New USCV criticism

[davidgmills]

www.geocities.com/lizzielid/WPEpaper.pdf

Here's what I don't buy right out of the box:

"Assessing how representative a sample of the selected precincts were can be relatively easily ascertained after the election simply by comparing the estimate made by the actual results from the sampled precincts with the actual results from the totality of precincts."

In other words, match the sample of the selected precincts to the fraudulent results. Easy as pie.


And more criticism:

http://www.dailykos.com/story/2005/4/19/81031/0928

[mgr]

1. Missing the point

David:

The quote you cite summarized NEP's methodology for the exit poll. The purpose of the paper and Febble's technique is for the second comparison method, the within precinct WPE, an explicit distinction made by the author. The point of the paper appears that one cannot separate bias from patterns suggestive of fraud--it may weaken USCV's argument, but it may be far more damaging to Mitofski's claim that exit polls may detect fraud. If you read further into the thread at Kos, discussion touches on what may be required to develop a measure that can distinguish between fraud and bias, but it is not followed up on.

As to the ad hominen angle, Febble is a graduate student advanced to candidacy (awaiting a PhD.)in psychology that worked with USCV. This paper is still awaiting formal peer review, so if there are problems with it, it will be detected at the meetings in mid-May.

Mike

[davidgmills]

3. I was asked to post this in a hurry so I did before reading it thoroughly

I still stand by my point and have been told by others that it is a point to be well taken.

She makes the argument that the exit poll is more like loaded dice than flipping a coin. I thought about starting a new thread with this title: flipping a coin vs. loaded dice. Maybe we would get a greater response.

But I concur with the poster below. TIA's new analysis that shows Kerry won not only the national polls but the state polls as well and that the two are within .06 percent of each other, which can only mean one of two things: Mitofsky's bias or election fraud were damn consistent.

How could this "bias" have been so consistent on such a large scale? Reluctant Bush responders the same everywhere? I don't buy it. Did Mytofsky load the dice? I would think that if he truly is the expert people say he is, he did not, at least not intentionally. And if he didn't do it intentionally, was it just dumb luck that the loaded dice favored Kerry? Was it just dumb luck that they have been consistently favoring dems for some time now?

[mgr]

5. But the bias is not consistent

David:

Ignoring what Febble argues, and going just on NEP, the bias is not consistent, only that it is strong and favors Kerry on a 2:1 ratio compared to precincts with strong Bush bias. The key is that these are precincts with a WPE greater than 3%, in other words they exceed the acceptable sample bias necessary to have the design MOE.

These are ordinal categories, not rational. Your argument has merit only if they were rational.

Do I really have to refute TIA's argument when WPE is at the basis of both election estimates? If the same error affects both the national and the state polls, then they should be consonant with each other based upon the error, and not much else. You should have been capable of that critical assessment.

It's the old logic question, if a steel suspension bridge expands and contacts with temperature, and you measure it with a metal ruler, will you know how much it will expand or contract?

Mike

[davidgmills]

7. What metal is the ruler made of?

[davidgmills]

8. A challenge to you Mike

The reason we have so much difficulty seeing your point of view and why it is so easy to see TIA's instead of yours is numbers.

His language is numbers. You, Mystery Pollster, Rick Brady and the like, all use verbal language to say TIA's numbers are wrong. It doesn't compute to even us guys who are verbalists not mathematicians. You say there is bias, well put up the numbers to prove it. Prove the bias mathematically. Do the hard mathematical work. Statistics is a science of numbers.

Don't show me graphs or computer models or pie charts. I want to see the raw numbers that you say prove bias. National level, state level or how ever you want to do it. Until then, I'm believing numbers I can see with my own two eyes. Numbers do that to you. We are conditioned that way.

[mgr]

20. For someone who describes themself as innumerate

You sure put a lot of faith into math. I've already explained to you why you should not. However, logic rests at its foundations; and everything TIA has done lacks that logic. If your premises are unsound, then anything you do is unsound.

But logic cannot persuade you, you now want mathematical computations that you have no expertise to judge? I think you have reached the nadir of critical reasoning and argumentation. What you are asking is that I reinvent NEP's work. I am more interested in rehabilitating it if that is possible. I don't measure sand grains. My work is coming.

Statistics is the science of comparison. I would suggest you pick up a copy of my stand by, Sokel and Rohlf's Biometry, and try to peruse the first three chapters.

