Time to get a new one.

Alternatively, instead of sniping from the sidelines you could just say what you think. Was the election fair and balanced? Is the exit poll "analysis" just tin-hat conspiracy? Is W just misunderstood? Watcha think?

It's printed in black and white in the WP exit poll summary.

Let's deal with what we know:

1) Mitofsky said the MOE is 1.0% in the notes to the "preliminary" exit poll of 13047.

2) Freeman and 10 Math/Stat Phd's agree.

3) The theoretical formula is: MOE =1/sqrt(13047)=0.86%

4) After including a 20% "cluster" factor, the MOE = 1.03%.

5) Professional pollsters agree that the historical record of exit polls is proof that they are much more accurate than standard polls.

6) Ballot spoilage occurs in every election. A large percentage of the spoilage occurs in minority precincts.

7) Ballot spoilage is a significant component of the average 2-3% discrepancy between the exit poll and vote count.

8) The discrepancy most heavily impacts Democrats, who always obtain fewer vote tallies than the exit polls indicated.

"As for the Margin of Error, Mitofsky has had to adjust has preliminary data much more than 1% in elections for many years. He might mean that the final results have a 1% MOE. I’ll ask them at the NEP web-site. Sometimes they do answer questions".

Yes, it's true that Mitofsky adjusts the preliminary data by more than 1% - but not because the exit poll is wrong. No, it's because the vote count is inherently wrong .

The vote count has NEVER reflected the intent of ALL the voters. Do you not accept the fact that an estimated 2% minimum ballot spoilage has occurred in EVERY election for as long as records have been kept?

So yes, Mitofsky weights the preliminary exit polls.
Unfortunately, he does so to match a vote count which does not reflect the intent of all the voters who went to the polls, who did not realize at the time they were exit-polled that their votes would not be counted since the ballots were spoiled, due to either faulty machines or faulty humans - or both.

The random selection applies only to the "in precinct" sample. This data must be weighted to replicate a national sample not to match the vote. The weighting is based upon the design of the poll. A later final weighting adjusts the exit polls results to the actual vote.

The sample size does effect the MOE, but it is only a part of what determines the margin. The formula that TIA uses only applies to completely random samples. Exit polls don't conform to such an analysis.

Ballot spoilage is a given. It happens in every election. It is part of the actual vote and it is part of the exit poll design. It accounts for no discrepancy.

Exit poll designs are primarily developed from prior vote results along with some other factors. To say that the exit polls are more accurate than the vote is not logical. Exit polls are a reflection of the vote.

Unfortunately, voter intentions are not measurable with exit polls.

"Ballot spoilage is a given. It happens in every election. It is part of the actual vote and it is part of the exit poll design. It accounts for no discrepancy".

BALLOT SPOILAGE ACCOUNTS FOR NO DISCREPANCY?

ONCE AGAIN:
THE PRELIMINARY EXIT POLL IS A REFLECTION OF VOTER INTENT.

THE FINAL EXIT POLL IS JUST COOKING THE BOOKS TO MATCH A BOGUS VOTE COUNT WHICH DOES NOT INCLUDE SPOILED BALLOTS AND BBV VOTES LOST IN CYBERSPACE.

YOU REALLY HAVE A PROBLEM ACCEPTING THIS, DON'T YOU?

JUST LIKE THE FAMOUS ACCOUNTING EQUATION:
ASSETS = LIABILITIES + EQUITY

WE KNOW THAT:
TRUE VOTE = RECORDED VOTE COUNT + SPOILED/OTHER VOTES

THE FINAL EXIT POLL MATCHES THE BOGUS RECORDED VOTE COUNT, NOT THE TRUE VOTE.

Professor Christensen produced accurate exit polls that matched the vote in Utah. Mitofsky's preliminary exit polls were off by over 4%.
You said:

THE PRELIMINARY EXIT POLL IS A REFLECTION OF VOTER INTENT

Mitofsky gave Kerry 30.5%. The BYU poll gave Kerry 26.5%

Which poll is a reflection of the voter intent?

Please post the link to the data you are representing?

TIA has done a wonderful job of exposing the facts that the poll data provide, by utilizing the absolute and irrefutable "laws of mathematics". He is doing all the work, please respect that. Produce argument based upon fact and if a source is quoted please include it.

http://exitpoll.byu.edu/ The exact quote is from an e-mail to me form Mr. Christensen.

