I learned something about exit polls today.

[TruthIsAll]

The Exit Poll is not really a poll.
No. It's not.

You see, it's weighted to match the actual vote.
If that's the case, who needs it?
Why bother?

Why not just use the vote counts from the get-go?
The votes are extremely accurate, aren't they?
There is no reason to assume otherwise, is there?

After all, the actual vote is a 100% sample-size.
That's a ZERO margin of error.
Very accurate.

No wonder they weight the exit poll to agree with the actual vote count.
Otherwise, you might conclude they are lousy pollsters.

It all makes a great deal of sense to me now.

Early exit polls are unweighted.
To analyze them is a waste of time.

The exit polls must be weighted.
After all, why analyze a poll when you already know the vote?

But why even bother to match the exit poll to the actual vote?
When you already know the actual vote?

The Republicans are right.
Exit polls are a total waste of time.

Let's get rid of exit polls.
Permanently.

[MissBrooks]

1. Translation?

Was this translated from the original Latin?

[New Earth]

2. looks like plain English to me

:hi:

[TruthIsAll]

24. Now I realize that the NEP randomly-selected sample of 13,047

is a worthless sample.

That 1.0% Margin of Error is misleading.
It really is much higher than that.
Mitofsky is just kidding.
He rounded off the true MOE- by an order of magnitude.

It's a VERY ACCURATE sample.
But it's not accurate enough.

It's a weighted sample.
But it's not weighted enough.

It has lots of data characteristics.
But they all signify nothing.

54% of women voted for Kerry.
But they really didn't.

The mix was 38% Dems, 36% Repubs, 26% Indies.
But it wasn't.

It's a random sample.
But not random enough.

Let's sample this Bizarro Universe instead.

[TruthIsAll]

166. IF ITS 38 DEMS /35 REPUBS, ITS 303 MILLION TO 1.

NEP 11/04
13,047 respondents,
randomly selected,
1.0% MOE


               HORIZONTAL		WEIGHTED		
PARTY ID	
     MIX	Bush	Kerry	Nader	Bush	Kerry	Nader
Dem 	38%	9%	90%	1%	3.4%	34.2%	0.4%
Rep 	35%	92%	7%	0%	32.2%	2.5%	0.0%
Ind	27%	45%	52%	2%	12.2%	14.0%	0.5%
	100%				47.77%	50.69%	0.92%
				
Probability: Poll(47.77%) to vote (50.73%): 0.00000000329447

         ********* 1 in 303,538,508 ************

Prob = 1 - NORMDIST (.5073,.4777, .01/1.96, TRUE)

[Duncan]

69. Its sarcasm. Thanks TIA. (n/t)

[tngledwebb]

88. 500 posts in just over a month, anodyne b-boomer user name,

humorless sarcasm. Are you sure you're at the right polling station, our MsBrooks?

[Karenca]

169. Simple!;

:boring:

TRANSLATION: B*SH STOLE THE ELECTION AGAIN.

But I think YOU already know that.

[Goldeneye]

3. I've been thinking the same thing.

It really just doesn't make sense. Why even put together polls before hand if there's a good chance they're going to be wrong. Might as well wait for the results to come in.

After 2000 they were upset that the exit polls were wrong...not the demographics, but the exit polls. Isn't that why they set up the new polling pool? So they could get better data? And yet they were wrong again...hmmmm.

[qwghlmian]

4. The exit polls in the US exist in order to

analyze the election voting patterns - what income levels voted for whom, how did the hispanic moms vote, how did NASCAR dads vote, what percentage of Jewish vote went for Kerry, etc etc etc.

In order to do that, the raw voter data that is gathered in just a handful of precincts (I believe it's something like 50 per state) is weighted in order to match the actual election results, thus making sure that all that various voter preference data that I described above is weighted correctly as well.

If you wanted to have the exit polls predict the results or verify the results, you would conduct them very differently. You would canvass a LOT more precincts, and you would pick the precincts you're looking at a lot more carefully. You would also make sure to bring down the non-response rate to a lot less than the 50% that it is today. The weighting to actual election results corrects a lot of skews and biases in the data that would otherwise screw it up completely and would require a lot more expensive methods to rectify.

[Name removed]

8. Deleted message

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

[BeFree]

13. You, again, euler?

You are spreading your ignorance. Anyone who believes exit polling has to be weighted to be accurate exposes their ignorance about exit polling.

[euler]

36. It's not a belief

See post #29

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=288785&mesg_id=289017&page=

[Karenca]

23. Euler, have you no

sense of irony or humor?

Or is it that you just cannot stop spreading that old propaganda machine?

[me b zola]

165. LOL

GOP=grand old propaganda:beer:

[Fly by night]

14. I disagree. Exit polls exist to identify the winner sooner ...

... than waiting for the final votes to be counted, reported and certified -- processes that may take weeks. The networks don't pay for exit polls to determine the voting patterns of demographic sub-groups. That is a secondary benefit, but they pay for exit polls in order to determine (with heretofore a great deal of accuracy) who will win each state on election night. The other analyses are important but they are secondary to the purpose of the exit polls -- to determine (again, pre-2000, with an astonishing degree of accuracy) which candidate won the state or the national race on election night.

Speaking of which, if the primary purpose of exit polls was to determine the voting patterns of demographic and political sub-groups, there is no way to adjust those exit poll results with information from the actual votes cast because OUR VOTES ARE CAST ANONYMOUSLY, FREE FROM ANY DEMOGRAPHIC OR OTHER IDENTIFYING INFORMATION ABOUT THE VOTERS. Our ballots do not contain any demographic or political sub-group information -- they contain only our votes. The only place we learn anything about the voting patterns of demographic or political sub-groups is through exit polls (or some other pre- or post-election poll) where the characterIstics of the voters can be matched with what they say are their actual or intended votes. The actual votes cast (and the information reported up the chain to the state election office) contain no information that would allow any adjustment whatsoever in refining the demographic/political sub-group analyses.

Exit polls have always existed to allow us to know who won the political contest sooner than we would otherwise know it by having to wait for the votes to be counted, reported and certified. Heretofore, skilled exit pollers have succeeded in identifying the right mix of voting precincts that had to be sampled (and in what proportion) to have an accurate estimate of who won. (And those skills were present again in 2004. They just were not accompanied by the honesty and integrity necessary to report those results to the public in a timely and un-"massaged" manner.)

Face it, folks, exit polls worked exceptionally well back when the votes were counted "as is" and not "weighted" by whoever contols the touch-screen machines or the tabulators -- and whoever controls the controllers -- who now "adjust" or "weigh" the consent of the governed to reflect the "evil massa's" wishes. Anyone who says otherwise has something to hide -- either their ignorance of the primary purposes of exit polls or their true intentions in denigrating the importance of exit polls as a measure of the honesty of the "reported" vote count.

[qwghlmian]

19. Correction

on both counts.

1. The "astonishing accuracy" of exit polls:

http://www.washingtonpost.com/wp-dyn/articles/A64906-2004Nov20.html

"The networks' 1992 national exit poll overstated Democrat Bill Clinton's advantage by 2.5 percentage points, about the same as the Kerry skew."

"I learned early in my Washington Post career that exit polls were useful but imperfect mirrors of the electorate. On election night in 1988, we relied on the ABC News exit poll to characterize how demographic subgroups and political constituencies had voted. One problem: The exit poll found the race to be a dead heat, even though Democrat Michael Dukakis lost the popular vote by seven percentage points to Dubya's father."

2. The purpose of the design of US exit polls:

http://election04.ssrc.org/research/InterimReport122204.pdf

"Rather, exit polls as currently designed and administered in the United States are not suitable for use as point estimators for the share of votes that go to different candidates."

...

"Nevertheless, some analysts inappropriately attempt to use current exit poll results to investigate whether the results in a locale (state or country) are accurate or whether fraud might be involved in an election.10 A certain form of exit poll could be used for this purpose, but again the designs would have to be different. To validate results in specific precincts or from particular machines, the designs would have to incorporate larger numbers of interviews with voters leaving the polls for precision."

[RaulVB]

21. Parroting talking points without adding your own rationale? (n/t)

[Fly by night]

26. I stand by my original post

1) Exit polls exist to predict the winner earlier than would be possible by waiting for the votes to be counted, reported and certified. That is why they have always existed. That is why the networks pay big bucks for the exit polls -- to tell Americans on election night WHO WON with heretofore an exceptional degree of accuracy. (Even Mitofsky has stated that, prior to 2000, his own polling expertise and experience was sufficient for him to be accurate in identifying the winner 99.5% of the time.) They do not exist primarily to entertain the American public with minutia about how left-handed right-wingers (and the plethora of other population sub-groups) voted.

2) If exit polls exist to answer the minutia, it is impossible to adjust them at all with information from the "reported" votes because those votes are cast without any identifying information whatsoever (other than precinct) that is over and above the votes themselves. You can make no adjustments to the minutia-related information with anything available from the "reported" votes. They are cast anonymously and without any identifying information.

So try as you might, you still win no cigar. You're huffing and puffing, but this methodological house is made of bricks. Why else is Mitofsky now saying that he will release his final report to the corporate media but neither he nor they have decided if we poor gullible citizens should ever get a look at the analyses, much less the raw exit poll data tapes themselves.

So paternalistic, so unscientific, so dishonest, so co-conspiratorial, so evasive, so unethical --
so Republican.

[RaulVB]

27. Yes...

"So paternalistic, so unscientific, so dishonest, so co-conspiratorial, so evasive, so unethical --
so Republican."

- So true...

[qwghlmian]

30. Argument by assertion. Doesn't fly. n/t

[RaulVB]

31. False statements don't "fly" either (n/t)

[euler]

42. What about true statements and historical records - do they fly ?

See post #29

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=288785&mesg_id=289017&page=

[euler]

38. What about argument by facts ?

See post #29

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=288785&mesg_id=289017&page=

[Fly by night]

47. I'm sorry. What point are you trying to make?

Taking my second point only (which is related to my first but it may be easier for you to focus on just one for the moment), if the primary purpose of exit polls (as you tag-teaming energizer bunnies keep saying over and over again) is to assess the voting patterns of sub-groups, then the justification of any "adjustments" or "weighting" would be to improve the predictability of the exit polls to describe the voting patterns of those sub-groups. But absent any more data tying votes cast with demographic or other sub-group identifying information, you cannot improve your stated primary purpose of exit polls by drawing on information from the actual votes cast. There is no sub-group information captured with the votes cast -- zero, zilch, nada. Your argument is not "reality-based" in an election system where votes are cast anonymously, without identifying information. That's not an assertion, that's a fact.

Welcome to reality. Now go home.

[qwghlmian]

49. That's a misconception -

the theory is, as you adjust the raw data for the candidates' performance in the precincts that you canvassed for the exit poll to match the actual results, for each precinct that you polled, you simultaneously and proportionally adjust the weights of the demographics for those precincts in the final exit poll results. If you do it right, the biases and skews that are in the raw exit poll data because of the clustering effects would be corrected. No demographic data for the actual election results is required.

This is really the basics of exit polling. You are dismissing this without any backing from any source whatsoever. Please show me some source that supports your assertions above. You are displaying a basic lack of understanding of how polls work.

[RaulVB]

52. You are falsely representing exit polls on purpose (n/t)

[googly]

168. You should refrain from denegrating a poster unless you have

a valid coroborrating narration to substantiate your attacks.

[Fly by night]

60. Correcting for biases and skews without relevant information ...

to make those corrections is not possible. Just like getting you to see my point seems to be impossible. But I'll try once again.

No researcher can adjust for biases or skews in a sample survey without having some basis for knowing that biases or skews occurred -- in essence, that the sample surveyed was not representative of the whole. The only way to do that would be to have an independent, and presumedly less biased/skewed, information source. You continue to state that this can be done with the final vote tallies in a precinct. But those final vote tallies tell you NOTHING about the final demographic makeup of the voters in that precinct who showed up to vote or how representative the voting patterns of the sub-group representatives interviewed in the exit poll were of the voting patterns of the entire voting sub-group because you have no demographic information tied to the votes themselves. Thus, you have no information tied to the vote totals available to adjust for any biases or skews within the sub-groups surveyed in the exit polls -- it simply does not exist.

If, however, your primary purpose is to estimate the winner of the contest (once again, the primary reason why exit polls are conducted), you do have information available to adjust the representativeness of each precinct chosen to participate in the exit poll. And that information relates to the differential turnout of all voters within each precinct. Thus, the representativeness of each precinct's exit poll interviewees can be adjusted (or weighted) by the overall voter turnout in that precinct. That is the principal way in which raw exit poll data are adjusted. No direct adjustment is possible for the voting patterns of demographic subgroups because no direct information exists to make that adjustment. It can only be implied by the differential turnout by precinct, with the overall demographic makeup of the precinct being used as a surrogate for the voting patterns of the subgroup members themselves. And that adjustment is dicey because it steers perilously close to the "ecological fallacy", something that those of us trained in the social sciences are well aware of.

