Mathematical Proof: TIA is Wrong (Part 2)

[Nederland]

Part I is here:

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x342146

To recap, the conclusion of Part I was this:

Regardless of your sample size, if your sample is not representative of the total population, exit poll results can be outside the MOE.

So now let's look at how we determine whether or not the sample is representative. Recall our example election:


Precinct ___ C1 ___ C2 ___ E1 __ E2
----------- ---- . ----
Precinct #1 6500 _ 3500
Precinct #2 6000 _ 2000
Precinct #3 6500 _ 3500
Precinct #4 6000 _ 6000 _ 592 _ 608
Precinct #5 5500 _ 4500
Precinct #6 5500 _ 4500

Totals ___ 36000__24000__ 592 _ 608
Percent ____ 60% __ 40%__49.4% 50.6%


The sample is clearly representative of Precinct #4. We know this because when we compare the sample results with the recorded result we get an error that is within the MOE.

Given a sample size of 1200, MOE = 2.88%.

Now the actual error was:

E = C1 percent - E1 percent
E = 50% - 49.4%
E = 0.6%

Which is within the MOE. Therefore we have a representative sample within precinct #4. Outside precinct #4 however, we have a problem.

The actual error there was:

E = C1 percent - E1 percent
E = 60% - 49.4%
E = 10.6%

Which is outside the MOE. Therefore we have a sample that is not representative.

Now lets add fraud into the mix. Again, keeping things simple, we will assume the fraud was accomplished simply by adding F votes to one candidate. We could choose separate fraud variables (F1, F2...) for each precinct and choose other variables to represent votes subtracted from the other candidate but the results would be the same.

The recorded vote looks like this:

C1: 36000
C2: 24000

The intended vote looks like this:

C1: 36000-F
C2: 24000

And C1 percentage of the vote looks like this:

C1 percent = ((36000-F) / (36000+24000-F))

Now when we try to determine if the sample is representative and within the margin of error the calculation looks like this:

E = C1 percent - E1 percent
E = ((36000-F) / (36000+24000-F)) - 49.4%

As you can see, you cannot calculate the error. It is dependent on F, which is unknown. In a perfect election F would be zero, but we all know that this election was far from perfect. Given that F is unknown, we cannot calculate E and cannot determine whether or not the sample was representative.

Which leads us to the conclusion of Part II:

When fraud occurs, it is impossible to determine whether or not an exit poll sample was representative.

[skids]

1. Gee, one would think...

That when your premise in part 1 was disproven, you wouldn't ignore that fact and turn around and use it as the basis for part 2.

[Nederland]

2. Let me get this straight

You believe the following statement is incorrect?

Regardless of your sample size, if your sample is not representative of the total population, exit poll results can be outside the MOE.

Funny, no one in Part I said it was incorrect. Care to point out the post# were they did?

[skids]

3. Mathematical proof that the statement is incorrect.

Let A represent the statement "your sample is representative of the total population".

Let B regresent the statement "exit poll results can be outside the MOE"

Since the case of B being true when A is true is outside the scope of your argument, which uses an error term based on the premise that A is false, it is portrayed logically as:

B iff not A

...which can be rewritten as...

A -> ~B

However, from the 2004 NEP report, we know that A is true, so substituting:

TRUE -> ~B

Therefore, or as we say in math:

___:cry:

:cry: :cry:

B is FALSE, rewriting into english:


"exit poll results can be outside the MOE" is FALSE

or

"exit poll results cannot be outside the MOE"

(EDIT: about part 1, it was me, in a post labeled "Wrong." Which, if I recall correctly, is mostly synonymous with "incorrect")

[Nederland]

5. Missing the point

This is the statement:

if your sample is not representative of the total population, exit poll results can be outside the MOE.

A = if your sample is not representative of the total population
B = exit poll results can be outside the MOE

If A Then B

Whether or not you believe A is irrelevant, the statement is correct.

That's just basic logic.

[skids]

7. Oh oh, am I "Off topic?"

No sir. You cannot escape logical consequences by dividing an argument into two halves. Mathematics is based on the process of logical induction. If you proceed to the next step based on faulty premises, then the remainder of your work is without basis and thus is, at best, and excercise in hypothetical conjecture.

