Ignore the RW spin. Calc the SAMPLE-SIZE for any MoE and confidence level

[TruthIsAll]

There is more disinformation on this topic than anything else because the nay-saying crowd wants you to believe that exit polls are not accurate due to sampling margin of error. They are intentionally misleading those who are uninitiated in statistical sampling theory.

http://www.raosoft.com/samplesize.html
This is the best interactive explanation of Statistics 101 as applied to surveys that I have seen.

Input these values to calculate the required sample size for the National Exit Poll:

Margin of error = 0.875%
Confidence level = 95%
Population size: 200,000,000

The formulas displayed at the bottom of the site are the ones I have used as the BASIS for all of my probability calculations for recorded vote deviations from the legitimate, pristine, non-biased, uncorrupted exit polls - up to and including the 13047 sample of 12:22am on Nov. 3 which had Kerry winning by 51-48%.

The Final 13660 National Exit Poll was weighted to match the vote.
It turned the scientific poll of the first 13047 respondents around on its head.

What's the point of doing a survey if you are not going to accept its results to within a pre-designated margin of error?

Would you ever trust a scientific survey of any kind if you knew that the results were adjusted in any way? Let's try to use a litte common sense here.

So spend a few moments and design your own sample-size. You will learn a lot by doing so and you'll see right through the slick misinformation promulgated by those whose agenda is to keep the best statistical evidence away from the general public.

THIS IS NOT ROCKET SCIENCE!

Don't let the trolls pull the wool over your eyes.

[In Truth We Trust]

1. Truth is truly ALL there is. Keep spreading the word TIA and thanks for

doing so.

[Botany]

2. TIA thank you again.

Can you repost the data you had on your last post about the #'s and
%s of "new votes" * had to get in order to reach the numerical # of votes he
had.

In other words did he have to get + 67% of all the new vote and none of his
2000 vote could have died or voted for Kerry in '04?

[davidgmills]

3. This was fun but

I know the naysayers would say that this can be used for a random sample only and that the exit polls were not random; they were clustered.

But you would think that highly selected clusters could approach a random sample, so that these numbers would not be that far off.

On the issue of the 13,660 being representative of 120 million, I would think that a good check on their representative status would be party ID. In the 13,047 sample party ID, as I recall was about 38 Dem and 35 Repub and the remainder, Ind.

One of the big issues before the election was how to weight polls with party ID. Zogby usually weighted his polls about 38 or 39 Dem and Rasmussen and Gallup came close to an even split when Gallup decided not to give the Repubs a huge lead.

I can remember arguing with a Freeper about Zogby being right and he was emphatic that since 9/11 Repubs had caught up with Dems or maybe even surpassed Dems.

So, I have always had this question since: What is the true US ratio of party ID now? Has it changed or are the Dems still ahead by two or three points? As you know, I suggested many times a re-exit poll and the answer always was they can't be trusted because people will lie about who they voted for in favor of the winner. But what about a poll now on party ID?

If there were an accurate poll today, which people had difficulty disagreeing with, that showed the Dems ahead by two or three, I think it would go along way to convince the naysayers that the 13,047 stats were dead on and that these 13,047 were truly representative of the population as a whole. What do you think?

[On the Road]

5. It's More Than Just Being Clustered

the exit polls are not samples of the officially counted votes. That is the ONLY situation under which probabilities can be assigned.

[davidgmills]

6. It always amazes me about how the "lie" factor always creeps in

People will "lie" about how they voted so you must have their actual ballot to prove they did not lie.

Lying just seems like a convienient excuse for everything.

After all, we Americans are just a bunch of liars. We have no morals. We have no scruples. We will lie about anyting just to lie. We lie when there is no money in it. We lie for the fun of it. We lie when a brownie scout gives us a sheet of paper to fill out. We lie when a hippie gives us a sheet of paper to fill out. We are the Lying States of America.

[On the Road]

7. It Doesn't Matter

you simply cannot say that an exit poll is the same as sampling the actual votes. Meaning that you cannot assign a probability to it based on the standard distribution.

[davidgmills]

8. My point exactly

we are liars. We can't be trusted to tell someone what we did. So therefore the exit polls are invalid.

It is so good to be a nation of liars because it makes a mockery of statistics.

[TruthIsAll]

9. So an exit poll is NOT EVEN A SAMPLE of the actual votes?

Ok, you are right.

It's a sample if how people THOUGHT they voted.

Wow.

[On the Road]

10. It's Not a Sample of the Actual Votes

A lot of the people involved in this effort come from the statistical side and are looking only at the mathematics. There's a hidden assumption that a sample taken by exit polling is the same as a sample of the vote.

I come from an experimental psych background in which you couldn't possibly get away with publishing a claim that a poll, especially under these conditions, is the same as a sample of the actual vote. I worked on one issue that used to be hot when I was in college in which the whole controversy, extending over several hundred papers, had to do with eliminating response variables. People kept claiming that that results showed an effectm, and others were able to show the effect was caused by response variables. It's not easy.