The basis for my remarks is graduate training at (and/or degrees from) Vanderbilt, Texas, Stanford and Berkeley; and three decades of experience in field research. I won't waste my time looking up citations for you because citations are not necessary when my statements are logical, experience-based, reality-based and accurate as is.

Besides, I learned two things a long time ago: You can never win an argument with a "true believer". And you can always tell a "true believer" -- you just can't tell them much. Fortunately, I have also learned here at DU that while I may not be able to penetrate the ideological fog of a "true believer", I can put them to sleep. Snooze away.

[anaxarchos]

61. True believers...

...and grad students ;-)

[qwghlmian]

66. Your training is obviously not in polling -

otherwise you would not be stating something that is contrary to the exit poll methods that have always been used in the US. I suggest you do some reading that was done by polling professionals

http://www.mysterypollster.com/main/2004/11/the_difference_.html

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

[euler]

80. Conversations here are lively,...

...if somewhat inane, until someone posts something that actually proves them wrong. That shuts people up fast.

[Fly by night]

92. Some folks don't stop, even when their illogic is pointed out repeatedly

You and qw are our best (and most recent) examples. Neither of you have addressed my points (or really anyone else's) directly, preferring to send us off to other sources, even though I have repeated my points three times so that I can be sure that you, like fourth graders, understand the points. Just address my points. They are straightforward, and can be addressed in a straightforward manner, if that is your intent. I don't have time to stay on DU all day -- my silence is because I have another life, including one that is focused on overthrowing the Bush coup. And while reading and commenting on DU threads is certainly a necessary tool in that process, it is not a sufficient one. And bantering with you two is the least useful part of this process, as neither of you seem capable of taking the cotton out of your ears and putting it in your mouths, even for a moment.

As far as my experience with polling, I have been involved in designing, conducting and supervising the conduct of polling at the state and national levels since 1975, including multi-state polling conducted while I was employed by the Gallup Organization, CDC and the National Institutes of Health. Are either of you even that old?

Come to think of it, I wish I weren't. I could use a dose of the "arrogance of youth" every now and then. Nah, on second thought, the side effects (hubris) aren't worth the fleeting pleasure of believing again that I know everything. Just watching the two of you in the DU playground reminds me that age and experience do have their benefits.

Now it's nap time for you too. Bye bye.

[mgr]

120. I could do with the arrogance and hubris myself

Let me respond or clarify. I am not sure who you addressed this post to, Euler or TIA.

For those who state that exit polling in the US is not appropriate for characterizing the overall election outcome, there is no need to explain how the reweighing or adjustments are made at the precinct or sub-national level. All that needs noting is that it happened. If one inspected the exit polling pdf for four o'clock at Scoop, one will note the words "weighted". Since no election results were reported at four o'clock, we can know that they are not being weighted against voting results, but some other known or presumed variables at the precinct or state level.

I think it is for those that take the position that the US exit poll for 2004 accurately characterized voter intent better than the actual outcome, to explain what the weighting was that did not taint the exit poll's purpose is needed.

As I have pointed out before, I think that what we have are precincts that have been sampled repeatedly in past elections, demographic patterns and changes are known through the census, registration numbers known, and expertise in addressing these matters. One may, with the appropriate experience, anticipate what variables require adjusting based upon past patterns, changes in voter base, necessary to characterize the precinct; but to extrapolate these adjustments to the overall election may be fallacious.

What we also have is a subsample of the overall sampled population used to estimate the national election outcome. How this subsample is selected is unclear, and sampling by phone 500 absentee/early voters seems too small for accuracy.

The weight of skepticism largely supports the naysayers.

Mike

[Fly by night]

121. Using other information to adjust our views of the "reported" vote

My comments regarding hubris were aimed at euler, not TIA. And I had written a long response to your post, which unfortunately was just eaten by my computer (trying to do too many things at once, I suppose.)

My guess (and that is all it can be without Mitofsky coming clean) is that any early (and even later) weighting of exit poll data that you refer to had to do primarily with the differential turnout reported within the precincts being sampled in the exit polls. That is the principal weighting procedure used to adjust exit poll data to extrapolate to the universe of voters in order to derive an estimate of the final outcome -- the main reason why exit polls are conducted.

However, your comments about using other historical information to adjust the exit polls still do not negate my continuing point that the final vote tally can do nothing -- in and of itself -- to adjust any understanding of the dynamics between socio-demographic characteristics of voters and the votes they cast. As you point out, there are other sources of information with which to do that, but most examples you cited (past elections, census data) are quite dated and are much less useful than pre-election surveys to assess the stable and shifting dynamics between voter characteristics and voter choices.

However, rather than repeating the points in my deleted message, I think it is more telling to look at the most recent Washington Post/ABC poll to assess the current attitudes of the American voter regarding Bush. If you believe that four year old census and four year old prior election data are predictive of voting patterns, then you might concede that a poll conducted two months after the election might also reflect the attitudes of voters in November. Here are some salient quotes from an article reporting that recent poll:

"Fewer than half of those interviewed -- 45 percent -- said they preferred that the country go in the direction that Bush wanted to lead it....

"Bush said in an interview last week ... that the 2004 election was a moment of accountability for the decisions he has made in Iraq, but the poll found that 58 percent disapprove of his handling of the situation to 40 percent who approve ....

"Of all presidents in the postwar era who won reelection, only Richard M. Nixon had a lower job approval rating at the start of his second term ....

"A majority of Americans express disapproval of Bush on other key measures of presidential performance. A slight majority -- 52 percent -- disapprove of the way Bush is handling the economy, and half or more also are dissatisfied with the way Bush has dealt with the budget deficit (58 percent disapprove), immigration (54 percent) and health care (51 percent).

"Overall, the public expresses more confidence in Democrats in Congress (50 percent) than in the Bush administration (37 percent) to deal with problems in the Social Security system."

Given this recent national poll, I would suggest that the sentiments of Americans regarding Bush were more accurately reflected in the exit polls released to us by our Kiwi friends than in the "reported" vote we continue to be asked to believe.

The methodological issues you raised in your post can only be addressed when Mitofsky and the corporate media release the exit poll data and methodology for scrutiny by academicians, investigators and the American people. Their continued reluctance to release that information (and the implication that this information may indeed never be released) continues to strengthen the impression shared by most of us here at DU (and a growing number of Americans) that this administration and its exit polling and corporate media lap-dogs do indeed have much to hide.

"Come out, come out, wherever you are. We'll find you out soon enough."

[anaxarchos]

125. Your point is dead on.

Even more, the lack of a "bandwagon effect" in the post-election polls is striking in and of itself. I was surprised at how little coverage this got in the U.S.

Not so internationally, however. This is Jamaican journalist, Ken Maxwell who has stalked many Caribbean elections for fraud.

http://www.jamaicaobserver.com/columns/html/20041106t200000-0500_68984_obs_a_lobotomy_for_democracy.asp

"One crucial statistic made me quite sure that the election was stolen. It is a well-recorded phenomenon that after an election result is known, more people will claim to have voted for the winner than actually did.

After this election, is a remarkable fact that only 51 per cent of the US electorate said they were happy Mr Bush had been elected. The post-election bandwagon effect is well documented.

"Response error tied to over-estimation of voting is one of the oldest and most persistent types of response error to be documented. . reports that such response errors tend to range between 12 and 16 per cent. with the error tending to be larger the closer a survey was done to the election". ( Robert H Prisuta,

A post-election Bandwagon Effect 1992 and Stanley Presser: Can Context Changes Reduce Vote Over-reporting?; Public Opinion Quarterly, Wier 1990)

In this case, and as far as I can discover, only in this case does the percentage claiming to have voted for the winner fall below the percentage actually voting for him.

The US press in its cocoon of fantasy, pretends to believe that this result is possible and accurate.
No one can - without his consent - be deprived of his rights. It says so even in Third World constitutions."

[euler]

139. So, you agree that some people lie to exit pollsters ?

[mgr]

161. That is a conjecture, not a fact

What you have is a universal statement, that after any election there is a bandwagon effect. The universal is informed by existential events, each previous election this occurred. The situation is similar to the statement "all swans are white" which held up until Austrialia swans were noted to be black.

You have to demonstrate that this election did produce a bandwagon effect, and that current polling indicates that effect. If there exist one counter example, your universal is disproven. I believe that the re election for Pete Wilson as governor of California falls into this category, that Wilson's approval ratings remained flat after the election. Bush's approval ratings appear to follow the same pattern.

This does not mean that discrepancies did not occur, but that this would not be proof, or only supporting documentation with stronger evidence, but by itself....

Mike

[anaxarchos]

164. mgr, I'm surprised at you.

First your beef is with Ken Maxwell (my citation) and only indirectly with me. You can write him and he will probably respond. Very knowledgeable guy.

Second, Maxwell cites sources. For his primary source (Prisuta), here is a direct link:

http://www.amstat.org/sections/srms/Proceedings/papers/1993_194.pdf

You can look up "bandwagon effect" in the literature and you will find that it is both generally accepted and the link above is the most commonly cited.

Third, "bandwagon effect" is particularly noted in Presidential Elections (your Pete Wilson comment is not really relevant) and is not linked (at least in what I've seen) to other races. It is so well understood that there is a term for it. It is called "post-election bounce". Here is Eric Boehlert, at salon.com (12-21-04):

http://hnn.us/roundup/entries/9175.html

"Since his 3-percentage-point win over Sen. John Kerry, Bush has experienced a complete lack of bounce in the polls. In fact, in at least one national survey, Fox News' Opinion Dynamics poll, conducted Dec. 14-15, Bush's approval rating has fallen five points in the last month, to 48 percent. In other polls, including Washington Post-ABC, NBC/Wall Street Journal, Pew Research Center, Associated Press-Ipsos, Zogby, and Gallup, Bush's already soft approval numbers have flat-lined since the election. That phenomenon stands in sharp contrast to U.S. history, when presidents voted into office for a second term, even after close elections, routinely have received robust approval ratings."

If you don't like Salon, you can look in the professional literature.

Fourth, I'm not sure what to do with your "proof" thing. I'm not sure you can "prove" evolution but we keep digging up these dinosaur bones...

Fifth, if you want to duel on existential proofs, I accept. Choose your weapons... Kierkegaard or Sartre?

[mgr]

160. Sentiments do not equal deliberate intention

I think the issue with national election results being discrepant from exit polling is a red herring. The absense of the Mitofski methodology cuts in both directions for those who question its applicability, and those that affirm it. I find in its absence, the patterns of sampling and analysis questionable for characterizing the aggregate of voters attitudes at the national level.

I agree that the weighting of sample results to precinct characteristics would be normal and logical practice, however it is a filter that undermines the precision of measurement beyond the specific precinct. Extrapolation to larger demographic units from filtered precincts incurs its own impacts to accuracy, and cannot account for micro scale differences in unmeasured precincts. When projecting from samples at an order of magnitude of 10,000 to voting outcomes in the order of magnitude of 1,000,000, the likely error could exceed 10% if you had two or three precincts wrong. It makes the national sample a joke.

A pre election survey of selected precincts, coupled with a reported election day refusal rate of 50% (particularly if the pattern is not random), give me little additional confidence. Timing is always an issue, and no matter how close to the election you assess a precinct, certain variables and attitudes may remain fluid up to the moment in the voting booth. It may be that little OBL rant did more to galvanize Bush support than we know.

An election can only capture priorities, attitudes, and values at that time. There is no independent measure if the election got it right except the election itself. You make the valid point that census and registration reports are static representations from the past, but were quiet on my third criterion which is the variable that is most discounted in these discussions, and that is the exit pollers' expertise. The poll is theirs to design, and unless you take them as unreliable, they seem to be stating that their design is not to characterize national election moments, but localized patterns. For us, as more and more voters opt out on casting votes on election day, exit polling can never accurately match an election's outcome.

The point is not that Bush is abhorrent, but somehow voters appear to have found Kerry more so at that time. Cherry picking recent polls since that event would not address this salient issue if not addressed head on, and compared with who was actually voted for. With your experience, you should recognize that how a question is put, who it is put to,and what types of open or closed responses are associated with the question can produce very discrepant patterns--look at the recent flap over how much moral values had influenced the election.

I'm from California, we have had two governors with weak public support who were re-elected. Negative campaigning appears to take its toll in this manner, by making the other selection less desirable. Not being from a battleground state, I cannot say with any relevance whether negative campaigning was conducted by both sides, but the Swifties probably did their damage.

As I have said before, the answers lie with precinct level affidavits and poll book comparison, the hard work is just starting, and this issue with exit polling is fluff and a distraction from it. The comparison of the hand counts to the actual machine counts in Washington suggest that exit polling there over-estimated democratic strenght in the governor's race.