Oh, by the way, have you read the NEP report yet? I'm going to keep asking until you answer, you know.

[Nederland]

12. I have read it

It's here if you like to read it yourself:

http://www.vote.caltech.edu/media/documents/EvaluationJan192005.pdf

[skids]

13. If you have been on this forum for more than two weeks.

You should know that I have read that report, and more.

[Nederland]

17. If you were familar with the report

...you would know that the NEP applied weighting to the samples.

Question: If the NEP believed the samples were representative, why would they weight them?

[skids]

23. NEP never claimed each precinct to be a bellweather.

NEP selected a set of precincts and calculated prior weights before conducting the poll, based on their past performance.

These weights were already decided before the poll took place.

After the vote results were in -- the ones reported by the elections committees -- they plugged those numbers back into their preselected precinct weights to doublecheck them.

They checked out. The precincts were actually biased about one half percentage point towards Bush, which only makes the discrepancy worse.

[Nederland]

24. You didn't answer the question

If the sample was representative, why would they weight it? The fact that the weights were calculated a head of time is irrelevant--the NEP still believes that weighting was necessary, therefore indicating that they believed their samples were not representative.

In any event, this entire discussion is moot. The point of this thread is this:

If fraud occurs, it is impossible to determine whether or not an exit poll sample was representative.

[skids]

25. Here's the point.

The sample was chosen by professionals to be representative on the level that it was required to: state and national. There are numerous other ways to show this.

Your attempt at a "mathematical proof" is of course doomed to failure, because in order to reduce this to a mathematical argument, you have to ignore all the externals factors. So are any "mathematical proofs" offered to prove the opposite doomed to failure.

You can only show that the most rational conclusion is X or not X, and this is not a purely mathematical task. Noone can prove anything. Rarely can anyone prove anything in the real world, which is why proof is not required in a court of law. Even in murder cases, "proof" is qualified with "beyond a reasonable doubt" and as such cannot be interperated in the logical pure sense of the word.

[Nederland]

26. Missing the point

The sample was chosen by professionals to be representative on the level that it was required to: state and national. There are numerous other ways to show this.

Yes, there are numerous ways to show this, and they all involve the assumption that the recorded vote is accurate. Just read the NEP report. Everything in it--the calculations of WPE, statistical bias, gender adjustment--everything assumes that the recorded vote is accurate. If you believe that fraud occurred, then that entire report is meaningless because it is based upon a false assumption.

[skids]

27. Not all do.

Demographics, for example, do not.

WPE's report does not go into detail in this area, unfortunately.

[Nederland]

29. Fair enough

You can get determine whether or not the sample was Demographically representative. You have to admit however, that knowing that is not sufficient to declare that the entire sample is representative enough to correctly predict the outcome within the MOE. Hell, all you have to do is oversample Kerry voters by 6% and the poll will fall exactly where it did. That is trivially easy to do, but impossible to prove if fraud occurred.

[skids]

30. Except that's not what happened.

No indicators showed a biased undersampling, other than demographic groups that were anticipated and for which mechanisms were utilized to correct. In fact the NEP report gave pretty good reason to believe that there was no bias, or in fact bias opposite to the "reluctant bush voter" theory. Then they did a complete about-face and asserted the "reluctant bush voter" theory contrary to their own evidence.

Anyway, I'm told that there will be more analysis by people more qualified than either of us in a few days.

And it's starting to get to be TV time. So have a good night :-)

[Nederland]

70. Question

Did the NEP report assume that the recorded vote was correct?

Please answer yes or no.

[kobeisguilty]

32. If it were that simple

then why didn't Mitofsky simply conclude that the sample was not representative? And while the exit polls skewed towards Kerry, why did the exit polls accurately predict total amount of voters, or voter turnout?

[Nederland]

69. He did

He concluded that the exit poll sample over represented Kerry voters and under represented Bush voters.

[johnaries]

36. That is not a correct statement.

Exit polling and Balloting are both data sampling. If the difference between the two are outside the accepted margin of error and confidence levels, that demonstrates that one of the 2 methods is flawed.

