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

[TruthIsAll]

Their baaaack.

The "State Exit Polls (73,000 respondents) and
National Exit Poll (13047) were lousy samples" crowd.

So, let's reverse the spin.
Are they already backing off rBr?

Want a particular MoE? Calculate your own sample-size. No rocket science here.
http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=356832


And here's what Edison-Mitofsky say about MoE:
Surprise!Its 1% (rounded) for 8,000+.
We calculate it as 0.87% for 13047.

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x366149


And here is what MP said about Reluctant Responders BEFORE the election. The bias, if any, is FACTORED in by the pollsters.

THEY CORRECT FOR IT.

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x365678

[Febble]

1. OK, now define Margin of Error

In words. What does it mean?

And how does it relate to sampling error?

And what is sampling bias?

Hint: sampling bias has absolutely nothing to to with MoE.

[TruthIsAll]

2. You NEVER responded to this, Febble. Now's your chance.

And here is what MP said about reluctant responders before the election. The bias, if any, is FACTORED IN in by the pollsters.


THEY CORRECT FOR IT.


http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x365678

[Febble]

5. I'm still waiting for you to define those terms

when you can, I'll respond.

But unless I know you understand the difference between sampling error and sampling bias, there is no point in me responding to this post.

[TruthIsAll]

7. You will never respond to my post of a week ago if you have not

yet done so. We know that. You are usually very prompt in your responses.

And please don't pull a Sawyer and try to infer that I don't know what I'm talking about. That would be a very big mistake.

My challenges to you have been in the form of hard numbers. In the form of sensitivity analysis, if you know what I mean.
Not emotional sensitivity - numerical sensitivity.

And you have never ONCE refuted the math. In fact, you agree that the math is correct. So you are reduced to calling my data GIGO. That's pure chutzpa, a word you won't hear much in Scotland. When you are reduced to calling the exit poll data Garbage-in, Garbage-out, that tells me you have nothing more to say.

You refer to the state exit polls are garbage-in.
You refer to the national exit poll from 8349 to 11027 to 13047 as garbage-in.

Did you ever refer to the Final 13660 as garbage-in?
That is the only one in which weights and percentages were changed to impossible values to match the vote. Of course, you won't touch that.

Febble, some at DU may think you're a sincere truth seeker.
Well, this DUer is not convinced.

Soon the truth will emerge.
Make sure you keep playing both sides of the fence.

[Febble]

9. You have never once

told me what you think the MoE means.

You have not told me what you think sampling error means.

You have not told me how you think sampling error differs from sampling bias.

My answers to your questions depend on both of us understanding what we mean by those terms.

When you have told me what they mean, I will answer your questions to the best of my ability.

I am not saying you do not know what you are talking about.

I am asking you what you are talking about.

But I am finding it odd, to say the least, that you will not tell me what you think those terms mean.

[FULL_METAL_HAT]

12. It's like there needs to be some kind of new "Turing Test" for "election

intelligence" or something.

I find it en-heartening that both of you are still in open dialog despite the dialog being "hey you got chocolate in my peanut butter, no you got peanut butter in my chocolate -- what's the definition of peanut-butter cups??!?!?"

Febble, I do appreciate your long answer to my questions in another thread, but when you said "we can simulate sampling bias, and it's what I did in my model. We can't measure it though, and we can't predict it" it made me wonder whether you consider this overall sample size of this survey to be less than significant?

Ironically in looking for sample size of the UK election exit poll, I got to the Mystery Pollster page referring back to one of your DU posts! :) "Also consider that the previous problems with the UK exit polls did not produce election fraud conspiracy theories. Why not? One reason, as our friend Elizabeth Liddle points out, is that the count in the UK is "utterly transparent.""

That lead to one thought I had about the obvious in-traction between TIA and yourself: are you yourself aware of the modus operandi of the characters here in the US and in the whitehouse? I hate to say it, but as a cognitive scientist (that is right, no?) hidden variables such as societal sociopathy might throw you off the track. I know that rather goes against the notion of naked objectivity and analysis of the prima facia "facts", but I've studied enough cogsci myself to know that "fixing reality" these people are used to, can easily throw of any honest analysis of events. That itself is their MO. With hidden histories distorting the perceived reality, typical good-thinking ideas can actually be way off base from the actual.

Four cases in point:
1) 1933 Smedley Butler and the Coup to overthrow the US government
2) 1962 Project Northwoods "False Flag" terrorism planning
3) 1964 Gulf of Tonkin Lies to get into a war
4) 1967 USS Liberty Coverup *
(Google or wikipedia any of these)

All of these items are true, and all are actually very significant to history, but few people are aware of them and even fewer have looked at the pattern of deceptive projection they create. What I mean by this is ANY SCIENTIFIC ANALYSIS done to understand the reality in and around the above events would not be accurate as the published "facts" were bald-faced lies. Daniel Ellsberg was a academic in the Defense establishment during the 60s who got access to the "true histories" and worked on real analysis. He was so deeply offended by the deception that he leaked it to the public. The Pentagon Papers is a chilling read I highly recommend.