My gravest issue with the advocates of the exit polling is that they cannot frame the argument that would persuade them they are in error, though I have framed a case that would persuade me that they have it right. It tells me that they are leading with their emotions, not their logic; and emotions don't count a fillip with the courts or the media.

If that come out comment is directed to me, I hope you understand that I have internet access only at work. I do appreciate a careful argument, and I would agree with you, in general, that there appears something's wrong with the election, but it may be as simple as we two overeducated geezers can't fathom a stupid attention deficit electorate, swayed by emotion and mudslinging (I doubt it).

Mike

[euler]

124. Huh ?

Neither of you have addressed my points (or really anyone else's) directly, preferring to send us off to other sources, even though I have repeated my points three times so that I can be sure that you, like fourth graders, understand the points. Just address my points.

I may be mistaken, but I don't think you have ever asked me anything, directly or otherwise, let alone three times. I'm being honest here. I seriously don't think you have ever addressed a question to me. If you have, I haven't seen it. You certainly have not addressed me in this thread except in the post I'm replying too. Have you addressed something to me in another forum?

I really don't get the "or really anyone else's" part. I do nothing here except reply to other peoples points. I sort of have the feeling, that you are confusing me with someone else.

Who is qw ?

Please tell me the points you want me to address.

EDIT: I notice that mgr in thread #120 also wonders if perhaps you have me confused with someone else.

[Fly by night]

129. Look at my posts above -- you've already commented on several

"qw" is qwghlmian. You posted several times, tag-teaming with him, on the portion of the thread where I posted today. Many of my posts are repeated attempts to make the same points, since I didn't appear to be getting through to you two.

However, if you want to review my posts 14, 26, 47, 60, 92 and 121 and comment, please do. By my reckoning, you've already commented (at least indirectly) on my post 26 (twice), 60 (once) and my post 92 was a direct response to you. Not sure where your confusion comes from or why "mgr" got confused as to who I was responding to with post #92 -- the dotted line to you is pretty clear to me.

[anaxarchos]

141. Got through to me.

Hiya FbN... Busy in here today, ain't it?

[Fly by night]

142. Yes. It feels a little like a merry-go-round in Bizarro World. n/t

[euler]

152. agreed.

[qwghlmian]

132. Here is the relevant information from the link I provided:

http://www.mysterypollster.com/main/2004/11/the_difference_.html

5) "Corrected" Exit Poll Tabulations - Once the actual results have been counted in the wee hours of election night, NEP re-weights the results of each exit poll so that the vote preference on the poll matches the actual count. They then release new cross-tabular tables for each state to the general public. In theory, weighting to match the vote preference to actual results makes the complete exit poll more accurate.

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

The guy who wrote the above, the "MysteryPollster" is a well known
Democratic pollster, Mark Blumenthal, who has been in the polling business for 18 years. He lists his credentials here: http://www.mysterypollster.com/main/2004/09/about_mystery_p.html

Are you saying in the paragraph above he does not know what he is is talking about and you know better than he does exactly how NEP weighs its final data?

[Fly by night]

144. Please re-read my posts. Your reference doesn't relate to any of them.

My relevant posts are 14, 26, 47, 60, 92 and 121. I know what weighting of polls means and I know what voting information is used to weight polls. As I have said repeatedly, if the purpose of exit polls (as you state) is to assess the voting patterns of socio-demographic sub-groups (and not to predict the winner faster than counting the votes allows), there is no information available from the final voting results to re-weight anything having to do with socio-demographic characteristics because votes are cast anonymously and without any identifying information provided by the voters. However, if the purpose of the exit polls is to predict the winner of the race quickly, re-weighting the raw exit poll data from precincts to account for the differential turnout in, and size of, those precincts (which would impact the final results of the race) makes sense and is done. You're not re-weighting the exit polls based on who won -- you are re-weighting the raw exit polls using the final reported vote counts in the precincts to account for the differential turnouts at, and the size of, those different precincts. That is the principal "weighting" that occurs using the final vote to adjust raw exit poll data. It's not who wins, but who turns out (and in what numbers, and where) that is how the exit polls are adjusted.

I have now said this more times than should be necessary to debunk your arguments regarding the primary purposes of exit polls and to clarify the weighting procedures that must occur to make the raw exit polls more reflective of the final vote (not the final result, but the final vote). You don't adjust the exit polls to reflect the final vote -- you adjust the raw exit poll numbers to reflect the differential turnout within, and the overall size of, the different precincts (and by extrapolation, to all those precincts for which the sample precincts are assumed to be representative), both of which should improve the accuracy of the "weighted" exit poll data for predicting the winner, absent any untoward influence on the "reported" vote.

That is and will remain my final attempt at saying the same thing in as many ways as I can tonight. I will understand if you still cannot understand that which you do not want to understand. Understand?

[qwghlmian]

146. Yes, you repeated it half a dozen times -

but every time it was a naked assertion, with no citations to back you up whatsoever. There are, however, experts in the field who contradict you (see http://election04.ssrc.org/research/InterimReport122204.pdf). When faced with a choice between an assertion by an anonymous poster and statements by known experts, which do you think should be picked?

And another thing: the link I gave you explicitly contradicts your assertion. It says clearly that the "corrected results" (the ones that were weighted to conform to actual election results) serve the purpose of "provide a resource to help reporters and the general public interpret the results of the election.". Later he says that "The various "corrected" releases ... serve the third function, providing an analytical tool for reporters, scholars and the general public."

He also says that "Once the actual results have been counted in the wee hours of election night, NEP re-weights the results of each exit poll so that the vote preference on the poll matches the actual count. They then release new cross-tabular tables for each state to the general public.". What are the "cross-tabular tables"? Those are the tables that "that show how respondents answered each question and the vote preference calculated across answers to each question. The cross-tab tables play no role in projecting winners. Rather, networks and newspapers use them to prepare "analytical" stories about the election.". This directly contradicts your assertions.

[Fly by night]

148. You continue to ignore the content of my posts. I can't help you anymore.

I am the only one (of the two of us) who has laid out his credentials and experience in this thread. All you've done -- aside from ignoring the content of my posts -- is to cite the same link over and over.

Here's a link for you to choke on: http://slate.msn.com/id/1000228 /

So from now on, I not only cannot help you. I cannot hear you. You have given me my first reason to use the "snooze" button. My guess is that this "petit mort" will feel as good, and last much longer, than the other kind. Bye bye.

If the rest of you DUers want to keep playing with this guy, be my guest. But my guess is that my way is the better way. Ah, Republi-Nazi "true believers", we can't live with them and we can't ... (yet).

[euler]

154. See my...

...post to Fly by Night. Post #153.

[euler]

153. More of the same.

Have you posted any links to someone who agrees with you? I'm not close minded, but I don't know you. Give me a quote and link to someone I can Google, so I can verify their credentials. I've given you about 20 quotes with links and author names so far. I haven't read all of Mr. qw's posts but I know that in at least one of them, he provided some quotes and links from real exit poll expert. You can Google our supporting evidence and check credentials of the authors we quote. Give us something we can check out too.

[Fly by night]

158. Go up two lines in this thread. A link was provided for you there. .

Other than you and qw, the vast majority of the posters on this thread agree with the points I have made (and TIA and others have made before me, on this and other threads). Contrary to your statements, I have posted my experience and my credentials in this thread (something which neither you nor qw have done -- care to do so now?) but you are now admitting that you don't even read the other posts that come before -- even those of your tag-team buddy. The fact that you're asking me now for a link when I just provided one to you two lines above (a line that was posted before your latest comment) says to me that you're on auto-pilot or that you guys don't debrief at shift change thoroughly enough.

Fortunately, this is not a plane ride I want to continue, and since others are not responding to you anymore either, I guess I'm late for this decision. So as another DUer has said before, time to...

alert, snooze, puke, move on. Bye bye.

[TruthIsAll]

110. WHY IS IT THESE "PROFESSIONALS" NEVER CONSIDER THE POSSIBILITY OF FRAUD?

They always assume that the tabulated votes are correct.
If there is a deviation, it must be due to polling error.
No matter how many deviations.
Always polling error.

In FL, PA, OH, NV, CO, NC, NM....
All had polling errors.

Or they say, the polling is good, but people lie to pollsters.
Or CEO's vote late.
Or women vote early.
Or Democrats like to talk.
Or Republicans hate to talk.
Or Bush is a popular war president.

Seems to me, they use inverted logic.

Their premise is that we must assume that BushCo is honest and would never defraud America.
Twice, no less.

Just who is being naive here?
And who is being rational?

[euler]

126. Actually, they aren't saying anything...

...about fraud one way or the other. There may be fraud, or there may not be fraud. They do not endorse either viewpoint. What they are saying is that you can't use these exit poll results to prove it one way or the other. I thought this was obvious.

[TruthIsAll]

128. There may be fraud? There are over 40,000 documented "anomalies"

And there may be fraud?

You know damn well there was fraud.

[anaxarchos]

147. Have you seen this?

Right-wing blogger... Screeching halt after the scoop data hit.

The gist is:

"Omigosh! The fraud guys aren't tinfoil nut-jobs... They were right! This is serious. The exit polls create a historical legacy. Turn the guns on NEP immediately. This poll has to be screwy.", etc.

They absolutely embrace Blumenthal & Kaus. The "next big issue" is NEP trained the pollers incorrectly (omigod, by telephone). I wonder how long before that "issue" gets here?


http://orsa.blogspot.com/2005/01/exit-left-2004-exit-poll-failure.html

Thought you might get a kick out of it. There is a ton of shit like this out there... all shifting from "the bloggers are wrong" to "NEP is wrong". A lot of it echoes Brady.

[emcguffie]

98. Bravo!

Thanks for doing that.

[euler]

151. Same problem as your other posts

[Carni]

65. I thought exit polls were used partially to detect fraud?

Not trying to be a smart ass...Weren't exit polls what tipped people off in the Ukraine election?

[qwghlmian]

67. No, this is a popular misconception

that is encouraged by some people on DU. What "tipped people off" in the Ukraine election was documented, blatant, fraud.

http://www.telegraph.co.uk/news/main.jhtml?xml=/news/2004/11/28/wukra28.xml

[Carni]

68. Well apparently even the MSM is deluded

Because that is where I heard it.

[qwghlmian]

70. Why would you expect MSM to be accurate in

what it reports? MSM is like candy as opposed to real food - it is immediate gratification, but not nutrition.

The exit polls in Ukraine were at best a small part of the "proof" of fraud and at worst completely irrelevant. Even in the re-vote the exit polls (three of them) that were done were completely inaccurate, and outside of margin of error.

[Carni]

73. So now the exit polls were a small part of the proof?

Where as previously they had nothing to do with serving as a red flag (according to you)

Forgive me... but your post count and your talking points sort of screams agenda.

Proving there was no fraud seems to be a passion of yours.

Common sense tells me that a President with an approval rating hovering in the 40's did NOT win the plurality of the popular vote.

[qwghlmian]

74. Did you read the article about Ukrainian election fraud?

There was blatant, open, and documented fraud in Ukrainian elections. When there are incidents of pouring acid into ballot boxes or beating up voters and election officials, or documented and massive multiple voting, why would you need exit polls as "proof"?

Apparently to you citing facts presumes an "agenda". Way to have an open mind.

[euler]

81. Yes, you are...

...correct. The exit polls in such countries are designed differently than ours making it possible to use them to verify the election.

In the US, pollsters don't design exit polls for 'fraud detection' because the media pays for them. The media doesn't want a 'fraud detection' exit poll. They want a demographic media-driven exit poll. If we want a fraud detection exit poll, we have to come up with the cash to pay for it - MSM certainly won't pay.

[euler]

150. You have several misconceptions I think.

"99.5 percent of the time"

I think your referring to the confidence interval for calling a state, which doesn't mean what you say it does here.

Your #2 is just wrong.

Here are a few links.

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=279921#280137

[Peace Patriot]

82. Go back to original premises

Re: the current discussion, Dr. Freeman states that there is no reason the exit polling that was done (voter choices and demographics) cannot be used to verify the election--an unqualified avowal by Mitofsky himself:

"Most of their (Mitofky's and Lenski's—the pollsters') public pronouncements have been qualified; their only unqualified statements have been that (1) the exit poll was not designed to verify the election can hardly be seen as conclusive that it could not be used to help verify whether or not the election was clean; (2) that the poll was conducted correctly and, more recently, that (3) their data confirm the exit poll-official count discrepancy." --Dr. Steven Freeman, "Hypotheses for Explaining the Exit Poll-Official Count Discrepancy in the 2004 US Presidential Election," (2nd paper, draft)

Freeman 1st paper: http://www.truthout.org/unexplainedexitpoll.pdf
(also at: http://www.buzzflash.com/alerts/04/11/Expldiscrpv00oPt1.pdf)
Freeman 2nd paper (draft): http://www.appliedresearch.us/sf/epdiscrep.htm

Freeman's 1st paper on the exit poll discrepancy discusses the reliability and accuracy of exit polls, which are used worldwide to verify elections. In his second paper, in addition to debunking the reasons that have been invented for the 2004 US exit poll discrepancy, he discusses how elections are conducted in other democracies, for instance, in Germany, paper ballots are placed in glass containers where they remain, visible to all, until they are hand-counted, also visible to all, and all changes in tallies and exit poll data are visible to all.