I assure you, if the Ohio poll was the one that showed a Kerry win, and the NEP poll showed a Bush win, TIA would be claiming right now that the Ohio poll was the accurate poll.

It's useless to argue with TIA on this issue, though useful for others to see. I'm convinced TIA feels he has no choice at this point...if he admits his analysis has a flaw in its basic assumptions, he loses all the perceived popularity he has gained here at DU, and feels like all his hard work over the last several months was for nothing. He therefore CANNOT admit he made a mistake, from his perspective.

Exit poll analysis: astronomical odds against Bush win

Dr. Steven Freeman: Professor, Center for Organizational Dynamics, Univ. of Penn.; Karel Steuer Chair for entrepreneurship, Univ. de San Andreas, Buenos Aires; Professor of Management, Central Amer. Inst. of Business Administration (INCAE),
http://www.appliedresearch.us/sf/epdiscrep.htm
"The Unexplained Exit Poll Discrepancy," and "Hypotheses for Explaining the Exit Poll-Official Count Discrepancy in the 2004 US Presidential Election"
http://www.inthesetimes.com/site/main/article/1970

Dr. Ron Baiman: Economist/Statistician - senior research specialist, Institute of Government and Public Affairs at the University of Illinois at Chicago; teaches at the University of Chicago.
http://www.freepress.org/departments/display/19/2004/997
Baiman: "I conclude that, based on the best exit sample data currently available, neither the national popular vote, or many of the certified state election results, are credible and should not be regarded as a true reflection of the intent of national electorate, or of many state voters, until a complete and thorough investigation…."

Dr. Webb Mealy: http://www.selftest.net/redshift.htm (Bush vote skewed to the Electoral Votes that were needed to win.)

Nine Ph.D's from leading universities call for investigation of 2004 Election:
http://uscountvotes.org/ucvAnalysis/US/USCountVotes_Re_Mitofsky-Edison.pdf

Josh Mitteldorf, Ph.D. - Temple University Statistics Department
Steven F. Freeman, PhD - Center for Organizational Dynamics, University of Pennsylvania
Brian Joiner, PhD - Prof. of Statistics and Director of Statistical Consulting (ret), University of Wisconsin
Frank Stenger, PhD in mathematics - School of Computing, University of Utah
Richard G. Sheehan, PhD - Department of Finance, University of Notre Dame
Elizabeth Liddle, MA - (UK) PhD candidate at the University of Nottingham
Paul F. Velleman, Ph.D. - Department of Statistical Sciences, Cornell University
Victoria Lovegren, Ph.D. - Department of Mathematics, Case Western Reserve University
Campbell B. Read, Ph.D., Professor Emeritus, Department of Statistical Science, Southern Methodist University
Kathy Dopp, MS in mathematics - USCountVotes, President
Also Peer Reviewed by USCountVotes’ core group of statisticians and independent reviewers.

And by corroborating evidence of electronic election fraud and racist vote suppression:

Florida: 130,000 to 230,000 phantom votes for Bush--paper vs. electronic voting—calls for investigation:
http://ucdata.berkeley.edu
Report issued by Dr. Michael Haut, & UC Berkeley Quantitative Methods Research Team; Haut is a nationally-known expert on statistical methods and member of the National Academy of Sciences and the UC Berkeley Survey Research Center
Press release: http://www.commondreams.org/news2004/1118-14.htm

Democratic Underground (ignatzmouse):
(North Carolina: absentee ballot vs. electronic, inexplicable 9% edge to Bush in electronic:)
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x45003
(also at:) http://www.dailykos.com/story/2004/11/12/233831/06

TV networks alteration of the Exit Polls to fit the "official tally" (& Zogby prediction of Kerry win):
http://www.exitpollz.org/

Johns Hopkins report on insecurity of electronic voting:
http://www.scoop.co.nz/mason/stories/HL0307/S00196.htm#5

Easy demo of how insecure voting machines are:
http://www.chuckherrin.com/hackthevote.htm

"Myth Breakers: Facts About Electronic Elections" (2nd edition): www.votersunite.org

Ohio vote suppression: http://www.bpac.info

Documentation of widespread machine fraud and dirty tricks in over 20 states: http://www.flcv.com/ussumall.html

57,000 machine malfunction/vote suppression complaints to Congress:
http://www.votersunite.org/article.asp?id=3961

"Kerry won – just count the votes at the back of the bus!" – by Greg Palast
http://www.tompaine.com/articles/kerry_won.php

Jonathan Simon:
http://www.scoop.co.nz/mason/stories/HL0411/S00142.htm

--------

You wrote:

"It's useless to argue with TIA on this issue, though useful for others to see."--Mistwell

So, at least one your purposes here is to tarnish TIA? That's certainly useful to know.

