Stephen A. Lawrence wrote:
Consequently, in an opinion poll they might SAY they'd vote for him, but then when they're all alone in the voting booth they'll vote the other way. This can skew the opinion polls to make Obama look more popular than he really is.
Exactly right. That's the Bradley effect.
(AFAIK nobody knows how large this effect is.)
Well, it is possible to make crude estimates of it. These estimates show that it has probably diminished since the Palin nomination, probably for the reasons I described.
This is actually a part of a larger problem with opinion polls, which is they're not "secret", and apparently not anonymous.
Some of them are. The answers come out quite differently depending on whether they are anonymous or not. You have to compare results from different methodologies to determine the extent of the bias, and you have to compare results to actual voting (or purchasing, or whatever you are polling for) where data from some other source is available. That way you resolve which method is more accurate in which situation.
I think most polling place exit polls are anonymous, although they usually ask you to check off a box for your sex, age and race. (It is anonymous because they do not look at the paper you fill in. They aren't supposed to, anyway.) Until this year, polling place exit polls were extremely reliable but there is so much advanced voting in many different states this year that the accuracy of exit polls will decline.
The choice of the interviewer can also have a profound effect on the responses, depending on whether its political content or not. For example if you ask a black man to interview black people you are likely to get very different set of answers than if have a white woman asking the questions. This is true even when the interview was conducted over the phone, because most people can tell the race of the person they are talking to. This was discovered in the 1940s by white and black statisticians (including my mother) in carefully controlled tests.
Even in an exit poll, some people are inclined to lie or misrepresent their vote, out of spite or confusion or because they forgot who they voted for. However, as my mother used to say, "we know they lie and we take that into account." They calibrate for it. It is hard to fool a public opinion researcher. They were always well aware of the Bradley effect, even before it had that name, but that raises the question as to whether they should try to fiddle with the statistics to eliminate it. Some do, and some don't. That is one of the many reasons election polls vary all over the place, by as much as 10%.
My mother also said that in response to polls and census forms, people tend to say whatever pops into their head. Here is an example of her writing in J. American Association for Public Opinion Research. You can see that for some inexplicable reason she sounded a lot like me:
http://www.jstor.org/pss/2747532 - Jed