I know it takes away from the perceived "science" of exit poll analysis to bring in the psychology of malfeasance, but I do believe that's what adds the extra impetus to TIA's analysis and refusal to accept the one single "answer" we're talking about here when it comes to the 2004 Exit Poll "anomalies", that being the "answer" of "reluctant responders" is ultimately indistinguishable from "vote embezzlement".

Without a scientific framework for fraud analysis, its hard to include it as a scenario to be tested.

Fraud is never a friend of science. The basis of science is "objective truth" and the basis of fraud is "subjective lies", two great tastes that DO NOT go great together.

Sorry its late and the metaphors are weak.

I can only hope this diatribe might have actually shed a little more light onto this debate. I do believe "scientific naivete" is frankly just as valid as "reluctant responder" :( and I hate it.

* Far more shocking was Washington's response. Writes Bamford: 'Despite the overwhelming evidence that Israel attacked the ship and killed American servicemen deliberately, the Johnson Administration and Congress covered up the entire incident.' Why?

Domestic politics.

[Febble]

13. The sample sizes were large

so the MoE is small. Sampling error goes down with increased sample size.

However, the sample sizes at precinct level were small, so the MoE was much larger. However, these were not used to make the prediction, only to identify where the problems occurred. They are therefore very noisy data points.

But the point I keep making, at the risk of sounding like a kid fighting over Reese's pieces, is that sampling bias is real, and it has nothing to do with sampling error. Whether sampling bias occurred in this election is precisely the point, but you cannot conclude it did not occur from the MoE, because the MoE is a measure of sampling error not sampling bias.

It may well be that the kinds of rogues you currently have in the White House make vote-count corruption a likely explanation for the discrepancy. It's the reason I got interested in the first place. But that doesn't alter the logic - you cannot conclude that no sampling bias occurred because sampling error was low. The two phenomena have nothing to do with each other. We know, from the numbers, that either sampling bias was high, or fraud was high, or both were moderate. Something happened.

But just to repeat once more - you cannot conclude that sampling bias did not occur by calculating sampling error, which is what TIA repeatedly does, to many decimal places, I'm sure. You can probably only conclude it if you find the mole who hacked the election.

Which doesn't mean it didn't happen! :)

Thanks for the post, anyway.

[Febble]

15. re:

You refer to the state exit polls are garbage-in.
You refer to the national exit poll from 8349 to 11027 to 13047 as garbage-in.

Did you ever refer to the Final 13660 as garbage-in?
That is the only one in which weights and percentages were changed to impossible values to match the vote. Of course, you won't touch that.

Either all three of these are garbage (first two dirty garbage, last one cleaned up garbage) or the vote count is garbage.

We do not disagree here.

But I am still waiting for you to tell me what kind of error you think a margin of error is a margin of.

And how it relates to sampling error (or standard error if you like).

And how (if at all) it relates to sampling bias.

[mgr]

6. yes, by increasing the MOE n/t

[TruthIsAll]

8. How do you know that? Can you get a link? n/t

...

[Febble]

11. Well, up to a point

and only by reducing alpha to a more stringent level.

If you think sampling bias is going to be a possible confound, you might use 99% confidence limits (wider MoE) instead of 95%, before acting on your estimates. In fact, after the catastrophe in 2000, of "calling" Florida for Bush and Gore in turn, E-M were extremely cautious, and in fact no states were called erroneously.

This does not mean that the MoE got rid of sampling bias, it means that because sampling bias was a possibility they used a more stringent test of probability before "calling" a state.

The MoE is simply the standard error multiplied by whatever standard of probability you want to set. If you make it stringent enough it will include any other errors, including sampling bias - but then it probably won't tell you anything at all unless the "effect size" - the difference between the candidates - is large, because you will have reduced your statistical power.

The point is that although you can set the alpha for the MoE to make it as wide as you want, sampling error (as measured by the standard error) itself will be unchanged. It is a totally different animal from sampling bias.

And even if you set a very stringent alpha, and all your results are within your rather generous MoE, TIA will still, rightly, say - aha - but all the errors were in the same direction. This can't be chance.

And it won't be. It will either be sampling bias, or vote-count error.

There is no way of computing sampling bias in advance. We can compute it retrospectively if we know the "true" answer. But if we do not know that the "true" answer is truly "true" (i.e. that the vote count is correct), then the math will not tell you whether the bias we compute is in the sampling or the vote count.

Regression analyses conducted on a measure of bias (preferably mine), with good hypotheses and predictors, should help though. If we had the data.

[CottonBear]

3. Why don't you define the two terms? n/t

[kansasblue]

4. prepare the book!

Remember the Conveys Report. This will be the TIA Report!

[kster]

10. Please one debate at a time

lets finish the numbers debate first,then some other day we can start a new debate on the meaning of words.

[Febble]

14. There is no point in debating numbers

unless we know what the numbers mean.

I don't have any quarrel with the numbers. I want to know what TIA thinks they mean.

I have asked TIA to say what he means by Margin of Error. All he has done is to give me the formula.

I know he can calculate it.

I just want to know what kind of error he thinks it's a margin of.

[kster]

16. I understand, but

can we all agree on one thing. That the US of A needs a full scale investigation into the election of
Nov 2,2004.

[Febble]

17. Yes

But you have to start somewhere, and I'd start in Ohio.

It stinks way more than the exit polls.