Why don't you complain to all those Ph.D's in statistics, mathematics, quantitative research and public policy, above?

Only TIA gives you a forum.

All prior early exit polls (ALL) had a bias that favored Democrats. If it happens every single time, then there is no surprise it happened this time.

http://www.washingtonmonthly.com/archives/individual/2004_11/005178.php

Year / Exit Poll / Results / Dem Lead / Dem Actual
1988 / Dukakis: 50.3% Bush: 49.7% / +0.6% / -7.7%
1992 / Clinton: 46% Bush: 33.2% / +12.8% / +5.6%
1996 / Clinton: 52.2% Dole: 37.5% / +14.7% / +8.5%
2000 / Gore: 48.5% Bush: 46.2% / +2.3% / +0.5%

“As you can see, the raw exit poll results always overstate the Democratic vote, sometimes by as much as eight percentage points. So the fact that the raw results this year overstated Kerry's actual vote tally is hardly cause for alarm.”

The leading expert on the subject is Warren Mitofsky, the father of exit polls. Mitofsky is also a lifelong liberal and “apparently holds no brief for Bush.”

http://www.russbaker.com/TomPaine_com%20-%20Election%202004%20Stolen%20Or%20Lost.htm

Mitofsky does not believe the early raw exit poll data (the stuff showing a Kerry win) indicates fraud. Indeed, he thinks the early raw data is inaccurate, and not useful until it is weighted with his formula. Once the formula was applied after all data came in, the election returned to being within the margin of error.

http://mayflowerhill.blogspot.com/2004/11/mayflower-hill-exclusive-warren.html

He has never, EVER claimed that the margin of error in his early exit polls (the ones that TIA likes to quote constantly) are set for detecting election fraud or even set for the vote itself. The MOE is set ONLY for the Demographic data. It is a total myth that the margin or error is accurate, or ever meant to be accurate, for the vote itself.

Lots of studies who that the early exit polls are seriously flawed:

http://www.exit-poll.net/election-night/EvaluationJan192005.pdf

and see also

http://www.scoop.co.nz/mason/stories/WO0411/S00140.htm

The subject of why the exit polls favor Democrats was studied, and a very reasonable explanation offered even prior to this election:

http://www.duke.edu/~mms16/non_response2000.pdf

And, your quotes of opinion writers doesn't mean much. Greg Palast is a fine writer and I agree with a lot of what he says, but he is no exit poll expert.

When it comes down to it, the actual exit poll experts all agree with Mitofsky that the way we conduct exit polls in the US, they just are not accurate enough to predict the outcome of an election. They are good for Demographics data only. And at heart TIA knows this...he just is too entrenched to back down at this point.

The point of my being on this board has very little to do with TIA. I'm just one of many Democrats who thinks we should not be wasting time on the theoretical claim that exit polls can prove fraud, and should instead be focusing on ACTUAL proof of fraud, and on the next election. That...and I love the science forum here, and the news I can pick up, and the spirited debate we can have on so many topics. I love DU...the TIA thing really is just a side thing.

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

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

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

I sent a transcript hours ago.

and passed it on. But I'm not drinking the kool-aid sorry. Mathematics is an EXACT science. TIA's analysis is irrefutable. Deal with it.

...but creating a sample that reflects the whole voter population is not. TIA has never proved that the sample NEP took was random and representative. We've been over this hundreds of times and TIA is incapable of even acknowledging that the sample needs to be representative. This reflects a disturbing lack of understanding of the nature of statistical analysis. The math is the easy part, its getting a representative sample that is difficult.

