What would polls look like if you took out 25% of non-whites? Answer here:

[troublemaker]

Have you ever asked yourself what polls would look like if you took out 20% of non-whites and replaced them, person for person, with white people making over $75,000?

Well, we no longer need to speculate because Gallup has done the hard work for us! What the hell does Gallup have against non-white people? More pointedly, WHAT DOES GALLUP HAVE AGAINST PEOPLE MAKING $50-75,000???

(2000 exit poll actual results in parentheses)

Income:
Over $75,000: 32% (28%)
$50-75,000: 16% (25%)
$30-50,000: 26% (24%)
$20-30,000: 11%
Under $20K: 9%

Race:
White: 85% (81%)
NonWhite: 15% (19%)
Black: (a subset of NonWhite) 8% (10%)

http://www.mydd.com/story/2004/10/19/195135/73

[KurtNYC]

1. I think this is self-reported income level

so probably what is happening is that people really in the 50-75 category lie and say 75+.

The bigger gallup problem is too many GOP in the sample (and they know it).

[troublemaker]

2. Mistaken on both points

People are at least as likely to lie about their income to an exit poller, which is the source of the exit poll numbers. (Both are self-reported)

The "too many GOP" is not the problem because there is no reason to adjust the sample for party ID. Gallup correctly defends themselves on that point by saying that party ID (self-reported) changes from week to week. What they don't advertise is a failure to weight fixed criteria.

The point of the linked piece is that Gallup does not weight for race or income. All reputable pollsters weight for race and income. Race and income, unlike party ID, are precisely the kind of traits you SHOULD weight, which is the point.