In the US in 2004, the country was testing out a new and highly controversial election system nationwide for the first time. The machinery for the election system was manufactured by Bush "Pioneers" and other major Republican partisans and sold to the states with contracts that stated that the programming code that runs the machines and that tabulates all the votes will remain secret, proprietary information. One third of the country voted on machines that produced no paper trail to be used in a recount (a nearly unauditable condition), and Republicans in Congress had blockaded this and other transparecy measures. This electronic voting system had been proven to be extremely unreliable, insecure and hackable (many studies including the Johns Hopkins study, the blackboxvoting.org study, the chuckherrin.com demonstration, and demonstrations of hackability on TV shows). Doubts about the system were widely known. Further, the state of California had sued one of the companies (Diebold) for its false statements regarding computer security and certification.

This situation CRIED OUT for an exit poll specifically designed to verify the election. Why didn't the TV networks, AP (which handled the "official results"), Mitofsky & Lenski, major newspapers, the political parties, or SOMEBODY commission a poll specifically designed to check this election system's reliability?

They not only DID NOT commission the most obviously needed type of poll, they altered the information that Kerry won the demographic poll that WAS conducted, to fit the "official results," so that Americans were denied that information.

What other countries take pains to create in their elections--transparency--was utterly lacking in the US 2004 election, so that experts and and ordinary citizens, voters and election activists are now having to scramble to obtain and analyze information to figure out what happened--while officials in at least two states (New Mexico and Ohio) are busily destroying evidence.

I think we need to backtrack to original premises. It's plain now that the election system was a fraud going in. No one can be expected to trust "results" when only major Bush backers and donors are privy to how the votes are counted. On top of this, we now have a mountain of data--not just the Exit Poll discrepancy--that the election was extremely unfair, and that the presidential result was wrong (and possibly some other races as well, such as the Senate race in North Carolina).

The TV networks, AP, Mitofsky & Lenski, major newspapers, and the two major political parties PERMITTED THIS TO HAPPEN: an inherently fraudulent election system, with no specific verification measures, even after the debacle of 2000 concerning which it has been proven that the wrong man and the wrong party was installed in the White House.

The "controversy" about the the type of exit poll that was conducted is a non-starter. The exit poll discrepancy is only one piece of evidence in an overwhelming case that this election was invalid and produced a result for president that is, at the very least, highly suspicious, and is very probably wrong.

An extremely fraud prone election system that produces a suspicious result = an INVALID election.

Exit Polls commissioned by TV networks are OBVIOUSLY commissioned to PREDICT the result. That's what TV networks DO--they produce instant news. So it is ridiculous to assert that the TV networks commissioned a poll for some long term investigative or commercial purpose (gathering demographic information about voters). If the Exit Poll was not correct, and was not predictive, WHY DID THEY DISPLAY IT ON ALL TV SCREENS ON ELECTION DAY? AND, why was there NO CAVEAT stated and posted with this polling information that it was NOT predictive?

Clearly, they considered it predictive. That's WHY they commissioned it.

So the questions are:

Why did they hide this information from the American public?

Why are they backtracking now, giving excuses like Republicans didn't want to admit voting for Bush (no evidence for this whatsoever), or it was a demographic poll (no reason it can't be demographic AND predictive--even Mitofksy says this--so, was it a built-in "excuse" for later, when they had to justify Stolen Election II?)

Why are the pollsters refusing to disclose all polling data?

Why did NO ONE commission a poll purely for the purpose of verifying the election?

Why are Democratic leaders nearly silent on these issues (particularly on the fraudulent election SYSTEM, and on the Exit Poll discrepancy)?

Why do Republicans not give a damn that the election system is fraudulent on its face--non-transparent and completely untrustworthy--nor about their own party officials' massive disenfranchisement of black and other Democratic voters?

Having been brought up a Catholic, and taught in Catholic schools, I am very familiar with the type of thinking that starts with the premise: AUTHORITY IS ALWAYS RIGHT. All facts, narratives, stories, homilies, teachings, directives, historical presentations and sermons flow from that premise, and all facts, narratives, stories, etc., that contradict that premise are therefore completely ignored, or, if they happen to come up, and "known to be" in error and wrong.

Did the Catholic Church order the slaughter of 20,000 Albigensians in the First Crusade (in southern France)? If you are a young Catholic in a Catholic school, this fact does not exist. It is never spoken of. The Crusades were all worthy causes in the interest of the "one true Faith."

AUTHORITY IS ALWAYS RIGHT, in the Catholic Church, means "the Church" is always right. This education was very valuable to me, because it taught me just what kind of language and propaganda methods to look for when people are LYING and COVERING UP their nefarious deeds.

Since the 1960s (Pope John XXIII and ecumenism), this rigid and inherently false-faced premise has been tempered somewhat. But it still serves as a template for how powerful people with fascist tendencies try to confine the human mind to a rigid set of beliefs that serve the interests of the powerful.

So, I think, that is what is happening here, with the election fraud of 2004. The Iraq War serves the interests of the powerful, and since it absolutely cannot be justified in any way, a DELUSION OF CONSENT must be created and defended at all cost, even at the cost of Truth itself (the corrosive basing of society on falsehoods), even at the cost of the basic sanity of the American people, and even at the cost of their democracy and their right to vote.

I conclude with some refreshment from the past:

"I know of no safe depository of the ultimate powers of the society but the people themselves, and if we think them not enlightened enough to exercise control with a wholesome discretion, the remedy is not to take it from them, but to inform their discretion by education."

-- Thomas Jefferson, letter to William C. Jarvis, September 28, 1820 (thanks to Thom Hartmann for this quote)


"I have sworn upon the altar of God eternal hostility against every form of tyranny imposed upon the mind of man."

--Thomas Jefferson, inscribed around his statue in Washington DC

[euler]

149. As Promised

You said:
exit polls worked exceptionally well back when the votes were counted "as is" and not "weighted"

When did we not weight exit polls? As far as I know, it's always been done. In fact, I proved that it was done in 1988, 1994 and 2002 right here: http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=288785&mesg_id=289017&page=

Also, in his paper, The Unexplained Exit Poll Discrepancy - Page 3, Stephen Freeman notes that:

If you go to the CNN website or any other website on which 2004 exit poll data are available, you’ll see numbers very different from those released on election day. This is because the survey results originally collected and presented to subscribers were subsequently “corrected” to conform to official tallies. The pollsters explain this as a natural procedure: the “uncalibrated” data were preliminary; once the counts come in, they recalibrate their original data on the assumptions that the count is correct

Natural procedure? Doesn't sound like a new exit poll procedure to me. They've been weighting to the actual vote since 1967.

You said :
whoever controls the controllers -- who now "adjust" or "weigh" the consent of the governed to reflect the "evil massa's" wishes

The weighting is done by the pollsters - Mitofsky and Lenski.

You said :
denigrating the importance of exit polls as a measure of the honesty of the reported vote count.

See here: http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=284055

some analysts inappropriately attempt to use current exit poll results to investigate whether the results in a locale (state or country) are accurate or whether fraud might be involved in an election. A certain form of exit poll could be used for this purpose, but again the designs would have to be different.

This exit poll was not designed to "measure of the honesty of the reported vote count."

Again, in Freeman's paper:

(Mitofsky and Lenski) have taken great pains to argue that their polls were not designed to verify election results

You said :
They just were not accompanied by the honesty and integrity necessary to report those results to the public in a timely and un-"massaged" manner.

Again, they have always been massaged. It's common knowledge among pollsters. The exit poll would be worthless for the purpose it's intended if it wasn't 'massaged.'

Go to the following link to see a step by step discussion of the exit polling process on election day.
http://www.mysterypollster.com/main/2004/11/the_difference_.html

Check out step 4:

Once the polls close, NEP gathers actual results for the precincts sampled in the exit polls and gradually combines the exit poll results and the actual vote counts into an evolving hybrid of projections.

Check out step 5:

Once the actual results have been counted in the wee hours of election night, NEP re-weights the results of each exit poll so that the vote preference on the poll matches the actual count.

Your second paragraph seems to say that the demographic data contained within the exit poll is somehow destroyed after the exit poll is reweighted to the actual vote. I disagree. The exit poll is always weighted to the actual vote count. If doing so, destroyed the demographics, it wouldn't be done.

Finally, the consensus among MSM is that exit plolls are no longer suited for calling states on election night. The demographic data is now the highest priority. I cover all of this here: http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=288785&mesg_id=289481

[mgr]

162. Things change

I can remember networks calling elections as the polls closed as far back as 1968. The problem with your observation is that the accuracy of projections or exit polls have been reported to be in decline over the decades (Bush-Dukakis unadjusted is a howler). I have speculated that the sampling methodology and precincts employed (used election after election, this was let slip over the air in 1976 by NBC) are developed from when Democrats were ascendant, and appear to still do a good job of predicting democratic support, but miss republican support by a mile (it apparently is not a binomial relationship, e.g. p, therefore 1-p). There was little concern with this until 2000, when Florida blew up in everyone's face.

I think what is going on this year is that NEP is having to revamp more than just the methodology, and is scuttling a lot of the precincts that were employed in the past, so there is a greater deal of uncertainty and imprecision. Diminished expectations are that only local movements will be measured, to enhance the post election analysis.

Actually, exit polls are not the only way to know how a voting population behaved, the other is to violate the secrecy of your ballot by comparing votes with your neighbor, spouse, co-worker. I recall the faculty did this when votes turned out different than expected, and it was found the chair was making the results match his wishes. Highly impractical in a country of 200,000,000+, but do-able within this cyber community (of course, we are likely to self select other democrats, but if they all voted for Kerry....)

The reality appears that the exit poll conducted for the election this year has problems being predictive of overall voter intent.

I hope someone reading this post doesn't raise the canard that these are republican talking points--they (with the exception of the pin headed comment that women were oversampled) were aired at DU first, and follow the usual basic assumptions and criticism of scientific measurement and statistics, and are part of the standard assessments/critiques of experimental design.

Exit polling serves its purpose, I just wish that this year it were more robust. And, I am surprised to note that neither party appears to conduct exit polling (at least openly).

Mike

[RaulVB]

25. False. Talking points are not your own

Try again, harder.

[Yupster]

41. 50 % non-response rate?

I've never seen that before.

Is that true or were you just saying a high non-response rate?

[qwghlmian]

45. Yes, 50% non-response rate in US exit polls

http://www.mysterypollster.com/main/2004/12/what_about_thos.html
...
The response rate for the 2000 exit polls was 51%, after falling gradually from 60% in 1992. NEP has not yet reported a response rate for 2004.
----------

I would presume the 2004 non-response rates would be similar. This is compared to less than 20%+ non-response rates for the German exit polls.

[99Pancakes]

107. Why did the Ukraine

rely so heavily in their exit poll results, then? Even BushCo referred to them. Hmmmmm.....

[noamnety]

5. It's like predicting the weather

(If you keep the forecasts secret, only report the weather after the fact, and adjust your predictions to match the actual results.)

[euler]

44. That's only necessary if...

...you want to study cloud demographics.

[imenja]

6. it's a statistical sampling

of how people SAY they voted. They depends on polling a representative population from representative precincts. They are, however, far more accurate than public opinion polls, most of which have samples of only about 1000 people, not all of whom will actually vote. Prior to this presidential election, exit polls have been very accurate. They had Gore winning in 2000, but that's because more people cast votes for Gore. It was the election process itself that was flawed, though ultimately that and the Supreme Court is what decided the election.
A lot can be said about the 2004 exit polls, and I'll leave that for others. Exit polls, however, are not a waste of time. If they were useful in the Ukraine, then why not here? Naturally the Republicans want to get rid of them. Then they have no accountability if they should fix future elections. Democracy requires that we be able to verify elections. Though certainly not sufficient in themselves, exit polls are one way of doing that.

[TruthBeTold22]

71. But then.....

How do we know the responses were accurate? If I am approached by an exit poller I have three choices to make:

1) be proud I was asked and tell the truth
2) be indifferent and give the pollster what he wants...most likely the truth on how I voted.
3) be offended that anyone has a right to know how I voted...most likly resulting in a no response. However, how many devious folks tell the pollster that they voted for someone they didn't rather than give no response?

[Peace Patriot]

87. TruthBeTold22: Good question!

"...how many devious folks tell the pollster that they voted for someone they didn't rather than give no response?"

Do you know the answer?