It doesn't mean that there's no reason to be suspicious or that everyone should just forget the last election. What's not correct is the assigning of probabilities -- eg, "there is a one-in-600-million chance of the exit polls being this far off."

If you're hearing a deafening silence lately, I think it's partly because people sense this but don't know how to phrase it. Personally, I think it tends to bury a lot of more specific reports that are very troubling, and which are simply being stonewalled.

There are ways to use exit polls, for example, to show differences among different variables -- voting machine type, party control, etc. I've seen different sources, and for every report that claims a difference among machine types, for example, I've seen others showing no difference between paper and evoting.

The lockdown north of Cincinnati may have been used to alter ballots. That might show up in a larger exit poll discrepancy in that area. You still couldn't assign an exact probability, but that would be an effective use of exit polls. JMO

[Sancho]

11. You're still leaving out something....

I've worked in related areas - not exactly political polling - but any competent pollster who's not an amature has...

1.) maps of distributions of party affiliation etc. to get stratified or representative samples
2.) embedded questions and observations to estimate lying or inconsistent responses (like the guy with the IBM briefcase and Bush button saying he voted for Kerry)
3.) demographic targets so that there isn't a volunteer effect
4.) historical experience at training pollsters and wording questions

If they have HUNDREDS of low-down lies and can't spot them they are more than incompetent...the trick is a representative sample by planning the samples and measures to start with....anticipating the "nonrespondents" who don't want to answer. These pollsters have the $'s to get it right.

If the statistical scholars (that's for you, Q) out there want some late night reading, find "Eelworms, Bullet Holes, and Geraldine Ferraro" by Howard Wainer....Journal of Educational Statistics, Vol. 14, No. 2

[On the Road]

14. Of Course, I'm Leaving Out a Lot

I'm just making the point that "one-out-of-X" probabilities cannot be projected from exit polls.

The professional pollsters have lots of context and history and a certain amount of control. None of those are foolproof. None of the pollsters are making the claims that exit poll discrepancies are proof of fraud.

The exit poll consortium claims that there was a model error during the day which allowed too many women into the sample. We don't know if that's true, but as far as I know there's no specific reason to doubt it.

If there were a clear anomaly in certain areas, like the Florida county in 2000 with over 10,000 triple-punched ballots, that would suggest fraud. You could also do make "one-out-of-a-million" statement about that situation, because it involved real ballots.

I would love to see it shown that GOP-controlled evoting wards, or all evoting wards, were especially skewed towards Bush. I've seen it claimed, but I've also seen it refuted.

[TruthIsAll]

16. You cannot assign a probability to a sample? What do all posters do?

They want to have a attain accuracy to within a pre-set MoE based on a desired confidence (PROBABILITY) level that the poll will produce a result within the MoE.

That's exactly how a pollster determines how many to sample.

You are so wrapped up in your spin that you ignore the lessons from the interactive sample-size calculator.

You have no clue as to the relationship between the required MoE and the confidence level.

You ignore the INPUT MoE.
That tells you how many you need to sample to achieve the accuracy you want.

You ignore the INPUT confidence level.
That tells you what the PROBABILITY is of the ACTUAL POPULATION mean of falling within the MoE.

You ignore the OUTPUT of the calculation:
THE REQUIRED SAMPLE SIZE TO SATISFY THE PRE-SET CONFIDENCE LEVEL THAT THE RESULT OF THE POLL WILL FALL WITHIN THE PRE-SET MARGIN OF ERROR.

You ignore everything.

[On the Road]

22. Your "600-Million-to-One" Claims Would be Correct

if you were referring to a random sample of official ballots.

Instead, they refer to an exit polling effort which sampled selected polling places, assigned relatively untrained people and told them to interview every seventh person who exited the polls. The results were then weighted to reflect the voting population.

One is not the same as the other. There's no way to quantify the how well the people were sampled, the response variables, or the weighting. You certainly cannot claim the *same* probability for sampling exit polls as for sampling ballots.

In practice, pollsters always use reported data from as random a sample as they can get. It's not always very representative. They know it's not the same as the actual data. They don't always take pains to make the difference clear, party because it's the same for every poll.

By and large, even pro-Democratic pollsters like Zogby do not believe the discrepancy between the exit polls and the official vote indicates fraud. That should tell you something.

There are situations in which systematic discrepancies would arouse suspicion. For example, if the divergence was greater in evoting polling places or in evoting areas under Republican control, that would be prima facie evidence of fraud. You still couldn't assign a probability to it if it were only based on exit polls.

This is what bothers me. I think it helps the effort much more to focus on qualitative facts that should arouse suspicion rather than making unsupportable quantitative claims.

[TruthIsAll]

23. Tell it to uscountvotes.org

You don't think the discrepancies arouse suspicion?

Ok.

[On the Road]

25. Of Course They Arouse Suspicion

and they should be fully investigated. Thanks for the site -- I just gave them $25 and hope they succeed in uncovering fraud.

I think the ElectionArchive.org people are wrong in attempting to give percentages upfront that the discrepancy would happen by chance. The numbers are not supportable and it weakens the argument. Appealing to authority and credentials of a partisan group is not the way to win the hearts and minds of the general public.

