2016 Postmortem
Related: About this forumWhen pollsters do their analysis do they include any kind of weight for
specific areas of a State that is the most populace and leans left or right? i.e. Pa. has a lot of voters, but the majority by far, are located in Philadelphia & Pittsburgh, and usually vote the opposite of the rest of the State. If calculations don't include that as a factor, then all the polls could be misleading.
Foggyhill
(1,060 posts)It should reflect the underlying pop
If you have more suburban - bias GOP
More land lines - older - bias GOP
More cell phones - younger, minorities - bias dems
More rural - bias GOP
More cities - bias dems
More south, mountains and mid west
- bias GOP
Call during day - bias GOP
Etc...
Making sure all those biases are removed is very hard, even harder when the number polled is low and there are huge variations in possible response
National polls should have at least 2000 polled in a manner to diminish sampling bias
State polls should be at least 1000
With a bad sample the margin of error becomes meaningless (except for the pop you actually polled (say white suburban voters)
Many think margin of error takes into account this bias, it does not. When the sampling is biased there is no way to determine how much you are off. That's why unskewing doesn't eork
Cicada
(4,533 posts)They make sure the sample matches the number of voters region by region.
Foggyhill
(1,060 posts)By age, region,
Most of those polls actually don't do that at all, when your calling 100 inside a state or even 50, even a small bias will be amplified
Most national polls are just garbage
PoindexterOglethorpe
(25,857 posts)are careful to sample representatives of all the sub groups out there. Done right, a surprisingly small poll can be very accurate. Done wrong, even a very large poll, can be hideously wrong.
The best example is the Literary Digest Poll of 1936. 2.4 million respondents. Showed Alf Landon to be the clear winner that year. It's very instructive to read up on that poll and how they got it so wrong.
Another thing to remember is back when there was no early voting, exit polls were very carefully and well done by the three networks, and they got the results very right, very early in the day. Now that there's so much early voting, plus three states that only vote by mail, exit polling is meaningless overall. It can still be useful in those states with no early voting, but they are so few as to not matter. Or be worth doing.
Foggyhill
(1,060 posts)You still need a good number of people polled in national polls to smooth out accidental biases
If you are polling 700 trough land lines, you are in fact almost guaranteeing a GOP positive result
Good pollsters can get decent results but their polls are buried in an avalanche of shitty polls and people lose their confidence inpolls
napi21
(45,806 posts)I've been using 538 as my base #'s and seer if most of the other polls are similar. Any that deviate far from what 538 found I see as an oddball. It sounds like the good polling methods you guys are speaking about are the methods used by the Campaigns. They always seem to know a lot more than the public hear.