Bob Dole's '96 polling firm, the old line Republican outfit Fabrizio-McLaughlin, has an
interesting poll in PDF format out. Their main polling (the type you see reflected in most national polls that still show Bush ahead by 1-3 points) has the election at 47.3% for Kerry and 47.1% for Bush as of October 25.
But... (and this is the big but) firm principal Fabrizio adds on this welcome comment;
However, as the data below illustrates, when the data is weighted to reflect minority turnout based on the 2000 exit polls, Sen. Kerry leads by 3.5% and if minority turnout is weighted to census levels Sen. Kerry’s lead expands to 5.2%
Weighted to 2000 Exit Polls Kerry-49.2% Bush-45.7% Nader-1.5%
Weighted to Census for 2004 Kerry-49.9% Bush-44.7% Nader-1.6%
Let's think about this for a second. If ethnically weighting to actual voter turn-out rates moves Kerry from a statistical tie up to a 5% point lead... does this mean that ALL the polls are undercounting the minority vote? In the past we've seen the meme passed around that black voters are somehow not real Southerners, but now it seems that a lot of national polls are being conducted on the assumption that minorities don't really turn out to vote in the numbers that they, um, historically turn out to vote in.
From working in market research I know that different ethnic groups can be harder to sample, based on the type of methodology being used and the hiring profile of the interviewing staff. But I would think that, after 50 years of professional polling being in existance, a few firms would have figured out how to accurately measure the actual vote projection without the somewhat risky practice of "weighting" their samples.
Apparently not.