At that point you must look at both methods of data sampling. You also have to consider the Potential Points of Failure. The explanations that have been presented for errors in the exit polling simply do not pass the "smell" test. However, there are many potential points of failure in the Balloting tabulation process, and there is substantial documentation that these PPOF's had an adverse affect on the results.

Although, again, we get into the question of semantics. This is not technically "proof". However, it is more than enough to make a logical determination.

So, your statement, "it is impossible to DETERMINE whether or not an exit poll sample was representative" (emphasis added) is incorrect.

Nederland, I'm not sure what you're trying to "prove", but at this point your best option is to "drop back five and punt".
:hi:

[johnaries]

37. An example of the "smell test".

Since I mentioned the "Smell Test" in my last post, I felt I should give a couple of pertinent examples.

Mitofsky said that one of the reasons the Exit Poll sampling was wrong was that they "over-sampled" Soccer Moms who tended to lean towards Kerry. Yet, one of the explanations of why Bush won over Kerry despite earlier polls was that Soccer Moms switched to Bush for "National Security" reasons.

Does that "smell right" to you?

Another reason cited was the "reluctant responder" factor, "assuming" that Bush voters were more reluctant than Kerry voters to participate in exit polls. However, it was demonstrated that exit poll participation was actually HIGHER in predominately Bush precincts than in predominately Kerry precincts.

Does that "smell right" to you?

[kobeisguilty]

38. can you quote the soccer mom discrepancy

from the Mitofsky report?

[johnaries]

39. Sure! Give me a minute... let me "get my Google on" n/t

[johnaries]

40. Sorry it's taking me so long, I'm having trouble with the

animals that live with me (dog - I want in, no, I want out, now I want in... cats - feed me, feed me, I know I just ate, feed me!). And, Google is not ALWAYS "your friend". Anybody else remember the Mitofsky "soccer mom" defense? Help me out!

I'll find it, it just may be later than sooner.....

[TruthIsAll]

59. You just don't get it, do you?

The preliminary exit poll sample WAS representative by design - and it showed that Kerry won. That was a real problem which had to be addressed quickly, for the poll weightings were at variance with the recorded votes.

So they had to re-weight the preliminary 13047 sample exit poll to the the final 13660 in order to match the bogus vote counts.

You have already conceded that the vote count did not reflect voter intent due to ballot spoilage. Why do you ignore the facts you are already aware of?

You are contradicting yourself left and right.

[Igel]

72. So what you're saying is that Mitofsky picked samples

that not only would be representative, but were representative; but the only way he could know that is if he new how they would vote.

This means either Mitofsky was in on how all his precincts were fixed, or he's God (assuming you think God and only God is omniscient or prescient, otherwise fill in the omniscient/prescient being of your choice).

His model assumed a known input, and weighted what are inevitably a non-representative sample to make it representative. If his input was wrong, or his model was wrong, he'd get wrong numbers.

And I still don't know when he "filled in" for the non-respondents.

[L. Coyote]

56. Isn't the weighting done after the respondents are known?

If 75 % of respondents are women and 25% are men, etc. etc., obviously weighting is required to make the sampling representational of the "expected" voter population (ie. more like 50/50).

One problem is that there is no way of determining if the "turnout population" matched the "expected voter population." If 75% of wonem turn out to vote and 50% of men turnout, contrary to an expectation that their turnout would be equal, weighting according to expectation will create errors (given the population subsets used for weighting voted differently).

Polling is inferential statistics. Margin of error is not a simple equation of numbers (sample size). For exit polling, it first requires that the sampling (or, in the absence thereof, the weighting) be representative of the "turnout population."

Bypassing this known statistical premise results in an argument based on sheer numbers to determine MOE. Such work has an underlying assumption that the sample is reprresentational, negating the utility of its conclusions. Ignore the sample size MOE pronouncements. Investigate the representational valitity of the sampling first. Then make a case.

[skids]

66. Here's how it works as far as I know.

First they decide on the precincts. They pick a set of precincts that is more or less representative of the larger area which they wish to poll. This is done by looking at past results, voter registrations, and other historical data. My guess is the size of bellweather groups was the size of their statewide "geostrata" regions.

