Statistical Methods

[barackmyworld]

Alright, I'm seeing a LOT of posts here using statistical tests to show a very small probability that the exit polls would be different from the election results. I've commented in a couple threads that exit polls are NOT random samples, and thus are not applicable in these types of statistical tests. Is there any reason why exit polls would be a random sample, or why you can use a non-random sample in a stat test? From what I have read, exit polls would definitely not be random samples because of the different times that people vote.

I've taken a couple stat classes at Harvard, and it's probably going to be my minor. I'm not going to claim that I know everything about statistics, because there are clearly a lot of accountants, etc doing these tests. Is there something I am missing here? The "rules" of statistical tests say they are only valid if the sample is random. If people wish, I can link some texts I've found online that say that.

[sixfive]

1. On statistical modelling

of exit polls and election results, I'm guessing tthe best apprroach would be an empirical one.

Look at the history of exit polls and how well they correlate to official counts to see what sorts of variances are typical. There will be trends which tend to change the essential accuracy over time, but this would be a decent starting point towards getting a handle on their statistical merit.

[Dogmudgeon]

2. "Randomness" is like the Platonic Solids

It really doesn't exist except as an ideal.

There is no such thing as a completely random sample. That's why statistics properly speaks of convergence to some central tendency. The criterion used to accept or reject a hypothesis in most statistical methods in the social sciences is the 95% confidence level. The criterion of "proof," of course, is replicability.

With exit voting, you have a large self-selected group. It is not truly random, but has enough intragroup variations for statistical methods to be useful.

One mismatch with an exit poll and the actual vote can be easily dismissed as a statistical fluke; but with over 1000 such results (as we had in the recent election) it is much more likely that there was some common variable influencing the outcome(s).

Good luck in school, too. If you truly understand what statistics measure, even if you're a little weak on the mechanics (math, formal logic, etc.), you'll be far ahead of the pack.

--bkl

[prof_science]

3. Well, I'm not Harvard educated, so...

...you probably can't believe a word of what I say. Anyway, exit polls aren't random, as you correctly state. Exit pollsters work their way around this by conducting what they call a "stratified sample." From what I understand, it's like saying "ok, when I've asked 100 women, I'll stop asking women and fill my quota of men." something like that. Google it, there's a load of information out there...

So you're at Harvard, but you're an Obama fan?

[barackmyworld]

5. I'm from chicago

Obama is my senator! and he'll be a great one :)

[prof_science]

7. OIC.

I'm from IL, too, but not Chicago. I find myself there every month or so (live in WI now) visiting friends, going to concerts, etc. I agree with you, Obama will be a force.

[barackmyworld]

8. on the stratified sample

those still have to be random. Like if you are picking 10 people out of every county in every state, that's ok if those 10 people are selected at random. Picking 10 women and 10 men wouldn't be random if those people were all voting between 9 and 10am, if there is a difference in who votes and different times during the day.

[prof_science]

10. From what I understand...

...experienced pollsters will weight the responses they've been given according to trends seen in past elections, among other things.

You know, I should really stop talking about things I know nothing about. I should and I will. If you ever have questions about brewing beer or restoring motorcycles, let me know :)

[DELUSIONAL]

11. bush is "Harvard" educated & Yale "educated"

As someone with a WEST COAST college education -- I am NOT impressed with East Coast colleges -- especially after learning about their legacy policy of accepting anyone who is breathing if their parent give a whole lot of $$$$.

Not to be insulting any individual -- but the west coast has some outstanding colleges and we can go toe to toe with any "ivy league" college.

So I'm a west coast snob -- so sue me!

As for Statistics -- exit polls are asking what someone just DID -- not what someone might do in the future. This can be thought of as a snap shot of something that happened. This is an eye witness account of an event (casting a vote) just moments after this event happened.

[barackmyworld]

12. heh, I agree

I am definitely *not* one of those people that got in by donating a dorm! But I do know that there definitely are people like that, because they have been in my classes. I think every school though (even you west-coasters!) admits a lot of "box of rocks" athletes that might be worse than the legacies.

On the topic now: even if people do say what they just did, the people that say that at 10AM might be different from the people that do it at 7PM, that's my main issue.

[Woody Box]

4. Could you elaborate?

exit polls would definitely not be random samples because of the different times that people vote.

How do you come to this conclusion? I can't help, but I think you don't understand the concept of randomness.

[barackmyworld]

6. It's worse than what can be accepted as "ok" random

because I have read that there are patterns to when people vote--women, elderly, unemployed vote during the day, and more right-leaning groups vote later. Is this true? I can't even produce sources because I just remember reading it in some other threads and hearing general TV/media talk about it. Is there solid evidence on it?

Maybe instead of showing that exit polls for state A don't match results for state A, somebody should find the probability that exit polls could be SO RIGHT in so many states, and so wrong in Fla, OH, etc.

[Woody Box]

9. Of course there might be timing patterns...

...that's why they're asking people at different times during the election. This method is called "randomizing". The more different time samples you take, the more randomized is the poll. "Complete" randomness is an ideal - see post #2.

You seem to imply that the polls are conducted at a fixed time. It's not like that.

And by the way, exit polls have always been very reliable everytime everywhere - except in the USA in 2000 and (increasingly) 2004.

[rman]

13. there's a couple of PhD's who'd disagree with you

http://www.buzzflash.com/alerts/04/11/The_unexplained_exit_poll_discrepancy_v00k.pdf

http://www.democracynow.org/article.pl?sid=04/11/10/1537201

Historicaly the margin of error of exit polls is 2%, and the errors have never been all in favor of one candidate.

[DELUSIONAL]

14. Repeat! "and the errors have never been all in favor of one candidate"

This is a smoking gun of sorts

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

and the errors have never been all in favor of one candidate

[barackmyworld]

16. then why aren't people showing tests on that?

it seems like there is a mistaken focus then. I am sure there are people who have done them, it just seems like every thread I see here is about exit polls vs. outcome, not the uniqueness of this year's exit polls.

[DELUSIONAL]

17. Read what Truth Is All has been posting

If these are true errors they should be 50/50 -- but to have ALL the exit poll oopsies in bushie's favor is THE virtual smoking gun.

[tritsofme]

15. The outfit that conducted the exit polls says MoE is 4%

on a state level.

The margin of error for a 95% confidence interval is about +/- 3% for a typical characteristic from the national exit poll and +/-4% for a typical state exit poll.

http://www.exit-poll.net/faq.html#a15

[barackmyworld]

18. thanks for the replies guys

the articles especially were good.

[Awsi Dooger]

19. Here's some good info on the methodology of exit polls

From a very good website, Mystery Pollster. Now identified as Mark Blumenthal: http://www.mysterypollster.com/main/2004/11/exit_polls_what.html