NEP asserts that their sample is random. That is what they sell. The proof is in the historical accuracy and in it's acceptance. Many news organizations have abandoned their own attempts at exit-polls in order to subscribe and most academics begrudgingly concede that Mitofsky is able to generate a random sample (while often bemoaning the "black-arts" that get him there).

As you must know, NEP recently published a report on precisely whether the poll results reflected a random population. They concluded that they did (or exactly the same thing, by arguing that any inaccuracy was due to in precinct error). That report has come under criticism but not, as far as I am aware, for the overall randomness of the poll but for the veracity of WPE.

It is not for TIA to "prove" that NEP polled a random population. That is for NEP to do and they have defended that. If you don't buy it, you have a bitch with NEP and those who do buy it. Not "believing" in exit polls in such a fundamental way is not common.

Alternatively, if you buy into NEP, TIA's calculations are legit.

The bottom line is that it is all data and all equally subject to test.

since you seem afraid to. This is the *link* jkd will not reveal, but has instead PM'ed to me....
----------------------------------------------------------------------

Reading message in your inbox
This is Professor Christensen's E-mail exchange with jkd
From: jkd
Date: Feb 27th 2005

Thanks for your kind words and your interest in our project. I will address your questions in the following message below.

Professor Christensen,

BYU students in the 2004 election polled my precinct in Heber City. They were a credit to the university. Are they paid or is this effort purely volunteer?


The students are purely voluntary labor with reimbursement for costs of transportation and a meal during the day. But that is a small payment for working throughout the day at a polling place. Often, cooperating instructors will provide extra-credit for their work, recognizing that there is some educational benefit to the experience.


I have several questions about exit polls. Are they simply a statistical function of the number of voters polled? If not, then what other factors are employed in determining the margin of error?


In a complex sampling situation such as the exit poll, the stages of sampling--the number of selected counties out of a congressional district, and the number of polling places selected--influence the margin of error along with the number of voters selected at each polling place.

What was the MOE for the Utah presidential election? What is the difference between a multi-staged stratified sample and a simple random sample?

Our margin of error with a 99% confidence level, was 3.51%. In typical reporting for ordinary opinion polling issues, attitudes etc., the margins of error reported in the media have a confidence level of 95%. To put ours on that comparative basis gives a MOE of 2.67%. The margin of error for other races was larger or smaller depending on the affects of the design and so forth.

A simple random sample is one selected such that every possible sample of the same size has the same probability. To implement it requires a "list" of the elements in the population; in the case of an exit poll, it would require a list of all of the voters who turnout. Since this list doesn't exist, a simple random sample cannot be conducted in an exit poll setting. We implement a stratified, two or three stage sample (the number of stages depends on which counties are determined to be a part of the sample of counties with certainty) with a systematic random sample of voters at the polling place. The first voter is selected according to chance which is different from voting place to voting place, and then every kth voter is selected from that point on. The value of k depends on the expected turnout at the voting place and differs, again, from polling place to polling place.


In your opinion were there problems with the NEP/Mitofsky presidential polls? Did their deviations from the actual vote suggest problems with their poll or with the election itself?


Mitofsky and his colleagues have prepared a report and acknowledge the "within precinct" error was larger than it should have been statistically, indicating there were possible selection biases. You can access his report at: www.exit-poll.net

The problems have to be with the polling process, not the election itself because the polling process is supposed to pick up the features of the election itself.

Your poll was very accurate. I thank you and all the dedicated students who produced it.





--
Howard B. Christensen, Chair
Department of Statistics
Brigham Young University
230A TMCB
Provo, UT 84602

Phone: (801) 422-7050
Email: howardc@byu.edu


Also here is a link to the BYU web-site

http://exitpoll.byu.edu/

I provided the source and the transcript to anyone who requested it.

JKD,
First a question, does the BYU exit poll use actual election results to weight/correct their data? If so, your points are not valid, if not, then you make potentially valid points.

The 4% difference in exit-polling between the BYU and Mitofsky polls is interesting. What is the published MOEs (95% confidence level is standard) for both? If the sum of the MOEs is less than 4% and two different samples produce such different results (4% difference), it does demonstrate your point that calculating the MOE may be misleading, since it raises the question how could two 'random' samples result in statistically conflicting (non-overlapping) data. We can even calculate the likelihood of this happening. But more importantly it suggests that there are important methodological differences between the exit polls that must be accounted for. These differences may help us understand the Mitofsky poll better, which then will help us eliminate any sampling errors that might have occured. Once we eliminate the sampling errors, we will have a better understanding what happenned on Nov 2 and possibly be able to evaluate the extent of possible fraud. Do you have any idea what the differences in method between BYU and Mitofsky are?