They are on much stronger grounds when they avoid the "one-in-a million odds" arguments and discuss, for example, the possibility of vote-padding in predominantly Bush precincts.

A disconnect of one out of twenty voters does not sound like a lot. In general, that could easily be attributable to misses and nonrespondents. But their modeling of the implied response rate by Bush/Kerry support shows how unusual the patterns are when you drill down to that level. That's what has to be headlined. It's a subtle point, and it's more difficult to get into a lead paragraph, but it can be done. That's the challenge.

They are also pretty good in discussing the disconnect between the presidential and senate vote, although it would have been better if they had not excluded North Dakota.

I'm not contending that there was no fraud, although I don't know if it was enough to swing Ohio or Florida. I'm saying that the weakest arguments are leading and the strongest arguments are being buried.

[davidgmills]

27. Did Zogby say this in print?

By the way, are you just now learning about USCountVotes.org? If you are, you are several weeks behind the curve. Maybe you need to read their article.



http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=351187&mesg_id=351187

[kansasblue]

4. PBS on Karl Rove

briefly mentioned the exit polls showing a Kerry win. They never directly explained it but they made it seem as if the exit polls didn't / couldn't reflect the surge in Bush's base. From your research is there anything to indicate that the exit polls wouldn't see that? I guess it might work that the exit pool establish which way the base (in percentage) is voting but the exit poll is unable to know that numbers are coming to the poll. So they calculate (for example) of 70% / 30 % going for Bush and normal turnout is X million. But the X million is what changed and the exit poll doesn't track that. Or is the exit poll itself very complete and would/could/should see Bush 'turning out the base'?

[Sancho]

12. If I understand....we're saying the same thing...

Pollsters should have a count of the number of males, whites, etc...coming out of the poll between 5 and 6PM (a demographic profile of where the poll sample came from)...even if only a few voters answered the poll...and so they should have SPOTTED any trends, lies, or confounding variables during 15 hours of polling!

It doesn't make sense to have pollsters standing there with Kerry winning and not having evidence of self-sampling errors, bandwagon lies, or last minute surges. All those Bush voters would be evident in observations of the pollsters, trends in the questions that correlate wth "Bush" voters, etc.

IF M-E has evidence of such; they've had lots of chances and haven't mentioned it...so the polls are accurate and M-E doesn't want an external examination of what's in the file drawer.

[TruthIsAll]

17. Kerry won the new voters. All exit polls said from 54-59%

So just what is this base you are talking about?

Oh, maybe Rove meant those 1.75 million Republicans who rose from the dead to vote for Bush once again (as indicated in the impossible the 43% Bush /37% Gore split).

Given that we have no evidence that these dead voters could resurrect themselves, Bush needed 64% of the 22 million new voters to win by 3 million votes.

But Kerry won the new voters by a 4 million vote margin.
If he did, and ALL exit polls indicated he won new voters by 9-20% (2-4 million) then we must believe that....

Those former Gore 2000 Democrats who came out to vote pissed off about the stolen 2000 election were part of the Bush base this time.

Right.

[Sancho]

13. Exactly...here a good example...Neilsen Ratings

Neilsen uses 5000 households to represent 99 million TV viewers....and on their web site they say, "Actually, a representative sample doesn't have to be very large to represent the population it is drawn from." and they they tell a story about soup...but the idea is the same.

http://nielsenmedia.com/whatratingsmean/

[libertypirate]

15. One more thing...

It's ok for republicans to site exit poles in other countries as evidence of fraud.

And it's not ok for democrats to do the same in their own country.

Mindblowing!

[StudentOfDarrow]

18. Thank you

and keep up the good work.

[kiwi_expat]

19. The NEP Methods Statement includes discussion of sampling error:

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


(I am not qualified to comment.)

[TruthIsAll]

20. The NEP table shows that the max. MoE for 8000 respondents is 1%

[kiwi_expat]

21. To see how NEP prefaced the table:

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=357345#357376

[anaxarchos]

24. TIA, I don't remember the thread but...

You have done this calculation once before. You had assumed a 1% MOE and the numbers came out different from Edison (if I remember correctly). Plugging in .0085 or .008 got a match. There was a short discussion about a 1% MOE being "too conservative". I could be wrong...

[TruthIsAll]

26. My calculated MoE for 13047 respondents was always 0.875%

That assumes a 50/50 characteristic split (gender) (the maximum). The MoE would be even lower than 0.875% for any other split.

Since Mitofsky used a 1.0% MoE in the notes at the bottom of the 10047 WP/NEP screen, I figured it was due to roundoff - or maybe am adjustment for a "cluster" effect.

Notice how ALL MoE's are rounded to 1% for 8000 + respondents.

In actuality, for N = 8000, the formula gives an MoE of exactly 1.095%, assuming a 50/50 characteristic.

[TruthIsAll]

28. How to take the Mystery out of the Pollster in just a few minutes.

.

[kster]

29. kick.nt