>Doug writes:
>"That's a version of what's known in the trade as a push poll, most
>commonly used in political campaigns. A rep of Candidate X calls up,
>posing as a pollster, asking: "How would you feel if you heard that
>Candidate Y likes to have sex with dogs?" Well not exactly that
>question, but you get the idea."
>
>You're implying that there's a way to do a neutral, clean poll, but
>I just can't see it. A poll occurs in the real world; even if it
>tries to strip itself of any context, it remains in the context in
>which it's given. If a poll-taker has been watching FOX news for the
>last few years, he/she will answer questions from a context in which
>Hussein is tied to Al Queda and is almost directly responsible for
>9/11. In this context all Arabs are actual or potential
>fundamentalist terrorists with whom you can't reason. All the U.S.
>wants to do is to give them democracy and freedom. Since they are a
>primitive, backward people, the only way to impress them with the
>value of democracy and freedom is to bomb the shit out them and to
>torture them because the only thing they respect is force. Once they
>understand that the U.S., which is also the most democratic and
>freest country in the world, is also the most powerful country in
>the world, they will understand that democracy and freedom ma!
>ke you strong and feared and they will choose it for themselves by
>choosing to serve the interests of the U.S.
>
>
>Now, would you like to answer a few perfectly netural questions?
But the poll is trying to measure what people think - those who've been watching Fox out of true believerhood, those who watch Fox sometimes for sick laughs, and those who don't watch Fox at all. Respondents' answers to an honest poll are shaped by forces outside the poll, not within the poll itself.
Of course, no real poll can match that ideal. The questions and answers are themselves shaped and constrained by an ideological context. But there are degrees. If you want to spin answers in a certain direction, you preface the money shot questions with a lot of ideological foreplay. One of the most dishonest polls I've ever seen was one sponsored by a health insurance company from the early 1990s. Other polls had been showing wide support for a Canadian-style system (when it was neutrally described). The insco poll showed very low levels of support. I was confused at first - until I saw the context. The policy questions were prefaced with a long list of questions about how well people liked their doctor, how pleased they were with their current health insurance, etc. This was well before the HMO takeover, so most people were pleased with their current sitch, even if they were worried about health care as a political issue. So the results were stacked in exactly the way the insurance company wanted them.
Honest pollsters also try rotating the order of questions, to see if any bias is being introduced by the sequence. Or by comparing responses to various wordings. Or by splitting samples, and asking different subsets of the sample different sets of questions, to see if any bias is being introduced. There's no way to avoid the biases introduced by forced choices (yes/no, A?B/C/D), but there are ways to do polls better than others.
Doug