The exit poll stands by itself and is completed before any results are announced. That is my understanding. The BYU MOE is 2.67% at the 95% confidence level. The poll actually missed the election results by one half of one percent. I have not seen any design model that BYU uses. I'm sorry, the only information I have has been obtained from their web-site.

I assume 5% is Mitofsky’s MOE for Utah. http://www.washingtonpost.com/wp-srv/politics/elections/2004/graphics/exitPolls_utah.html The sum of NEP + BYU margins of error is well over 4.0 % at 7.67 percent. It appears that they are both within the margins and do overlap even though they differ considerably. Sorry. Here’s a little more information about the Utah Colleges Exit Poll. http://newsnet.byu.edu/story.cfm/52808.

First, it is a Utah poll just as the LA Times poll is (mostly) a California poll.

Second, the precinct selection criteria is completely different. BYU does a multi-tier selection which starts out with random precincts, then a comparison to demographics, then a tuned precinct selection, etc. That is why the MOE is higher and the MOE calculation is different.

NEP dabbles in the black arts instead... they use a variation on the old pol's "key precincts" idea - precincts with a known record. Like a computer program with human "hints", the performance can be much better.

Third, the BYU is new and the record is a little spotty. Take a look at the Utah congressional races for 2004.

Unfortunately, BYU is a "red-herring"... appropriate for one of the "reddest" states.

You wrote:

"NEP dabbles in the black arts instead... they use a variation on the old pol's "key precincts" idea - precincts with a known record. Like a computer program with human "hints", the performance can be much better."

You mean that they use a historical model to determine precincts that are best representative of the sample? If the precinct worked well to model results in the last few elections then it is selected this year?? How do they control for how well this is working in the current election? Do they just multiply the actual election data from the selected precincts by some geographic/demographic weighting factor and show that it closely matches the election results overall? It seems that they would have to do this.

OK, if this is right, then when the NEP multiplies their exit polling data against the same model, it doesn't match the election. This means that they should be able to identify specific precincts which most contributed to their "error". These should be checked for fraud by a careful check of the voting equipment and recorded votes. Also, to check the possibility of polling error, they should compare the data from the worst precincts with the data from the best precincts to see if they can determine any source of error.

... a lot of that is "secret sauce", I think.

I'm not so sure that they chose on "results" so much as on the demographic mix they generate.

I wish I could remember who, but that is exactly what he said--that they simply shifted the party percentages.

to pull off given the differing state laws, machine types etc. Fudging around with more complicated approaches to election fraud would have been silly, since there are so many more ways to screw it up and fail. They only needed to *win* by 2%-3% so they kept it simple. Obviously it wasn't simple enough for Blackwell though...

...registrations (Dems 57% vs. Repubs 41%), and by the enthusiasm of the grass roots campaign to oust the Bush Cartel. Almost all Nader voters also jumped to Kerry (80+%). Gore 2000 repeat voters were also hot to oust the Cartel as well, due to anger over 2000 election fraud, and hideous events since then (100,000+ mass slaughter of Iraqis, mostly by our bombs, mostly innocent civilians; and widespread torture of prisoners). The Gore 2000 repeat voters were the ones who got family, friends, co-workers to register and vote for the first time--all to oust the Bush Cartel.

And Bush's idiot performance in the debates. And so much else.

So they had to use every fraud plan that had put into place in order to "win" (--save one, the "terrorist alert" they had so carefully prepped in the "news" prior to the election--did you ever wonder, as I did, why Dick Cheney flew to Hawaii, of all places, two days before the election? I think it was part of the "terrorist alert" plan that wasn't used--VP's plane in trouble over the Pacific, something like that = "terrorist alert" shutdown of California freeways on election day. --???).

Anyway, what I see is that the "old pols" didn't really trust the "new techies" and their promises--and so they had many backup plans, most of which they had to use, given the blowout win for Kerry that was developing (and that in fact occurred).

enter the "then they fight you " phase, great work. NGU

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