Each precinct is not itself a bellweather. Groups of precincts are bellweathers when considered in the aggregate. That is to say that they will go into a precinct that is 90% black in order to make sure that there are enough blacks in the survey, for example, but the set of precincts that contains that precinct will also have a whiter-than-usual precinct in it so that the group of precincts is either representative, or if they don't do that, they at least know beforehand that it will need to be weighted and by how much each precinct will need to be weighted. They may also have some demographic safeguards that apply weighting to keep early results from being sensitive to selection bias, like weighting based on refusal and completion rates in a demographic group.

That's because the exit polls are interested in measuring turnout as much as they are in measuring the way people vote. They want to know for example, if more asians decided to vote this year, compared to the number registered, than did last time.

They never released raw results until a month or so ago. They always apply the predetermined weights before releasing results. In addition they add weights based on an estimation of their interviewee selection bias, and these weights are updated with each release of the preliminary results. So the next-to-last results should already have been adjusted to some extent to account for that.

The initial predetermined weighting is what they fed through to discover that the selection of the precincts themselves were half a percentage point biased towards Bush. The preliminary results released election day are scaled by this weighting and also by any weights they determine are needed to correct for refusal/completion variance among demographic groups. That means specifically that the fact that old people had a high refusal rate should have allowed them to correct for this fact, and the pre-final results should not be biased towards Kerry based on old people (who vote more republican in most places) refusing to take the survey.

Then the actual vote results are collected and the weights are twiddled to find the least drastic variance which allows the exit polls to match the final vote count. That means the final results are only meant to be used to look at turnout and "fun facts" based on demographics.

Anyway, the fact that they had built-in a way to prevent grumpy old codgers from spoiling the results is why they had to go with the "selective non-response" hypothesis. The precautions would have caught it if old or young, black or white, male or female, etc had not been feeling too talkative. The only way they could stick their heads in the sand was to hypothesise that across all races and ethnic groups, across all wealth classes, across both genders, among all ages, all people behaved the same, and in each of these groups it was a universal constant that people that voted for Bush refused to take the survey more often -- that that one trait about a person, whether they voted for Bush or not, was the primary correlating factor in how willing they were to take the survey.

[Time for change]

71. Actually, the M-E report admits

that their poll oversampled Republican precincts.

[skids]

10. Oh the statement is just a statement, true.

Except that we know A is false, so the statement has no meaning, and you are welcome to proceed to ramble on about "what would happen if A were true" but that it just science fiction. If anyone wants that, they have only a few more hours to wait before SG1.

[Nederland]

15. Logic 101

You obviously need a lesson in logic.

Let's take the following statement:

If X is pregnant, X is female.

(Note: Please don't nitpick by coming up with some obscure biological exception, just run with the concept.)


Now, the statement above is true.

It is true if X is male.

It is true if X is female.

It doesn't matter what X is, the statement is always true.

That is just basic logic. Learn to live with it.

[skids]

16. Your condescending attitude doesn't get you anywhere.

I could just as easily say that perhaps you need a class in reading comprehesion, or an eye examination, since you plainly missed the context I proferred.

If you are going to tell me I need a lesson in logic, perhaps you should state your own credidentials on that matter... I'm waiting...

[Nederland]

18. I have a BA in Philosophy

What are your credentials?

[kobeisguilty]

19. stop

both of you.

[Nederland]

21. You're right

This is stupid. I apologize for wasting space.

[skids]

22. He threw the first punch.

Look, I don't care if people want to assail TIA's "odds" calculations.

However, if they insist on using unfair and disengenuous debate tactics in doing so, they will be responded to in kind.

[skids]

20. BE CSE

Logic is the core of my profession.

[johnaries]

42. BA? and you're PROUD of that? I chew you guys up and

spit you out on a regular basis! BTW, I never got my degree. However, I do believe in Continuous Education. I receive certification in a new area every 1 1/2 years, on average.

But that explains a lot, like why you're so anxious to continue on a doomed premise.

Sorry, kid. You've got a LOT to learn!

[kobeisguilty]

43. i'm on my way to finishing a BA in Philosophy

chew me up and spit me out.

