Ok so then let’s say the population in my state is 10 percent black but in my 1,000 random calls I only connected with 75 black voters. Do I then go back and give those 7.5 percent of black voters a 10 percent weighting in my poll?
You make a good point: We know some things about the population; e.g., how many people are black. BUT ... the percentage of blacks who vote may not be the same thing as the percentage of blacks in the population. The possible difference is called turnout.
Black people actually vote pretty regularly nowadays. I’d say more uncertain than black vote is hispanic vote, and more uncertain then either of these is the youth vote.
And, if you assume these demographics vote with the same frequency as they did in 2008, a monster year for turnout, you wind up with a very Democrat-looking sample.
A better way to do this is to first weight the sample by demographics and second to tease out the likely voters from the sample. Think of the first step as using demographic weights to insure that you have a sample that is representative of the population, and the second step as allowing the likelihood of voting to be forecast, in addition to the forecast of how those who vote, vote.
Going the opposite way, first teasing out the likely voters and then weighting the sample by demographics and possibly also partisan affiliation so the sample of likely voters “looks like” the voters who showed up in 2008, will have the result of predicting that whoever won in 2008 will also win in 2012 as well.
No you keep polling until you get the number of respondents that fit your weighted sample. In your case you would keep making calls until you got 10 self identified blacks.
Here is the other thing about polls. It assumes people are telling the truth. Maybe they are, maybe they are not.