|
Post by Chicago Jake on Oct 30, 2012 12:15:48 GMT -6
I tend to trust them more than most of you seem to. I can't recall a race in recent memory where the actual results differed significantly from the most recent polls. And while it's true that society is ch-ch-changing, statistical methods and polling are also evolving.
By the by - All these polls have a "margin of error" of some number of percentage points. From my understanding of statistics, a margin of error must have a confidence level associated with it. For instance, you might say the results are "within +/- 4% at 90% confidence" or something like that. Gordon (or anyone), any idea what alpha levels they use in these polls?
|
|
|
Post by Ardbeg... innit on Oct 30, 2012 12:23:30 GMT -6
I believe that most are 3 standard deviations or 95%
That means that if they repeated the same poll with the same techniques, from the same population, 20 times, they would get the same results within that "margin of error" 19 out of the 20 times. In other words, there is a 1 in 20 chance that the real numbers are outside that "margin of error".
The question is never about the calculations, the question is always about whether your technique accurately samples from the population... that is, does eveyone in the "population" have an equal chance to being questioned.
|
|
|
Post by Tex on Oct 30, 2012 12:31:54 GMT -6
2% and 95% is what Rasmussen uses I believe. Like Gordon, I studied statistics in school and actually used it quite a bit when I did independent audits. I agree with Gordon that it is not the calculations that are the problem, but the assumptions you make before calculating. I am not claiming that the polls are off in one direction or the other, but that the varying demographics of who voted in the last few elections combined with ch-ch-changes since the last election make it very hard to know what a good sample looks like.
|
|
|
Post by Chicago Jake on Oct 30, 2012 13:17:55 GMT -6
I believe that most are 3 standard deviations or 95%. Except that +/- three sigma would give you 99.73%. I doubt they use that level of confidence. 95% confidence would be closer to two sigma. And I agree with both of you that mathematical confidence is one thing, valid poling technique is quite another.
|
|
|
Post by Ardbeg... innit on Oct 30, 2012 14:09:49 GMT -6
You are right 2 SD
|
|