[johnaries]

46. You haven't given me reason to.

So far, your logic has been pretty solid. Which may or may not have anything to do with your proximity to a BA. People who think that their opinions should be respected simply because they have a BA without regard to the merits of the opinion itself...

Ummm, Lunchtime!

[Name removed]

61. Deleted message

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

[berniew1]

45. I likely have more advanced degrees from more Univ. than anyone

on list and in Statistics, Applied Math, and Demograpy to go along with my Engineering UG and work experience. I've been Univ. prof. at several Univ. teaching stats and logic. But that alone doesn't make any argument more valid than someone else. But I seldom make logical errors. But data is also an issue. I'll try to look at this tomorrow. Too late for tonight.

[kobeisguilty]

48. i would have jumped on you

if it weren't for the "that alone doesn't make any argument more valid than someone else" part. Glad you understand that. Degrees don't mean shit in themselves, its about how the person utilizes that knowledge. Looking forward to your contributions.

[L. Coyote]

57. Read Thomas Kune on the Structure of Scientific Revolutions

Kuhn, Thomas S. 1996 The Structure of Scientific Revolutions. The University of Chicago Press.

Degrees can mean one is fully acculturated in false paradigms!!

What may be needed is a paradigm shift. This tome may be more useful than all the TIA rantings put together. What will it take to create a paradigm shift?

[Chi]

54. I know how to ask "ju'want fries with that' any credit for that 8) n/t

[kobeisguilty]

6. just for the record

Nederland is right, although a more accurate way to say this is that exit polls utilize clustered sampling ( http://en.wikipedia.org/wiki/Cluster_sampling ) which increases a margin of error somewhere between 30-80% of the original MOE calculation.

[skids]

8. Nice, but...

That's not what he is stating. To borrow a phrase from another DU poster, who's handle is mgr, you're off topic.

[Nederland]

9. Thanks

That is a much more accurate way of saying it. And yes, that is precisely what I am saying.

[skids]

11. No it isn't.

It has no bearing on what you are saying at all, now if you would like to talk about clustered sampling, do so, but your original post does not.

[Nederland]

14. Ok

Perhaps I didn't explain it correctly. Just so we are clear now, I agree with the statement by kobeisguilty above.

[Jack Rabbit]

31. Actually

Regardless of your sample size, if your sample is representative of the total population, exit poll results can be outside the MOE (note the affirmative antecedent) is also a true statement. It's just less likely to be. We are dealing with statistics and probability, not categorical statements.

Having missed Part I, I'll stop there.

[johnaries]

35. Go back and look at my posts.

You asked me to "prove it" with my own figures, and I simply posted a thread where a MIT professor had already done it.

OK, I'll state it simply: your statement condradicts itself - it is not true.

"Regardless of your sample size, if your sample is not representative of the total population" - it your sample size is 100%, then by definition it is representative of the total population. Of course, only the actual ballot count is "supposed" to be a 100% sample. But, as we know from numerous reports of "irregularities" this is not accurate.

I go back to my original statement of "randomness". Randomness is the key. Sure, if you "cherry-pick" your smaller samples are less random. However, as you increase your sample size randomness increases, simply because it's more difficult to "cherry-pick".

Eventually, you hit the point of diminishing returns where you have to increase your sample size by very large amounts in order to increase the randomness and decrease the margin of error by very small amounts.

So, if you are waiting for someone to say quote, "your statement, 'Regardless of your sample size, if your sample is not representative of the total population' is not correct", I'll say it. Consider it said.

I know you have a whole presentation ready that you want to present, and you're chomping at the bit to give your presentation. It really sucks when someone points out that your basic assumption is wrong and therefore all the hard work that you are so proud of goes right down the tubes.

Trust me, I've been there. I learned to accept it, and I made sure my assumptions were better in the future.

[L. Coyote]

58. "...as you increase... sample size randomness increases.." False

if you do not poll a more representative sample of the population.

In the case of exit polls, the sampling is of selected precincts. Asking more people how they voted at those precincts might not change the results.

[rzemanfl]

4. Well then, tell this to all the dumbshits in the Ukraine and all the

BushCo folks who agreed with them. I have driven a car for years, if someone steals my car, it doesn't mean I don't know how to drive. People have been picking representative samples for years and know how to do it.