[imenja]

117. you don't know absolutely

but people who refuse to answer filter themselves out. Some, as you suggest, may indeed lie. Statisticians figure the probability of that into their equations. It seems to me the answer is not to abandon exit polls altogether but rather to engage more organizations to carry them out rather than the single consortium the networks use. The only possible reason I can see for ending exit polling is to facilitate election fraud.

[anaxarchos]

133. It is really not an issue for the reasons you suggest.

Of all of the "black magic" associated with exit-polls, "non-response" is the easiest to deal with.

First, it is much more innocent than some people suggest. The difference between various national responses (80% German versus 50% U.S., etc.) is due to differences in national temperament (if you don't believe that, you haven't been to Germany), whether elections are held on workdays or holidays, the location of polling places, etc. The biggest difference, which may account for the gradual slide of U.S. response from just over 60% to just under 50%, is the growth of the survey questionnaire as U.S. polls have become more and more detailed.

Second, "non-response" is obvious. The pollster knows and corrects for it immediately. It is not something the polling organization has to guess at and correcting for it is fundamental to the methodology. This also applies to correcting for a gradual decrease in "responsiveness".

In order to impact the exit-poll, you have to show not just "non-response" but a CHANGE in "non-response" for the 2004 election that is much greater than that which was accommodated by the exit poll methodology AND that "non-response" must be tilted towards one party (or else it would even out as you say) AND cannot be skewed towards any other criteria (otherwise it would be detectable).

As you might guess, there is no empirical evidence for this whatsoever.

"False-response" is even rarer and it may even be a myth. The typical reference is to a single election in the U.K. in which Conservative voters were purported to have reported voting Labor because they were "too ashamed" to report their actual vote. Even this one instance is controversial and based largely on anecdote. It has no application for the U.S. 2004 elections which were essentially just a rerun of 2000 (and thus corrected in methodology).

Non-response and false-response are just spin. More properly, they are "mussahappened".

As in:

"Since the vote count CAN'T be wrong, the exit poll HAS TO BE wrong. Non-response, that's what 'mussahappened'".

Yeah, that's it.... indubitably.

[qwghlmian]

134. Here is a description that one DUer posted

of the exit polling process that he has observed himself:

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=97506&mesg_id=97575&page=

really does not instill a sense of confidence in the methodology, does it?

[anaxarchos]

140. My dog really likes exit polls however....

That brings it back to even.

Go be silly with euler.

[BeFree]

7. Hahaha

Look here, son, exit polls are big business in this here country. That's why $10M was spent on the exit polling this time around. Don't know who has been bending your ear, but exit polls are scientifically designed to be accurate within a margin of error of something like 1 point.

Exit polls have some of the best mathematicians designing and analyzing the numbers taken from selected polling places around the country. Those places are selected based upon history, History I say, and you can't argue with history.

:/)

[jkd]

15. Where did Mitofsky say MOE +/- 1?

The NEP website gives + or - 4% MOE for their state polls. Have they published different MOE's somewhere else?

[TruthIsAll]

17. SMOKING GUN: 1% MOE, RANDOMLY-SELECTED 13,047 POLLED.

I guess you have been away from DU for a while.

[jkd]

43. I been in sunny Cal. without computer access.

I believe these are adjusted poll numbers, so the MOE of course would be adjusted. Mitofsky first had the Hispanic vote at 45% and later adjusted it down to 41%. I believe these are those later adjustments.
When did NEP release this data? TIA you keep moving back and forth between raw and weighted polls to fit your purpose. NEP states that when they make final adjustments the numbers might change. Perhaps then the MOE will be 0%. But, these aren't those numbers you keep quoting to show the margin of error.

[TruthIsAll]

46. Why don't you just read the notes at the bottom?

Welcome Back.

Oh, this was up on the WP site on Nov 3.

It's still there.

[Igel]

93. Error "is somewhat larger for subsamples."

1% error for the nationwide figures. State exit polls are subsamples.

[TruthIsAll]

94. No kidding. That's why I calculated the individual state MOE's.

They are prominently displayed with the sample size and standard deviation in my state polling analysis.

I'm sure you've seen it. 20 states deviated to Bush BEYOND the State Exit Poll MOE (NM was not included because it was within the MOE by a hair - .01%). In any case, the probability is a big fat ZERO.

Regardless of wheteher you look at the individual state polls or the national exit poll, you come up with the same result:

............I M P O S S I B L E ...............

[euler]

39. I'm glad you agree. Here's a little history for you.

See post #29

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=288785&mesg_id=289017&page=

[bleever]

9. Something is so seriously wrong here, I keep failing to really

comprehend it.

On November 3rd, when I first heard that the final exit polls were "adjusted" with real votes, I wondered if I was crazy.

I came here, and I knew that if the answer was "yes", it wasn't because I was disturbed about the exit poll "adjustment". It was as f*cked up as the government saying they're going to shorten the inch.

Why?

Why no raw data, as demanded by their own standards?

Why say they're not accurate, when they are painstakingly designed to be so, for millions of dollars, progressively better every time?

Why say something IS an exit poll when it is a mixture of exit poll and NON-exit poll data? When it defies every scientific bone in any concerned body?

Why would this occur? Could this only happen within an enclosed system, where everything has been commercialized, and commerce is so easily controlled by the people in power for their own benefit?

:mad:

[Nothing Without Hope]

10. TIA, you sound really tired and down.

The exit poll data do have value, though they are not enough on their own to prove fraud occurred.

[TruthIsAll]

18. Come on. Down? Hehehehe. Read the post again. n/t

.

[RaulVB]

22. He sounds pretty upbeat to me (n/t)

[go west young man]

145. Its called sarcasm. HELLO!

TIA you rock! I am glad to see you continuing the good fight. I've seen you taking a lot of flack from Euler and others but you continue to do good research, make great points and keep the issue at the front of our minds. It doesn't need to go away as some would suggest. It could unravel yet and then those naysayers would have to DU with a permanent poop stain on their faces for having their heads up their asses and trying to dissuade you. The worse that can happen for you is you could be wrong. But I doubt you are. My personal biggest reason I believe it was election theft is simple. Because the mainstream media doesn't ask if it was. Anyway you have a supporter here.

[Nothing Without Hope]

157. Looks like I was the one that was tired. Thank goodness!

In my muddled state it looked like TIA was giving up on exit poll analysis, which is NOT GOOD!

Fight on, TIA!

[Patsy Stone]

11. TIA, you're wrong and you know it :)

I still want proof. Call me picky, but... If nothing's wrong here, then cough it up. Show me the numbers. Prove to me that there was absolutely no way that I could have figured out who the winner was going to be. PROVE IT! They can't and they won't, because they are criminals. What are they hiding?

The exit polls were right in 2004, just as they were right in 2000. More people did vote for Gore in 2000, or (in the case of our Palm Beach, Pat B. voters) thought they were voting for Gore.

NGU

[valis]

12. Geez, the media said that from day 1... The exit polls are

weighted, by using a bunch of assumptions... Why cannot we find a way to sample that does not require all this weighting BS? I suppose it's because of the cost involved to get a truly representative sample...

[Nederland]

16. Finally

You are making sense.

It's a shame you are probably being sarcastic.

[RaulVB]

20. Your ignorance shines (n/t)

[euler]

33. But not quite as bright as some.

[anaxarchos]

28. They are accurate enough...

...to help convince 35% of the American Public (up from 20% immediately after the election) that the elections were questionable.

They are accurate enough to have triggered a right-wing campaign against exit-polls themselves.

They are accurate enough to have many people still talking about them 3 months later.

[marions ghost]

62. anaxarchos can u post

here, or as another thread--a link to the poll that shows this 35% figure? Or who did the poll, and I can look it up? Am not questioning you, just that I can't find the source and need it for something else.

Also--you make 3 very good points here about exit polls IMO. I will use them as well. thanks mg

[anaxarchos]

84. Here's 30%....

http://www.annenbergpublicpolicycenter.org/naes/2004_03_mindset-bush-kerry-supporters_%2001-05_pr.pdf

I don't remember where 35% was but I will look it up.

[euler]

29. HOW TO LOSE CREDIBILITY IN A SINGLE POST.

You may have learned something about exit polls today, but it wasn't enough. If they weren't reweighted they would useless for the purpose they are intended. Reweighting makes perfect sense to people who know something about them.

Exit polls have always been re weighted to the actual vote, and they always will be. That's how exit polls are conducted. It's not a secret.

For crying out loud TIA, Stephen Freeman notes the fact that exit polls are reweighted to the final vote count in his paper: The Unexplained Exit Poll Discrepancy - Page 3

If you go to the CNN website or any other website on which 2004 exit poll data are available, you’ll see numbers very different from those released on election day. This is because the survey results originally collected and presented to subscribers were subsequently “corrected” to conform to official tallies. The pollsters explain this as a natural procedure: the “uncalibrated” data were preliminary; once the counts come in, they recalibrate their original data on the assumptions that the count is correct...The pollsters have taken great pains to argue that their polls were not designed to verify election results, but rather to provide election coverage support to subscribers – as one set of data that the networks could use to project winners and to explain voting patterns, i.e., who voted for whom, and why people voted as they did. Whatever the merits of calibrating exit poll data, it obscures the issue of why the uncalibrated polls were so far off and mostly in the same direction. Although this calibration process may seem perfectly natural to NEP, it confuses nearly everyone else.

Well, there's no doubt that it confuses a lot of people. But it wouldn't confuse someone who has posted hundreds of documents in this forum related to exit polls - or would it ?

Look at these official exit polls from previous elections. All are available at the Roper Center FTP site here:
ftp://ropercenter.uconn.edu /

NBC News/Wall Street Journal Poll: National Election Day Exit, 1988

ftp://ropercenter.uconn.edu/United_States/NBCWSJ/USNBCWSJ1988-NATELEC/version2/usnbcwsj1988-natelec.pdf

Paragraph 1 states: The data were weighted to correctly reflect the outcome of the election.

Voter News Service National Election Day Exit Poll, 1994

ftp://ropercenter.uconn.edu/United_States/VNS/USVNS1994-NATELEC/version2/usvns1994-natelec.pdf

Precincts in the national sample are weighted to regional totals during the day so that incomplete regional tallies can be combined in their correct proportions to form a national tally. Tabulations are meaningful only if they are weighted. Unweighted tabulations will make no sense. Ouch.

VNS 2002 Election: National Election Day Exit Poll

ftp://ropercenter.uconn.edu/United_States/VNS/USVNS2002-NATELEC/version3/usvns2002-natelec.pdf

the weighting and processing procedures were similar to previous years: there is a non-response adjustment for age, race and sex, and the survey is forced to the final outcome in each region within meaningful strata.

Check out this link:

http://www.mysterypollster.com/main/2004/11/the_difference_.html

It explains all the steps exit pollsters take throughout the night:

Check out step 4:

Once the polls close, NEP gathers actual results for the precincts sampled in the exit polls and gradually combines the exit poll results and the actual vote counts into an evolving hybrid of projections.

Check out step 5:

Once the actual results have been counted in the wee hours of election night, NEP re-weights the results of each exit poll so that the vote preference on the poll matches the actual count.

Moving on to your next misconception. You are fond of saying that if exit polls are weighted to the actual vote, why bother ? You seem to say that you think exit polls are useless if they are weighted. Well, you know very little about exit polling. These are basic questions. If you don't want to work to improve your understanding of exit polling, I'll explain it to you. Let me know.

EDIT: Removed a 3 word phrase that was argumentative.

[Nederland]

35. Even if TIA doesn't want to know

I do. Please continue.

I've disagreed with TIA all the way back since he told us that the 2002 election didn't agree with the exit polls. I also disagreed with him when he told us Bush wouldn't run in 2004. My reasons had to do with the refusal to admit that exit poll samples might not be representative, but your reasons seem to be different. I'd love to hear them.

[euler]

40. OK. I've posted extensively

I'll compile the links and post them here tomorrow.

[TruthIsAll]

48. You always disagree. And you are always wrong. Except for Bush.

2002. Max Cleland. Georgia. Oh, yes, I remember it well.
He was Diebolded right out of the senate.

But I think it's great that you challenge me every chance you get.
I see you have lots of company here.

I'm actually getting to enjoy this.

[euler]

63. THAT'S IT? That's your argument ?

I provided official US presidential exit polls from 1988, 1994 and 2002, all of which explicitely state that they were weighted to the final vote count. Yet you refuse to accept the that US presidential exit polls get weighted to the actual vote count. In the face of evidence that contradicts your 'analyisis', the right thing to do is defend yourself. Can you defend yourself ? If you can, then do it. If you can't then....

[BansheeDem]

76. Here are the cold hard facts concerning Max Cleland and his loss ...

I have seen a number of posts where you insist that Max Cleland was Diebolded in the 2002 Senate election.