Better yet, show Febble's paper to your father.

Mike


Better yet, run Febble's by your father.

[TruthIsAll]

21. As usual, mgr, words, no substance, just words..

And what will you call this seminal work: "The National exit poll: Rehabilitated"?

LOL.
Your pomposity knows no bounds.

The only thing you ever said which I agree with is that you believe fraud occurred in this election.

Let me ask you. How did you arrive at that conclusion?

WHAT WAS YOUR PREMISE?

[MontageOfFreedom]

22. What mgr seems to be implying is impossible....

If this was just a statistic based sampling for the whole nation, I could see the bias easily being affected in the larger cities and the more dense populations.

But this is a statistical bias that is in every single state.....I'm sorry, I can't see how anyone who even knows math could believe that this many democrats across the board voted for a democratic senator, and bush as president. Or if the ticket splitting needs to get even more divergant, a independant senator and bush as president.

Or how about no vote at all for anything but still bush as president? It doesn't make sense by any statistical means you can apply, when there is already this many confirmed undervotes and straight-ticket miscounted votes in the reports. http://www.votersunite.org

It defies the law of gravity itself to inanely believe that this entire oversampling problem, happened everywhere. It also appears to be the only option remaining for republicans to latch onto-- It's either the bush voters were silent, or there was decisive fraud.

I'm going to go ahead and say the obvious, decisive fraud and problematic exit polls across the board.

[mgr]

19. steel

[MontageOfFreedom]

2. I don't see the viewpoint being expressed here as being valid.

Its an opinion certainly, that seems to be well argued. But I don't see many things that even casually back it up.

If the exit poll demographics are still completely plagued by bias, then the state exit polls should not have any solidarity or show the exact same thing eg: as the national exit poll.

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x362208

It doesn't jive, and seems more tinfoil than just the straight exit poll report.

[mgr]

4. You are forgetting the bias affects both

The problem is with how the polls are taken. Presidential preference was asked in the state polls, as well as other ballot issues.

If both samples are affected by the same directional bias, then they will be consonant with each other.

What is interesting is the divergence between senate and presidential being attributed to vote splitting (Brady post)as an artifact of this,a nd being the underlying reason for why the senate exit polls were better estimators than the presidential polls. The vote splitting is interesting in that it suggests a movement to wards Kerry by republicans, which undermines the solidarity of the republicans. (when one hears ticket splitting, one automatically assumes democrats, but it does not work with the simulation).

Mike

[MontageOfFreedom]

6. Clarify: What is your meaning by "divergance"?

"What is interesting is the divergence between senate and presidential being attributed to vote splitting (Brady post)as an artifact of this"

What actual partisan "vote splitting" is there? In all of these places, and samples both the senator and president exit polls are all conducted. They are done using the same sampling institutions also, if I understand correctly.

A better assessment would be that the fact these "senator" vs "presidential" irregularities occur so consistantly, is because of biased straight-ticket voting which does not in any of these instances favor the democratic ticket for unknown reasons.

As per the individual reports on http://www.votersunite.org, there is seminal bias across the board which would suggest why senators are consistantly showing more polling response than presidents.

[mgr]

11. Linkage issue

It works the same way alleles on the same chromosome work.

The way to understand what I am saying is that choices for president and choices for senator are paired. The same respondent states who they voted for senator and who they voted for president. If there is a greater WPE for Kerry (6.5%) that for a D. senator (1% say), that implies that more of those sampled preferred the democratic presidential candidate over the democratic senate candidate. That is the divergence that is suggestive of ticket splitting, or the incumbent effect with senator preference. Bias affects both equally, so the difference in patterns is real.

My impression was that many more republican senators were up for re election. What it suggests is that many Republicans split the ticket going for K as president (what I think happened); or many democrats support the incumbent. Either way it suggests movement to Kerry within the electorate beyond party registration being the primary determinant.

Mike

[Sancho]

9. Edison-M. knows what happened...

Competent pollsters know about differential question functioning (technical name for a common process that detects bias)...so they can tell from a set of embedded questions about income, sexual preference, etc. (you can see those questions when you look at the surveys), correlates to demographics, and comparisons to who actually voted how much bias there was...and they can detect outliers from the sample, representativeness, etc.

If there was a reluctant respondant effect, the locations and amount of bias should be known.