[Joe Chi Minh]

33. They must be really

beginning to panic now about the statistical improbabilities.

TIA and others have ben manfully persevering with these threads, to keep them in the public consciousness - knowing that the neocons have been playing for time - and here they've come back swinging (albeit at the air). Great stuff!

[kobeisguilty]

28. True, but...

One must always accept that there are certain unknown variables which affect the vote count. For instance, in every election, some voters are dissuaded from voting, others register but are too lazy, and many other unknown factors which affect the final voting tallies, thus in addition to falsely recorded votes, there are also non-recorded would-be votes that affect this total as well. The crux of your argument rests on that if the calculation for poll error (which is different from representativeness, by the way) contains variables which cannot be solved, the poll error thus cannot be determined if these unknowns are present. However, this would invalidate every statistical survey known to man. All statistical surveys have unknown factors that affect the final tally, which we cannot account for. Mitofsky himself says that the Reluctant Bush Responder Theory is "impossible to quantify". Statistics is the science of correlation within definition of limits... we set these limits and adjust for possible unknown factors which we may never quantitatively know. This fact, however, does not invalidate statistics as a whole.

In addition, your example is flawed as only one highly dissimilar precinct is polled. Remember, these people have access to voter trends on a precinct level for the last 20 years in addition to voter registrations to predict voter turnout for 2004. More likely, they would pick precincts a LOT more representative then your example of Precinct #4, which is an extreme outlier from the other precincts you mention, or at the very least use an outlier favoring the opposite candidate, thus producing results closer to the vote tallies.


I conclude:

C1: The mere inability to quantitatively define unknown variables does not invalidate the process of statistical correlation.

Thus,

C2: Fraud does not corrupt exit poll sample representativeness to the point where they cannot be used to determine indicate such an event occuring.

[understandinglife]

47. Exquisite. Thank you. (nt)

Peace.

[anaxarchos]

34. With all due respect...

... you have just disproven the whole of social science.

[Dancing_Dave]

41. You have got to be kidding!

What a silly little shell game with hypothetical numbers!

What you overlook is that the representativeness of samples is not determined in relation to one election, the sampling methods have developed over time and been cross-checked and refined to where they have developed a track record of accuracy.

It is true that we should be allowed to know more about exactly what the sampling methods were--we MUST in future elections hire some open, honest, methodical pollers who are willing to reveal how they arrived at their numbers...just as all real social scientists must do to prove their salt to their colleagues.

[jkd]

50. Exit Polls not appropriate to judge accuracy of vote.

Mitofsky says that his state samples average 4% MOE and that the national sample is about 3%. The final weightings, which use information from the current election results, adjust the margin of error for the over-all national sample to about 1%. If the NEP’s methodology is so accurate, then why is the MOE so high?

The NEP model was basically created from vote results. Now, some would file a paternity suit. The exit polls may or may not accurately reflect the voter’s intentions, but the polls and the results are too closely related to be a check against the other.

http://www.exit-poll.net/faq.html#a15
http://www.washingtonpost.com/wp-srv/politics/elections/2004/graphics/exitPolls_national.html

[kobeisguilty]

51. you're using "corrected" exit poll data

http://www.exitpollz.org

Use the 12:23am release, as it is the last update prior to correcting against actual vote tallies.

[jkd]

52. yes

Then, they use that corrected data to tweak the exit polls for the next election. It's part of the same process. The polls and the results become intertwined.

[kobeisguilty]

53. exactly

could not be a more clear-cut example of circular reasoning.

[Nederland]

67. You don't have another choice

unless you'd like to offer up one...

[L. Coyote]

60. Right On. Bulls Eye. Admirable.

Academic standards need to be applied to the problem. Peer-review requires openness about methods. What is being hidden and why?

Give the facts to everyoine and let the review begin. This is where pressure needs to be exerted. Let's crawl out of the Dark Age of political exit polling into the light of openness.

The most poignant comment in this thread by far, Dave. Hooray for you. Your donkey has real kick!

[berniew1]

44. Where do any reliable statistics come from on how many R s & Ds

& Is and women and men voted?? Who has that data and what does it say by state?