Well, let me say this ... We worked hard for Max Cleland in the 2002 election in Georgia; but we did not fool ourselves into thinking that it would be a runaway for Max. The late polls clearly showed that the race was tightening, and that Chambliss had a shot at defeating him. Senator Cleland was successfully painted as a liberal by the Republicans and this showed in the polling data as the election drew near. What we were not able to do was put forth a clear and united message that appealed to the larger population of this red state. Max wasn't Diebolded; he was out-maneuvered. It's as simple as that.


In a Poll conducted by M-D on October 21, 2002

Sen. Max Cleland (D): 47%
Rep. Saxby Chambliss (R): 41%
Undecided: 9%
Sandy Thomas (Libertarian): 3%

MOE +/- 4%

While Chambliss' numbers have remained static from the last M-D poll, Cleland has been pushed under the all-important 50 percent mark. Greene blames Chambliss' negative campaigning:

[RaulVB]

104. Can you offer PROOF that you worked with Cleland's campaign? (n/t)

[BansheeDem]

118. I didn't say that I worked directly for the Cleland campaign ...

I said that we - meaning the grass-roots Democratic party in Georgia. However, I was involved in the GOTV campaign and went door to door in the last weeks to ensure as many people would be voting Democratic as possible. Like you, I know how to read a poll, and it was clear that Chambliss was closing the gap - so to me, it was no big surprise when he won.

[Carolab]

171. That's a pretty wide gap to close in such a short time.

Besides, there were other 2002 "upset" elections, not just Max's.

[valis]

37. There go all the calculations down the drain...

[sportndandy]

111. Euler has not established credibility

Why do the 1988 exit polls show Dukakis beating Bush 1? Why are the 1996 exit polls in close agreement with the results? Why are the 2000 and 2004 exit polls off? Why is it that exit polling loses its accuracy when someone named Bush is running?

But exit polling discrepancy is not a smoking gun. It is the evidence of a smoking gun elsewhere. It is the reason that more investigations into the real voting is necessary, not an excuse to scrap polling.

[k8conant]

163. Granted, weighted exit polls don't tell us anything about the vote...

euler, you wrote to TIA:

Moving on to your next misconception. You are fond of saying that if exit polls are weighted to the actual vote, why bother ? You seem to say that you think exit polls are useless if they are weighted. Well, you know very little about exit polling. These are basic questions. If you don't want to work to improve your understanding of exit polling, I'll explain it to you. Let me know.

OK, I'm not a pollster. If they are weighted, of what use are exit polls? Why don't we just wait for election results? If we want other demographic and opinion information, why aren't non-election-day polls used for that?

Explain it to me.

[bleever]

32. How many ways can you weigh the way the word "weigh" is weighted?

Wait: don't answer yet.


ed: sp, dang.

[anaxarchos]

34. They are accurate enough...

...to get TIA mugged every time he opens his mouth.

[TruthIsAll]

50. Why is it that only mathematicians like you agree with me?

I saw your post on MOE calculations.
Very thorough.

Hope you stick around for a while.
Backup helps when I'm being assaulted.

They see my posts and they pounce.
Like clockwork.
Orange.

[anaxarchos]

56. Don't take it too personally...

....the right seems to have discovered (3 months too late) just how important exit polls were.

I'll stick and try to control my temper. Don't know how you do it.

[TruthIsAll]

57. Oh, I learned.One has to be very careful. No flames.

I've been tombstoned.
You won't believe some of the names I've been called.

Don't be goaded into a flame war.

Keep repeating.
Just the facts.
Just the facts.

Say it a thousand times.
Like Howard Hughes.

[anaxarchos]

58. Ha... Or Dustin Hoffman in 'Rainman'... n/t

[euler]

64. They see my posts and they pounce.

Yes. Will you please pounce back? When I pounce, I expect you to pounce back with counter-arguments. You never do. Why is that? Hmmmm.

[TruthIsAll]

72. The numbers speak for themselves. No need for me to pounce back. n/t

Others here seem to get it. Why don't you?

Hmm...

[euler]

78. Your claims that exit polls are not d to the actual vote.

My claim is 'Yes they are.'

Who said anything about numbers? You're still at the 'taining wheels' level. We can talk numbers later.

[euler]

79. Will you speak at all to post #29 ?

You seem to be ignoring it. It isn't going away. Perhaps I'll start a new thread, and put it front and center. Would you adress it then ?

[googly]

51. So far everyone has missed the real reason why exit polls are done

At the risk of sounding pompous, I have actually worked from
the inside in democratic politics in Chicago. The main reason
exit polls are done is not to gain a few hours before the actual
vote counts come in from key precincts. The election results can
be predicted much more accurately once those key precincts are
tabulated and counted. That method of using actual vote counts
from key precincts has ALWAYS been accurate so long as the guys
picking the precincts are experienced and knowledgable individuals.

So why millions are spent on exit polls? You really want to know,
right? Ok, I will shed the light of knowledge on the darkness of
those who don't know. The politicians want to know how different
voting blocks such as white men, hispanic women, blacks with college
degrees etc. voted. That helps them direct the resources for the
NEXT election. It also helps them frame the issues for the NEXT
election. The actual vote counts have no such breakdown data for
various voting blocks.

This may help explain why Kerry and the other experienced pols
did not make an issue of exit poll accuracy as a lot of novices
seem to be doing.

[RaulVB]

53. False.

The networks here and around the world use them TO CALL RACES.

[euler]

75. TO CALL RACES - Not close races

Let's see what CNN says about using exit polls to call states.

http://archives.cnn.com/2001/ALLPOLITICS/stories/02/02/cnn.report/cnn.pdf

Exit polling is extremely valuable as a source of post-election information about the electorate. But it has lost much of the value it had for projecting election results in close elections... (We recommend that CNN) Cease the use of exit polling to project or call winners of states.

The key phrase here is 'in close elections.' The only states that MSM can call, are the ones that aren't close. But everyone already knows what the outcome will be in those states, so the capability for MSM to call those states is a hollow one.

The report contains interviews CNN personnel:

As to whether CNN would consider ending the use of exit polling to project winners, Bedingfield said more study is needed to find out if there is a real decline in the accuracy of exit polls which make them too flawed to use. He noted that Judy Woodruff, one of the CNN anchors, favors ending their use for projections but that her argument has not won out so far. It turns out that others of the on-air team also oppose exit poll-based projections.

The report contains a recommendations section:

in states where it is not possible to make a call at poll closing time based on exit polling, the use of the exit poll in that state would essentially be dropped from consideration.

Early calls serve no particular public or journalistic purpose. Taking a long view, it appears they do not even serve a commercial purpose. We believe the networks should return to reporting election outcomes based much more on actual vote counts and much less on the crystal ball of exit polls and sketchy returns.

On the question of exit polling, Schneider, who is CNN’s political analyst, said there are more serious problems with the accuracy of the polls than in the past due to non-responses and a rise in absentee voting. He said that taking exit poll information from one source (VNS) is “inherently risky,” and that he favors ending their use as “a primary tool in projecting results.”

Did media hold back this time when it comes to callig close states ?

The dirty little secret is that the exit polls alone are almost never used to project winners in seriously contested states, or (as the reader put it) in states where the winner was not already perfectly obvious 24 hours before the election....Only Alaska and West Virginia were called at poll closing time on the basis of exit polls alone and few still considered those states truly competitive by Election Day. All of the true "battleground" states were called later in the evening on the basis of actual results.

Mitofsky's recommendation ...was to use exit polls to project winners only when they show a candidate cruising to a huge landslide victory.

So why do the networks shell out millions of dollars for exit polls? First, they do care about making projections as quickly as possible ...Second, they use the exit polls to fill airtime on election night with analysis about why the candidates won and lost.

Total elimination of exit polling would be a loss, but its reliability is in question. A non-partisan study commission, perhaps drawn from the academic and think-tank communities, is needed to provide a comprehensive overview and a set of recommendations about exit polling and the linked broader problems of polling generally.

It seems that some people are more right than others.

[googly]

90. Thank you Euler, you have proved my theory, eloquently!

[googly]

83. And as everyone knows, networks have been wrong badly

in 2000 and other elections.

[TruthIsAll]

155. They were wrong? Gore won Florida by 50,000, but he still lost.

I wonder how?

How about...
65,000 undervotes.
110,000 overvotes.

75% of them with a Gore punch.
The networks were right when they called it for Gore.
They did not no that 175,000 punched cards were "spoiled".
They did not anticipate the butterfly ballot.

Once again, this proves you don't know what you are talking about.

[googly]

167. If you have proof that will hold up in court, I will contribute to the law

suit which must be filed on behalf of VP Al Gore, who
was wronged by Katherine Harris & Jeb Bush amongst others.

[TruthIsAll]

54. I guess you mean Kerry will win next time...

		HORIZONTAL			WEIGHTED		
PARTY ID	
     MIX	Bush	Kerry	Nader	Bush	Kerry	Nader
Dem 	38%	9%	90%	1%	3.4%	34.2%	0.4%
Rep 	35%	92%	7%	0%	32.2%	2.5%	0.0%
Ind	27%	45%	52%	2%	12.2%	14.0%	0.5%
	100%				47.77%	50.69%	0.92%
				
Probability	3.29447E-09	or  1 in 303,538,508

That Bush would go from 47.77% to 50.73%

[googly]

85. TIA, you are still using exit polls as accurate predictors

and they have been proven not to be so. IMHO their main use
is in learning about how different blocks are voting, to an
aproximate degree. Exact accuracy here is not paramount.

OTOH selected key precinct actual results have much-much
higher reliablity in predicting election results earlier than
waiting for total final count.

[TruthIsAll]

114. Pristine exit polls reflect the TRUTH. Tabulated votes reflect the FRAUD.

If an election is rigged, then the riggers must:

1- Claim the unweighted polls are innaccurate.
2- Rig the exit polls by contaminating them with "weighted adjustments" to match the votes.

Or perhaps a combination of both.

If YOU wanted to rig an election, what would you do about exit polls?

[TruthIsAll]

55. Oh, they want to know the voting blocs, but not the votes.

I suppose that's why everyone focuses on exit polls, to know as soon as possible how the Hispanic vote went - to plan for the NEXT election.

Sure. I can understand that.
That's important info.
They need to know it ASAP on election day.

Boy, I'm learning a lot today.
Thanks for the insight.

[googly]

86. Yap, it is too late to do anything if for example the Hispanics

are breaking in a big way against your candidate. But the
party can learn a hell of a lot on how to channel valuable
dollar resources for the next campaign.

[magellan]

59. We can't possibly do exit polls to direct resources

and frame issues for the NEXT election. Because as we all know, exit polls aren't reliable. They're high-priced fluff, good only for filling air time on election night while we wait for the REAL results to come in.

Funny thing is, I hear other countries use exit polls to help validate their elections. When discrepancies are found, as in -- was it Chile? -- and the Ukraine, the election is investigated.

Those places obviously don't see the entertainment value in exit polls like we do.

</sarcasm>

[euler]

77. I hear other countries use exit polls to help validate their elections

Correct. The exit polls in such countries are designed differently than ours making it possible to use them to verify the election.

In the US, pollsters don't design exit polls for 'fraud detection' because the media pays for them. The media doesn't want a 'fraud detection' exit poll. They want a demographic media-driven exit poll. If we want a fraud detection exit poll, we have to come up with the cash to pay for it - MSM certainly won't pay.

[googly]

89. I never said exit polls can be wayyyyyyyyyyyyyy off, what I said

is that exit polls are based on a very small percentage of
actual number of votes cast. Depending on which voters were
willing to stop and talk to the pollster, results can easily
get skewed. That is exactly why they keep adjusting and
weighting the exit polls to make them closer to reality.

Yes, I agree with you that exit polls can be used to prove
massive vote fraud, as commonly occurs in some countries.
BUT EXIT POLLS CAN NEVER SUBSTITUTE ACTUAL VOTE COUNTS. One
is a statistical projection relying upon skill of the pollster
whereas the other is actual vote count which as the recent
example in Washington state proved that it was within a couple
of hundred accurate out of 2.3 million cast (0.0085% accurate)
with THREE DIFFERENT countings. Exit polls would be lucky to
come in with 2% accuracy.

[TruthIsAll]

95. You show a complete disregard for or ignorance of the facts.

The National Exit Poll sample was 13,047 with an 1% MOE.

State exit polls ranged from 600 (4%) to 2846 (1.86%), an average 2.85% Most larger EV states had around 2000 respondents - that's an MOE of 2.2%.

You fail to take into account the fact that we have 50 states in which 20 exceeded the MOE in favor of Bush - and not one for Kerry.

I have already stated the probability: ZERO!

So your whole post is a fallacious waste of time.

[googly]

100. You may benefit greatly by taking Statistics 101 at any

accredited university, as I did during my engineering
curriculum at U of Iowa. That 1% MOE you state has tons
of assumptions behind it.