Why don't we get the raw data? I think that E-M has a good idea that there was a problem, and aren't going to give up what they have until someone leaks it or a court orders it. Their explanation was not convincing nor sophisticated for someone with the data in hand. The statisticians are working with secondary data, and still see evidence that the "sampling" effects don't explain...I agree.

The patterns of swing simply don't appear random, which is the biggest pointer to a serious investigation!

[mgr]

12. Raw data is available. N/T

[Sancho]

14. where....

all I've seen are tabulations and percents...the E-M website and all the links I've seen only show the results...

[Chi]

15. Raw data...

Raw data is available, problem is they refuse to give out the part that identifies which actual polling precincts they represent.
So I guess technically, it has only been partially released.

[mgr]

16. Available at Roper and Institute for Social Research since 3/2005

http://www.ropercenter.uconn.edu/2004_presidential_election_polls.html

http://www.isr.umich.edu/

At Roper you can buy the data on a CD; I have had no luck identifying the NEP poll data availability at ISR.

MIke

[Chi]

17. ISR data

"National Election Pool General Election Exit Polls, 2004 (ICPSR 4181)"
http://www.icpsr.umich.edu/access/quick-data.html

The fifteenth bullet down the page.

[berniew1]

10. NEP team did this assessment. Assessment was between results of

selected precincts to the state total to see if the total of results of selected precincts predicted the state results. This comparison was of total precinct results to "official state results", which were comparable. The analysis showed that NEP had chosen a good sample that matched the state results well. So problems not statistical, if there was a problem it was bias(fraud)in both the precinct results and the state total.

[Sancho]

13. Then I'd like the stack of surveys from that precinct's poll

If the stack matches the demographics of the voter rolls and embedded questions that are "predictors" (sexual orientation, party affiliation, education, income, etc.) are consistent with poll results: then the polls are accurate and Kerry won...

If there is bias or an "effect" like bandwagon or reluctant voter or whatever - it will be evident in the examination of the predictor questions and the same demographics.

I'm sure E-M knows this and has 100 pages of correlates and sampling statistics...I haven't seen anything like that...

[mgr]

18. I'm not sure what you are saying

There might be a sentence or paragraph that was inadvertently omitted.

Regardless of the demographics, the pattern of senator choice and presidential choice are linked at the level of respondent. You cannot separate the two. So if the presidential choice is out of whack with precinct patterns, so is the senate choice. The fact that the senate choice is in better agreement with the outcome is irrelevant, the same bias that resulted in oversampling Kerry supporters affects the senator choice equally.

Your concept of predictors does not hold water, if there was bias to wards Kerry support, then the demographics are tweaked in the same manner, and corrected for by reweighing. The consistency check would be with party registration, as this is the strongest predictor of future election choices, and is the easiest to obtain current information on.

NEP reports bias in their report (1/29/05), the bias is with poor sample design in that they did not adequately train the survey takers, in that Kerry support appears oversampled. All those correlations (multiple or single) are what brought them to that conclusion. I am not sure where you are going with your argument.

The bandwagon effect has nothing to do with this. It was applied as a partial explanation to address why the Bush support in 2000 came out so high. My explanation for this is that that particular variable has no informational weight for the next election, and was ignored in the reweighing, so it became a 'howler'.

Mike

[Sancho]

24. If we're talking about the same thing, I disagree..

1.) I think that a black female democratic union member teacher who voted for Betty Castor also voting for Bush doesn't make sense...how many of those stange voters are there by precinct/state/etc.

By the same token, the voter rolls are pretty easily sampled for any number of demographics. In theory, the NEP was targeted at representative precincts.

2.) Are there more/less "gays" or "democrats" or "women" or "minority" or "union" members in the polled sample in a precinct than in the voter rolls which record who voted. If not, then the sample is valid and the MOE applies and TIA is entirely correct.

That's simply the validation of the generalizibility of the sample. I believe that E-M has done this, and IF there is a biased sample, they know it already. Why would E-M NOT put out a table by precincts that were poor samples and why - instead of saying that "we polled a poor sample, but don't know why something is wrong"

Two things point to problems: inconsistent correlations indicating weird swing voters1.) or sampling effects that biased the results 2.)..... both are known to E-M (in theory). They haven't DEMONSTRATED either...only speculated. Why?

[tommcintyre]

23. RBR is moot until appropriate data to test can be gathered in the future

see this thread:
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x363706#363731