It doesn't take Exit Poll to know the election was not a serious election and the results were bogus. The case for major fraud and manipulation that swung millions of votes is documented by the case reports from the elections hotline database(EIRS). I've been following elections as a political coordinator(non-partisan) for a long time and this one had the most known fraud and manipulation of any election i've ever followed(25 years of such). Documentation at
http://www.flcv.com/summary.html (by state)
and
http://www.flcv.com/ussumall.html (some not in above)

and http://www.flcv.com/studentv.html student voter suppression

[johnaries]

49. Why are you doing this?

For every argument against your case, you are not applying logic, you are applying your own dogma.

Once you've decided on the outcome, it's easy to make the data predict it.

Nederland, what is the purpose of this thread? Is it just to "prove TIA wrong?"

As for posibilities, it's "possible" that I've been dreaming for the last 5 months and when I wake up Kerry will be president.

Hey, it's just as likely as the assumptions you're trying to base this whole thread on!

[Nederland]

62. Why I am doing this

I am doing this because TIA's methodology is fundamentally flawed. You cannot calculate the MOE simply by plugging the exit poll sample size into a formula. It is important to demonstrate this in an irrefutable way because TIA makes the Democratic Party look like a bunch of idiots that spout on about things that they don't understand. I believe that there was real fraud in this election, but we are not going to find it following TIA's lead.

THAT is why I'm doing this.

[kobeisguilty]

55. no response from nederland :(

[kobeisguilty]

63. Summary of Criticisms of Part II

One must always accept that there are certain unknown variables which affect the vote count. For instance, in every election, some voters are dissuaded from voting, others register but are too lazy, and many other unknown factors which affect the final voting tallies, thus in addition to falsely recorded votes, there are also non-recorded would-be votes that affect this total as well. The crux of your argument rests on that if the calculation for poll error (which is different from representativeness, by the way) contains variables which cannot be solved, the poll error thus cannot be determined if these unknowns are present. However, this would invalidate every statistical survey known to man. All statistical surveys have unknown factors that affect the final tally, which we cannot account for. Mitofsky himself says that the Reluctant Bush Responder Theory is "impossible to quantify". Statistics is the science of correlation within definition of limits... we set these limits and adjust for possible unknown factors which we may never quantitatively know. This fact, however, does not invalidate statistics as a whole.

In addition, your example is flawed as only one highly dissimilar precinct is polled. Remember, these people have access to voter trends on a precinct level for the last 20 years in addition to voter registrations to predict voter turnout for 2004. More likely, they would pick precincts a LOT more representative then your example of Precinct #4, which is an extreme outlier from the other precincts you mention, or at the very least use an outlier favoring the opposite candidate, thus producing results closer to the vote tallies.


I conclude:

C1: The mere inability to quantitatively define unknown variables does not invalidate the process of statistical correlation.

Thus,

C2: Fraud does not corrupt exit poll sample representativeness to the point where they cannot be used to determine indicate such an event occuring.

- me


What a silly little shell game with hypothetical numbers!

What you overlook is that the representativeness of samples is not determined in relation to one election, the sampling methods have developed over time and been cross-checked and refined to where they have developed a track record of accuracy.

It is true that we should be allowed to know more about exactly what the sampling methods were--we MUST in future elections hire some open, honest, methodical pollers who are willing to reveal how they arrived at their numbers...just as all real social scientists must do to prove their salt to their colleagues.

- Dancing Dave


Response, Nederland?

[anaxarchos]

64. Hello, kobeisguilty....

It's a distinct pleasure to make your acquaintance. Welcome to DU...

[kobeisguilty]

65. thank you :)

[Nederland]

68. Question

What you overlook is that the representativeness of samples is not determined in relation to one election, the sampling methods have developed over time and been cross-checked and refined to where they have developed a track record of accuracy.

And how do you "develope a track record of accuracy"? Are you not comparing exit poll results with recorded results? Isn't the recorded vote the benchmark against which the exit poll judged?

[kobeisguilty]

73. He is referring to...

the overall selection of representative precincts, I believe. I also believe that prior exit poll history is more of a benchmark then current vote tallies, as its comparing apples to previous apples rather than apples to oranges.

