On 03/20/2010 06:29 PM, Jed Rothwell wrote: > Stephen A. Lawrence wrote: > > > Oh, I don't know -- keep in mind that the "random" sample is actually > self selected. Have you ever consented to answer a phone survey? > > > It isn't supposed to be self-selected. There are ways to reduce that > bias. Granted, the methods are not perfect. > > Does it say these are telephone interviews?
Yes. Phone interviews work pretty well for political polls, presumably because willingness to cooperate with phone surveys doesn't correlate strongly with political alignment. In this case, however, I'm not so sure the correlation with ability to answer the questions correctly is insignificant. > Fully self-selected samples are, for example, when you set up > a questionnaire on a web site. Those are, indeed, self selected, and what's worse, they're nearly always trivially gimickable. With a little effort most of them could be flooded with a Perl script. In other words, they're totally worthless (unless the issue is one nobody cares about). > Fully non-self selected is the U.S. > Census, now underway. There are gradations in between. The commercial > phone surveys about radio stations with few respondents probably are > toward the self-selected side. The telephone polls before political > elections are remarkably accurate these days, despite problems with cell > phone. Of course, people who vote in an election are self selecting by > definition, aren't they? If you are likely to tell someone on the phone > what you think, you are also likely to vote, I suppose. Voters are > self-selected except in Australia, where it is against the law not to vote. > > - Jed >
