Reply #125: they actually do use 99%...

[c-macdonald]

MysteryPollster (who is a professional pollster with years of experience in polling) as well as other professional pollsters have said that they use a 99.5% confidence interval for the EXIT POLLING, not all polls in general. The reason is that using a 95% C.I. for 51 states (incl. D.C.) would result in 2-3 results falling outside the MOE due to chance alone. That's why on election night they rely on a 99.5% confidence interval to call a state for a candidate. This is cold-hard solid fact and, frankly, I'd listen to an expert on polling on this one.

Your discussion of MOE isn't exactly clear though. It all depends on whether or not you have a one-tailed test or a two-tailed test. I've been taking a hard look at Freeman's data (with input from MysteryPollster and another blogger) and I think a one-tailed test would be better to use (the blogger insists on a two-tailed test).

But in either case, you have to look at Bush's and Kerry's results separately, meaning that there's a 2.5% chance Kerry's vote would be above the 95% C.I. and 2.5% chance it would be below (and likewise for Bush). There really is no way (that I know of) that you can combine the Bush and Kerry results to get an MOE as you described.

I've discussed several times on here about your calculations of the MOE and how I think you've underestimated it. I've done my best to get an estimate of the MOE in each state and while the results aren't as overwhelming as they originally were, there are still some very interesting results. Once I do some more checking and verifying, I'll post on my findings. I'll even send you my spreadsheet of the data (full official election returns and Freeman's updated data) if you agree to have a civil conversation about your analysis.