[Nederland]

74. Ok

So you are comparing the current exit polls to previous exit polls. My question is this: how can you expect to verify that the exit poll samples are representative of the voter population if you never compare them to the voter population?

[kobeisguilty]

75. Of course you do

I was merely clarifying what it meant by a "track record of accuracy"... and that prior exit poll accuracy (as compared to voter turnout) are more important than one election's discrepancy... say if we did not realize that every presidential election since 1988 had a democratic skew, there would be no question that this is fraud.

[anaxarchos]

76. Every metric is a relation...

That's how it begins. Thus a kilogram is the weight of a liter of water. Past this initial derivation, a kilogram has an independent existence, abstracted away from its original relation. A kilogram expressed in iron may well be recalibrated against a liter of water if the measure is doubt. But, equally, an iron kilogram, previously abstracted, may be used to determine whether the weight of a "liter of water" is correct (whether, it is in fact, "water", a "liter", etc.). If both are in doubt, we have recourse to what is known about each, i.e. the next level of analysis.

What happens when a liter of water does not equal a kilogram in iron? Well, whatever happens, the last thing we do is abandon all kilograms because it is no longer "possible to verify" them.

What am I missing according to your logic?

[Nederland]

77. You are missing this

Your analogy would be correct if tracing things back you at some point brought you to the recorded vote count. However, that not what I'm hearing people say. Perhaps I'm not understanding correctly, but my interpretation of people's comments are that you should never compare to the recorded vote count because it is tainted.

From post #75 I sense that kobeisguilty agrees with me on this small point--at some point you must look at the recorded vote.

[kobeisguilty]

78. i do agree with you on that point

I'd rather see you address the issue of dismissing statistical calculations simply due to unknown variables, as per my critique of your post.

[kobeisguilty]

85. .....

[anaxarchos]

79. It's easy to agree with that...

...otherwise I wouldn't have written the above. But if you stretch that argument to suggest that the actual vote must be invoked in EVERY election or the exit poll is unverified/illegitimate, you doom everyone who wants to weigh anything to lugging around liter jugs of water as well as their scale...

[Nederland]

80. Question

If it is unnecessary to reference the actual vote in every election, why do you suppose every exit poll firm out there does it?

[anaxarchos]

81. Because that is what they are hired to do.

Their job is to make the scales balance, either to help call the election (despite all of their disclaimers) or to match the survey questions to the electoral outcome. In truth, Nederland, there is not a direct manipulation of the methodology because the exit polls were "off"; no 1% shift in an "overall fudge factor" because a particular poll was 1% out of MOE. The polling organizations adjust the polls to match outcomes and also adjust their methodology when they believe their sample is not correct but there is no direct relationship between the two (you can argue about indirect connections).

There is an L.A. Times (one of the few papers that still run an independent exit poll) document on this which I will try to find for you.

You must be aware of the controversy around Mitofsky as well. While many on this board are angry with him for not crying "fraud", there are others from the left side of the blogs who claim that by admitting the variance of the polls from the vote count and then not hiding behind "methodology" (i.e. "re-calibration") but instead pointing to the highest WPE in the history of the polls (with very ambiguous supporting evidence), Mitofsky is in fact coming as close to crying "fraud" as his commercial charter allows.

kobeisguilty is correct. The exit polls have a significant life of their own by now.

[berniew1]

82. After the fact to make numbers match; but EP as check on vailidity

of election results can only use the raw data; or as corrected for
the distribution of voters who voted.


But there is a little more to it than that; related to how the exit poll precincts and samples are chosen; and their voting history.
To know the degree with which the sample taken predicts the total population.

[Nederland]

84. Exactly

I'm glad somebody gets it.

[jkd]

86. yes

I am too.

[anaxarchos]

87. I'm glad you guys are glad... we're all glad. n/t

[TruthIsAll]

83. Descending into the nether world of Orwell-think.

"When fraud occurs, it is impossible to determine whether or not an exit poll sample was representative"

Unbelievable.
Banging my head against the wall.
Staring into space in utter disbelief.
Descending into the nether world of Orwell-think.

In my world, I believe that:
When fraud occurs, it is impossible to know....without a representative exit poll sample.