On Thu, 5 Oct 2000, Doug Henwood wrote:
> Marco Anglesio wrote:
> >I wouldn't give too much credence to an individual poll. Zogby-Reuters Oct
> >2-4 has Nader at 7%, while Gallup Oct 1-3 has Nader at 2%. A candidate
> >scoring low is apt to appear to fluctuate more strongly, but that's merely
> >because they're scoring low in the first place and therefore any gain or
> >loss in a poll is a massive gain or loss.
>
> The variations may also have to do with how they weight "likely
> voters." What if a lot of Nader supporters are young and/or former
> nonvoters?
That's true, but you would expect the same sample bias in each tracking poll, unless they modified the criteria for likely voters over the course of the tracking poll. I suspect that it's merely sample error, accounted for in the confidence interval. One should expect the poll numbers to rise and fall even if support within the sampled population is stable.
So, Doug, I agree with you that Nader supporters are potentially underrepresented in polls due to sample criteria, but I don't see it necessarily as a cause of sample variance.
Zogby has been criticised for his likely voter model before, actually - it seems that his name-making prediction of the 1994 mid-term election results may have been, for lack of a better word, a fluke. Has something to that effect been posted on LBO-talk?
Cheers,
Marco