On Fri, 6 Oct 2000, Brad DeLong wrote:
> So what the f*** is their definition of a likely voter? I know, I
> know, I know it's "proprietary."
Well, the model proper is proprietary. I think I read a journal article which touched on it a while (okay, about 3 years) ago. Essentially, likely voter models are of two kinds.
First, they can exclude people from the analysis (in other words, if you're not a likely voter - for example, if you belong to a minority, have a low income, aren't registered to vote, didn't vote in the last election, are of a certain age) - your answers are ommitted from the pool of valid answers. This is a pretty crude filter, but it's of quite common use.
Second, a statistical model can be applied, so your answers are weighted up if you belong to a group that votes disproportionately and down if you belong to a group that does not vote disproportionately. This is a somewhat finer filter, and has the advantage of requiring less respondents, but requires a much better technical touch.
A third kind would be a mix of the above.
Also worth reading: http://www.gallup.com/poll/fromtheed/ed0005.asp http://www.dartmouth.edu/~chance/chance_news/recent_news/chance_news_7.10.html http://www.cnn.com/ALLPOLITICS/1997/gen/indepth/polls/faq.html http://wcb.uww.edu/wcb/schools/400/820/mohanp/3/modules/page15.html
Marco