IMHO the repug voters for Bush in 2004 were particularly
reluctant to express their preference. The voting district
I live in is a prime example. I saw a whole bunch more Kerry
yard signs than Bush signs. And yet, this is a predictably
repug district and voted as such. In 2000, that was not the
case. There were the usual majority of Bush yard signs over
Gore signs. I don't know if it is the Iraq war or is it 911,
or something else. But Bush supporters have been noticeably
silent this time.

As for exit polls, if you look at the statistics, people in
higher income brackets preferred Bush except the super rich.
Typically the super rich do not have a 9:5 job. But those in
other high income brackets typically put in long hours at jobs
or businesses. These people gotta be reluctant to spend time
talking to a pollster. That is my theory why the exit polls
usually favor democrats since a lot of women who are full time
moms may have more time on their hands to talk to a pollster
than lets say a owner of a small business who is putting in 60
to 70 hours/week.

But, all that said, I respect your right to reach your own
conclusions regarding accuracy of exit polls.

[TruthIsAll]

106. I should take Stat 101? Tell me more about "reluctance" and "women".

What a joke.
So, you're are an engineer?

It's not very scientific to use Repuke talking points with those worn out, inane suppositions. They were debunked a long time ago.

Those pronouncements about voting behavior reveal that you have no serious argument to make.

You are grasping at strawmen.

[googly]

122. My main point is actual vote counts will always be more accurate

than exit polls based on less than 1% of voting public. But
obviously you are convinced it is otherwise. To bolster my
hypothesis I recited the recent 3 recounts for governor of
Washington. 2 were machine counts and 1 was manual count.
ALL THREE CAME OUT WITHIN 0.008% of each other out of 2.5
MILLION votes cast in state. Even if Dr. Einstein was given
the job of exit polling with a sample of less than 1% of the
votes, he could not come up with any thing resembling 0.008%.

Now if you want to talk about voter suppression, it is obvious
that was the case atleast in Ohio, where the lines in heavily
democratic districts were inordinately longer than republican
districts. Whose fault was that? Ofcourse Blackwell had some
thing to do with it. But I would also blame the local officials
in the democratic districts whoever they were. They should have
raised hell BEFORE the election.

Even Kerry and Edwards had the smarts not to make an issue out of
exit polls. Kerry did mention voter suppression and correctly so.
I think there are too many people wasting time on statistical
projections based on exit polls in such a close election.

[euler]

130. I don't....

...think he's an adult. Seriously. If he's not a teenager, I'll eat my hat.

[anaxarchos]

135. Wait a second....

.... I gotta hear more about the Yard-sign poll in the midst of this learned dissertation. You say Kerry won that one too?

Wow.... Go figure.

[Peace Patriot]

91. I don't understand why those who attack TruthIsAll don't discuss...

...all the OTHER exit poll discrepancy studies that have been done, and all the OTHER evidence of an unfair, invalid, fraudulent election! Let me just list SOME of them...

All question the 2004 election result on the basis of the Exit Polls:

Dr. Steven Freeman 1st study: http://www.truthout.org/unexplainedexitpoll.pdf
(also at: http://www.buzzflash.com/alerts/04/11/Expldiscrpv00oPt1.pdf)
Dr. Steven Freeman 2dn study: http://www.appliedresearch.us/sf/epdiscrep.htm
Dr. Ron Baiman: http://www.freepress.org/departments/display/19/2004/997
Dr. Webb Mealy: http://www.selftest.net/redshift.htm
Jonathan Simon:http://www.scoop.co.nz/mason/stories/HL0411/S00142.htm

130,000 to 230,000 phantom votes for Bush in FLA (paper vs. electronic voting)
Dr. Michael Haut & UC Berkeley stats team: http://ucdata.berkeley.edu

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

And I repeat my question from above: With the US testing out a new and highly controversial election system nationwide for the first time, WHY DIDN'T the powers-that-be who control election information to the American public conduct an exit poll SPECIFICALLY DESIGNED to verify this election--especially in view of the 2000 debacle--instead of MUDDYING THE WATERS with a voter choice/DEMOGRAPHIC poll that could later be "questioned" with easy excuses and "talking points" about its design?

The situation CRIED OUT for specific election verification. Why wasn't it done?

[Igel]

96. The last question is a good one, but a rather different one from what

everybody's been discussing. (If you can't get past the "exit polls are divinely revealed truth" you can't get to this question.)

I certainly can't give an answer, but to kick off the debate, I'm not even sure there's an administrative body at the federal level that would be able to commission such an exit poll. Maybe the DOJ? As far as I know no other administrative unit has any jurisdiction over elections (but I'm a linguist, not a political scientist or lawyer).

Congress could, no doubt, but I'm not sure they'd want to finance it in states that have used the same technology for a couple of decades(such as Ohio). And I'm sure that states would consider it a slap in the face, if the exit poll were carried out in just selected states: the federal government implying that it doesn't trust those states to administer the elections.

The states that adopted new tech could certainly finance the exit poll. But to do so at the same level would imply the SOS doesn't trust the BOEs, and presumably the voters that elected them.

As for most of the exit poll studies, they make the same, or same kind of, assumptions that TIA does, and have the same problems. You must assume that you have the raw data, and that the raw data can be easily interpreted. If you accept those assumptions, TIA's stats are unassailable (although his conclusion that the posted election results are impossible doesn't follow). The problem is that euler and others say at least some of his assumptions are wrong (purpose of the exit polls, 1% error for state polls, raw data require knowledge of the the model assumed in devising the sampling method and can't be intelligently interpreted without the model, and others).

(Excuse me, I'd like to go on but I have a baby in melt-down mode.)

[TruthIsAll]

97. Wrong again. I never used a 1% MOE for state polls. Do your homework.

And I have not done anything unusual with the polling data.

First, you guys argued that exit polls were not random polls.
They are: Mitofsky said so.

Then you said individual state MOE's were high.
But I calculated them using a simple formula based on sample-size.
And Bush exceeded 20 state exit poll MOE's.
The odds: merely ZERO.

Then you said the 1.0% National MOE was too low.
But it isn't. Mitofsky said so himself.

Using the 1% MOE combined with the Party-ID voting mix, the odds are 1 in 303 million that Bush would see his national exit poll vote morph to his final vote percentage of 50.73%.

Then I showed you that 43 out of 51 states deviated to Bush.
We can ignore the MOE's. Just consider the fact that 43 went to Bush.
I calculated the odds: 1 in 2.9 million.
You can't argue MOE on this one.

So what are you left with?

I guess we will have to wait for the Mitosfsky explanation.
But now we hear that the public won't get see the data - only the media will. We all know what the "liberal" media will do with the data.

Just like they did with the WMD data.
Just like they did with the Bush AWOL data.
Just like they did with the 2000 Florida data.
JUst like they did with the 2002 Senate data.

Not a damn thing.

[jkd]

99. You certainly implied it.

If you didn't say state exit polls were accurate within +/- 1%, then you certainly implied it in your answer to my post #15. I followed your suggestion to look at the Washington Post information. There, Mitofsky gives individual state MOE's. They all fall between +/-3to 6%. Perhaps you should revise the states that acceded MOE based on that data. Also some of your actual vote results need adjusting, but I can see you have already done some to that.

[TruthIsAll]

105. I was very clearly referring to The National Exit poll on the WP site.

You are not looking at the WP NEP site. You are confusing the NEP the individual state polls. The MOE is 1.0% in the NEP, because 13,047 were sampled. That really gives an MOE of 0.86%. But we can round it up to 1.0%

Haven't we had this discussion before?

For you to insinuate that I was "implying" anything else is total BS.
You should know by now that I provide ALL relevant data and calculations in my posts. I don't imply anything. I state it. If you do not agree with my data or calcs, show me yours.

I will not revise the MOE's because you say so. These were computed based on sample size. And I will stick to those numbers.

The MOE's are calculated the same way as pre-election state polls - purely based on sample size. But the exit poll sample-size is significantly HIGHER than that of pre-election polls. That's just one reason that exit polls are more accurate than standard polls.

You, of all people, must know that to be true. You're familiar with BYU?

The other is that the individuals polled are NOT undecided. For this reasons, my calculated MOE's are higher than equivalent pre-election state poll samples. That being the case, even if you apply a 30% cluster factor you will be right back where you started. So the MOE's are correct.

A total of 2846 were exit polled in FL. I calculated the MOE using the formula 1/Sqrt(2846). That's 1.84%.

And I suppose that is too low for you.
Well, too bad. Take it or leave it.

Where is your formula?

[jkd]

143. I'm leaving it

I’m confused. I asked a question about state exit polls and TIA responded with inappropriate national polls. He subsequently chastened me for not reading his post.

If exit polls are based only upon sample size, then why must pollsters publish their MOE? These polls are not simple random samples. The polls are both multi-staged and stratified. Mitofsky has published his MOE for each state. Is TIA actually saying that his margin of error for NEP’s poll is correct and they are wrong?

Mitofsky gives his MOE: http://www.washingtonpost.com/wp-srv/politics/elections/2004/graphics/exitPolls_florida.html. The MOE’s are between +/-3to 6%.

TIA gives his MOE: “A total of 2846 were exit polled in FL. I calculated the MOE using the formula 1/Sqrt(2846). That’s 1.84% …….Take it or leave it.

I’m leaving it.

[euler]

108. Yes...

...this statement is correct:

The problem is that euler and others say at least some of his assumptions are wrong (purpose of the exit polls, 1% error for state polls, raw data require knowledge of the the model assumed in devising the sampling method and can't be intelligently interpreted without the model, and others).

I do make these arguments. But even if I couldn't, I would still post my dissent.

I didn't know what to think argot this whole issue when I first dived in. I got over it by reading everything related to this topic. On about day 2 of my efforts, after learning that Mitofsky designed and implemented this exit poll, I came across statements by Mitofsky that made it clear to me that the type of analysis being done by TIA and others in this forum, are bogus. I subsequently heard the same from various other experts.

Now, Mitofsky designed and implemented this exit poll. He has conducted 3000 exit polls over a period of 37 years. He has written dozens of books and papers on exit polls, he is considered the father of the whole concept of 'Exit Polling' and conducted the very first exit poll ever conducted in the US in 1967. Who am I to question credentials like this ?

TIA, on the other hand, admits he is not a exit poll expert.

Even if I knew nothing at all about exit polling or statistics, I would still argue that TIA's analysis is bogus, based solely on the statement of the man who designed and implemented the exit poll in question - Mitofsky.

When I first started sending Mitofsky quotes to TIA so he could defend his analysis against Mitofsky, something bizarre happened. I got response like:

"I'm right, Mitofsky is wrong,"
"You're an idiot"
"I know better than Mitofsky about exit polling"
"Mitofsky is biased."
"Mitofsky works for the media. The media is in on it"
"Do the math stupid, prove me wrong."
"The exit poll is just a poll. Mitofsky is full of it."
"I'm correct."

So, TIA, who admits he is not exit poll expert, does away with the man who invented exit polling with the twist of his finger and a pop of the tongue. This is simply bizarre. I think it may be 'A' word

[googly]

101. Could'nt agree with you more! We have a mess on our hands because...

So long as we do not have a fool proof system of counting
votes, paper trails etc. We are letting people vote without
having to provide proof of citizenship, no way to verify dead
people voting. One person moved 3 times since last election
and received 3 absentee ballots in the mail. Another person
in the news here voted 32 times, by using different names! He
got caught, but how many more are there who did not??

Our voting system is a national disgrace.

[euler]

103. For me, it's simple

Mitofsky designed and implemented this exit poll. He has conducted 3000 exit polls over a period of 37 years. He has written dozens or more books and papers on exit polls, is considered the father of the whole concept of 'Exit Polling' and conducted the very first exit poll ever conducted in the US in 1967.

Your guys know statistics, but none are expert exit pollsters.

Mitofsky and Lenski have made public comments since the election that, in effect, debunk, the scholars in your list of experts. Freeman (to his credit) even acknowledges that Mitofsky and Lenski are not supportive of his analysis. Mitofsky beats your experts any day in my opinion. Of course there are others experts who also agree with Mitofsky.

You say:

WHY DIDN'T the powers-that-be who control election information to the American public conduct an exit poll SPECIFICALLY DESIGNED to verify this election

We live a capitalistic society with customers and vendors. The customer (MSM) went to the vendor (Mitofsky) and asked him to conduct an exit poll for them. MSM told Mitofsky what they would like the exit poll to do for them. Mitofsky designed and implemented the exit poll per customer specifications. MSM paid Mitofsky the agreed price, and Mitofsky delivered the service. So, your question boils down to this: "Why didn't MSM include election verification when they specified what they wanted the exit poll to be able to do?" Money.

