How does the poll works?

[Gnister]

Does anybody know exactly how the polling companies make up their samples for their polls? Is it based on last election turnout? Or how do they select how many latino, old, young, afro-americans they will choose for their sample?

I've heard all kind of things regarding this. Does anybody know?

[suston96]

1. Each pollster tells how they do what you are asking. Go there and look......nt

[Gnister]

2. But...

Usually all that they state is how large each subgroup is, i.e. how many females, males, latinos, young, old voters their polls is made up of. Not how they came up making their sample consist of for example 11 % young voters....or do they?

[suston96]

3. Did you look? Or do you want me or someone else to do it?

The validity of their poll depends on their sample being representative of the poll's intentions. If, for example, they want to poll what Latinos plan on doing in Texas they will make sure their sample is comprised of Latinos. And so forth.

The less representative their sampling is, the less dependable their results.

[mohc]

4. As an example

SUSA links to this statement of methodology on each poll. Most other pollsters have a similar disclaimer.

So for SUSA they:
1) Use random digit dialing
2) Weight answers to demographics based on previous census
3) Perform a likely voter screen

[HeraldSquare212]

5. Well, on TPM some people are saying that SUSA has the expected black turnout too low

they're modeling 17%, and apparently there was a 20% turnout for Kerry in 2004, even though Kerry has the nomination sewed up by then. I don't think they say why they chose 17%.

But that's just to make this point: that you need to know what guesses they are making in the make-up of their sample, and be able to compare those to prior races and some educated guesstimates about turnout this time around. That's really what it comes down to - a good guess. Recall that in Iowa, everyone projected around a 150,000 turnout; the Iowa Daily Register projected 200,000, and everyone said they were crazy. It turned out they were right, and that's why their poll was the closest to correct.

That doesn't answer your question, but it suggests the additional research you might have to do to check their assumptions and even after that, you still have to guess what the turnout this year will look like.