[mgr]

119. I believe there was an effort to address HAVA shortcomings

2003 Dems attempted a HAVA reform that I believe never came up for a vote. This was mentioned in passing with the electoral vote challenge. Not sure of the specifics, but you can guess the legislations may come up again. Push for what you think is necessary with your representative

With the exception of Cal, the other studies rely upon the same supposition as TIA's analysis, that a sample given certain assumptions should characterize a population. Of course, that key assumption is with the experimental design or methodology, which we know with exit polls in the US are not for election verification--unless you take the very ascientific position that the experiment's designer is a liar. The Cal study is specific to Florida, and requires some verification as to Jewish voting patterns to be accepted. Having lived in Jewish neighborhoods, and looked at voting for congresspersons representing these neighborhoods from Los Angeles, I have my doubts that Bush's support of Israel caused any significant voter defections.

I have put forth the argument that comparison should be from the population to the sample, and appropriately compared. If the sample's results can be encompassed with in the larger population without too much violence, these comparisons are normally known as goodness of fit tests. If the sample is discrepant, then I will accept that exit polling reflects widespread voting irregularities.

Mike

[garybeck]

102. and the purpose or recounts

is to make them the same as the original count just to show everyone that the first count was right

[deacon2]

109. Thanks TIA (and anaxarchos)

For laying out the case and enduring the circular arguments: if you point to the data, they impugn the methodology. When you dismiss that, they impugn your character and qualifications. Failing to move the discussion away from the data with this canard, they dismiss the entire discussion as unworthy. "Everyone but you knows that X equals..." Fine. Exit polls are just to make Judy Woodruff look intelligent and to better market politispeak to po' folk in the Ozarks. So let's try this - the majority of the votes counted in the U.S. on machines with unauditable secret code funded, designed, owned and operated by GOP heavyweights = the big MO - Motive and Opportunity. Motive and opportunity. Motive and opportunity. Go ahead, try it... Motive and opportunity. And once you get your brain around that, come on back and we'll revisit the rest of the equation: a freaking avalanche of "questionable" data that has moved numerous learned mathematicians here at this site and out in the real world to cry foul.

But hey, it's only our democracy.

[TruthIsAll]

116. Means, motive, opportunity. The ultimate TriFecta.

It's a pleasure to read a sensible post.
A diamond in the rough.

[anaxarchos]

138. Add "the will"...

Embarrassed about Katherine Harris? Hell no. Run her for Senator. Embarrassed about Blackwell? Hell no. Run him for Governor.
Care about election fraud? Hell no, except in Washington state.

Motive, means, opportunity, and intent.

Hiya deacon...

[bruised]

112. Dont ban exit polls - ban cooking exit polls

So, truthisall, you now want to get rid of exit polls after all that work you did on the data. The real truth is that in that data somewhere is still a lot of evidence but we haven't been able to find the smoking gun yet.
If we had the raw data and knew which precincts had been polled we could compare with the tabulated results in those precincts - and
I suspect there would be some pretty strong evidence there.
That's why no-one has released the raw data yet.

In the meantime we have now put lots more original exit poll data on www.exitpollz.org. (all but 4 states).
And it's not a smoking gun - but the Miami data extracted from the full Florida poll IS STONGER THAN ANYTHING ELSE that has been found yet.
But this will be a long haul but I am sure that those who keep on digging either the exit polls or other stuff like the turnout figures, will eventually build up a case so strong that something will come out of all this work .
It may be a banning of machines that don't leave a paper trail.
It may mean a weakening Bush's mandate - who know it might even mean more. But ban exit polls - NO Ban cooking exit polls!

[TruthIsAll]

115. I thought the sarcasm of my post was obvious. n/t

.

[bruised]

156. sarcasm

You're right,the sarcasm was obvious - don't know how I missed it.

[garybeck]

113. I got it,,, they should just not have elections... it's a waste of money

think of all the taxpayer money spent on elections.

they could take all that money and make the tax cuts even bigger instead.

Bush knows what's best for us anyway, so why don't they just let him pick the next president and save us all some money

[valis]

123. Right now I only believe what William Pitt says!!!

He's the only one who's been consistent in this mess...

[seaclyr]

127. Exi t poll disclaimer: FOR ENTERTAINMENT PURPOSES ONLY

Any connection to any electoral event, past or present, is purely coincidental.

[RaulVB]

131. Ha, ha, ha (n/t)

[valis]

136. The numbers depicted in this poll are purely fictional

[bleever]

137. Very nice!

Should they add disclaimers about no gambling, and you can't be related to the president of the corporation or any of its employees, and that no liability pertains to the corporation, and that these polls can only be used with the express written consent of Major League Baseball??

WTF?

[Igel]

159. OK, I'll dissect one assumption.

Since nobody on the "exit polls are descended from heaven" bothers to look at their low-level assumptions.... this'll get messy.

Let's assume that the exit polls says 98% of green-eyed glasses-wearing republican left-handed blondes voted for Kerry, with an error of 1.0%. (the actual percentages and categories are unimportant for this discussion)

I think TIA would say this represents the nationwide voting pattern. What do we have to assume or assert for this to be true?

This entails one of two assumptions, **if the numbers are completely unadjusted**, one patently false and the other highly improbable. For the time being, I'll assume the exit poll numbers are completely unadjusted, since that's so many people's holy grail.

The 98% +/1 1% error, can be read as: To a high probability (95/100 times, if it's at the 95th confidence level) the actual percentage of such women that voted for Kerry is 98% +/- 1% **for the set that was sampled**. The number doesn't necessarily predict anything about any other set that might have been sampled, so we can't assume that it does.

To say that this number holds nationwide, for the entire set of actual voters, means that it was the full set of voters that was sampled, and randomly so. Which means that Mitofsky didn't pick any precincts nonrandomly, that the precinct for each voter interviewed was selected randomly--so that the chance of a voter from any given precinct in the country (Maine, Hawaii, Ohio ...) being interviewed was the same as for any other voter in any other precinct. After all, that is what random has to mean.

But this is counterfactual. We know that Mitofsky randomly picked some precincts, picked other precincts because he considered them "key", and, furthermore, that he counted numerous voters from each those precincts chosen (let's say 10 voters from each of 1000 precincts, and none from the other 49k or so precincts is highly unlikely to occur randomly). Within each precinct, the voters were chosen randomly with respect to the categories under discussion--he chose every nth voter, without regard to eyes color, ethnicity, party affiliation, or any other variable he was investigating. We conclude that the voters interviewed weren't sampled randomly from the entire set of actual voters, but were randomly chosen from a non-random sampling of precincts. The first possible assumption isn't true.

The only other way to say that the nationwide (*unadjusted*) exit poll numbers reflect the set of all actual voters is to assert that the sample of precincts was, with a vanishingly small error, identical to the nationwide set of voters. The same percentage of Asians, dems, evangelicals, college grads, 20-30 year olds, whites, etc. I can't even begin to think how to successfully pick such a set of precincts, even given complete foreknowledge of how they would vote. Such a set (given a preselected number of precincts) may actually be impossible, if you want vanishingly small error. Doing this *ahead* of time, without foreknowledge ... yikes. It's wildly unlikely that Mitofsky managed to select precisely a set of 1000 (or however many he chose) precincts that matched in all respects the nationwide set of actual voters.

So there's no way to realistically claim that the unadjusted numbers clearly say anything meaningful about the set of nationwide voters; therefore I conclude the unadjusted numbers by themselves would be meaningless for detecting nationwide fraud (although they'd be useful to compare against manual recounts in the precincts sampled). Still, Mitofsky had the temerity to cite a 1% error. What could that possibly mean (still assuming the numbers are unweighted)?

Well, can we stick with the most obvious way of interpreting it: "the actual percentage of people in each category lies within 1% of the number cited, *for the set of voters voting in precincts chosen for sampling.* But who would care about the exit polls if this were the case? That particular subset of precincts just isn't necessarily interesting enough. It *might* turn out to have been perfectly representative, but that's highly unlikely.

What relation that set of voters has to the nationwide set of voters--our real interest--is unknowable a priori. It may be close; it may not be. You can make estimates, given enough historical and demographic information about the precincts you're sampling, concerning the relationship of the voters you sampled to the superset of all voters; you might even try to make your precinct sample come as close as possible to the previous election. But that's all you can do, because *any* deviation from the historical pattern for any reason is going to render your set of estimates (i.e., "model") inaccurate. Mitofsky's deductions/assumptions are a mystery; I don't think he's ever released detailed info how he maps his non-random sample to the full set of voters, and he probably won't (unless he retires and has nobody to carry on for him). In any event, the numbers probably change with each election (based on the previous election).

We know the numbers we keep seeing cited (early, late, final, whatever) *were* adjusted, and so the error can't have the simplest interpretation given above: Mitofsky claims they *predict* something about the nationwide superset of voters. While Mitofsky has percentages and errors based solely on the sampling, he also has to include any predicted errors in each term of his mapping function. The error he cites has to include not only error in sampling, but also errors he's been able to *calculate* for each category in his model. And we don't know those, either. Again, his *calculated*, predicted errors are based on historical voting patterns: if 20% more blacks turn out (nationwide) than expected, or 15% fewer evangelics, or 5% more green-eyed blondes ... or if a lot of Latinos vote in a previously 100% white precinct ... then his predicted errors are wrong, and his figures become inaccurate. How inaccurate? I dunno. I can't know. He knows this, and says *don't cite them*. But they get cited anyway.

To claim the nonsensical 98% figure (+/- 1%) above means something, I have to implicitly believe that how the figures for a sample of voters from a sample of precincts were translated into nationwide figures is very, very accurate. But we know there were surges in some sections of the population and some parts of the country, so we know his tranformation algorithm was wrong. I'd expect his mapping functions to have the most error where people turned out at variance with historical patterns: where they were highly motivated to turn out, maybe where there was a lot of advertising or or other aspects of large-scale gotv drives: that there should be a correlation between his actual error intervals and deviations from historical voter turnout. This assumes he based his model entirely on historical data, and that there was nothing funny with the specific precincts he chose (it may assume other things that I'm not aware of): he may have tried to factor in predictions, in which case turnout different from what he assumed is what's crucial. And then my hypothesis crashes, since it's completely untestable unless he releases what he assumed turnout would be.

If you have non-respondents, your error rates become problematic, as well (since you can't predict non-randomness, but people certainly act non-randomly), and when you resolve non-respondents you introduce even more error. I think he waits until voting is over to resolve non-respondents (but don't hold me to that; if true, it means you can expect shifts in the exit poll projects after the polls close). He can keep adjusting his model by having some sort of feedback between the exit poll data he receives from the field and his model, but ultimately he has to wait for the final turnout figures for at least his sampled precincts, if not all precincts, to come in. Remember: he has no data whatsoever from the 98% (or whatever the actual %age is) of the precincts he *hasn't* sampled.

Mitofsky's early exit polls are based on raw data from a limited, non-random set of precincts, with the figures from that sampling unknowably related to the superset of all voters. Of that there is no doubt. To make that non-random subset of voters say anything about the superset of all actual voters, he had to devise a function that he believed would relate the non-random sampling to the full set of voters. I find it hard to believe that Mitofsky can be dead-on, 100% accurate about that (when he's never been accurate about that before: we keep seeing his *final* numbers cited for past elections, and even those are too frequently wrong), especially when he implicitly says *he* doesn't believe his model (why would he keep updating his numbers, otherwise?). I find it almost easier to believe Bush will convert Kerry to Islam, find a loophole in the constitution so that Congress can appoint Kerry to the presidency, and then step down from the presidency (convincing Cheney to join him) in order to get sex change operations so they can join a Buddhist convent. I could start a "I believe" thread on that topic, but won't (since I don't really believe it, and wouldn't want it to happen: pity the Buddhists.) Once probabilities become vanishingly small, I stop distinguishing between them.

In short:

Mitofsky's early numbers are based on a random sampling of voters from a rather small *non-random* sampling of precincts. The unadjusted numbers are valid for those precincts, and we have no basis in fact or even logic (just in religious mysticism) to say they have any given relationship to the overall voting pool. To get those numbers to say anything meaningful about the overall set of voters, including those that hadn't yet voted when the numbers were crunched, requires a number of adjustments based on a set of a priori assumptions that we don't know, and may never know, but which we can be pretty sure are at least partly in error. To say that it's somehow unethical to adjust those assumptions based on actual numbers implies an unwarranted faith that Mitofsky's initial assumptions, pre-election day assumptions, are a better reflection of reality than any other number (note that even just incorporating actual turnout by precinct would be an improvement). Again it borders on religious mysticism.

So the evergreen thread "I Believe" is accurately titled, at least.

If I've screwed up assumptions or interpretations, I hope somebody better versed in stats will let me know.

[KerryDownUnder]

170. If you had told me there would be a summary at the end of your post

I could have saved myself the headache I now have.

;)

Thanks for the effort, but I think the whole issue of exit polls is going to remain one of faith for many who want to believe.