Friday's Washington Post finally caught up with what we've been talking about here at VDARE.COM since the election: the National Election Pool's exit poll claim that 44% of Hispanics voted for George W. Bush is implausible. It is now being viewed skeptically by professional pollsters. [Pollsters Debate Hispanics' Presidential Voting, by Darryl Fears, November 26, 2004]
Now I can explain why the exit poll results were internally inconsistent.
Of course, this won't make any difference to the Bushies. They will believe what they want to believe.
The problem with the exit polls: Bush's reported shares of the national and regional Hispanic vote were inflated compared to the sum of the state-by-state numbers.
As you may recall:
The cause of this internal discrepancy is that the national and regional numbers are based on
Across the 50 states plus Washington D.C., a total of 76,298 voters filled in exit poll questionnaires. Of these, 62,638 respondents (82 percent) answered the short form questionnaire, which contained only about two dozen questions. These 62,638 responses were used in calculating the state results—but not the regional and national results.
The regional and national results came from just the 13,660 respondents who filled in the long form (about 60 questions).
Clearly, this methodology is just asking for trouble.
One obvious source of error: there might not have been a large enough sample size of Hispanics among people who filled in the long form and thus got included in the national/regional results. The sample size of "national" Hispanics would have been around 1,100—bad, but not as good as the 3,700 Hispanics in the broken-out states.
Edison-Mitofsky, who conducted the NEP exit poll have a Frequently Asked Question list that includes this answer:
"The margin of error for a 95% confidence interval is about +/- 3% for a typical characteristic from the national exit poll and +/-4% for a typical state exit poll. Characteristics that are more concentrated in a few polling places, such as race, have larger sampling errors."
While telephone pollsters generally use random dialing to get a representative sample, exit pollsters have to guess ahead of time on the sample of voting stations they'll send their pollsters to.
Race/ethnicity poses a particular technical problem for exit pollsters—as opposed to, say, gender. Minorities are distributed in a lumpy fashion across the landscape. This increases the chances of coming up with an unrepresentative sample of that minority. For instance, if they send a worker out to measure voting at a military base, they are likely to come up with a lot of minorities, but also more conservative minorities than in the overall population.
The Edison-Mitofsky FAQ offers an additional warning: "Other nonsampling factors may increase the total error."
For example, the difference between the lengths of the questionnaires could have caused the skewing of the results.
We know that Hispanics who are well-educated and work in white-collar jobs tend to vote Republican more than Hispanic manual laborers. Perhaps GOP-voting Hispanic office workers breezed through the long form while Democratic-voting laborers found it more intimidating, and thus were less likely to turn it in.
That's pure speculation on my part. But it could explain why Bush was more popular among the "national" Hispanics who filled in the long form than among the "state" Hispanics who only finished the short form.
Skepticism about Bush's share of the Hispanic vote is spreading. Pollster John Zogby, for example, told the Washington Post's Darryl Fears in the article I mentioned above that he believes Bush's true share of the Hispanic vote was only 33 to 38 percent. As Fears reported, that's also in line with Bush's share of 34 percent in the Velasquez Institute exit poll. (For comparison, the 2000 VNS poll showed Bush with a 35 percent share of Hispanics.)
Zogby's estimate sounds a little low to me. I suspect Bush gained among Latinos in 2004—but only by about what he picked up among everybody else, i.e. about three or four percentage points.
For example, Robert David Sullivan of Massachusetts's Commonwealth magazine put together an interesting map of the U.S. divided into ten regions based on county-level voting patterns.
His "El Norte" region consists of most of the heavily Hispanic counties in the U.S. In this sprawling region along the Mexican border, where one-third of the population is Hispanic, Bush won 44.10%, up 3.33 points versus 2000. In the whole country, he won 51.03%, up 3.15 points (all results as of a few days after the election).
So Bush ran 0.18 points better in El Norte than in the nation as a whole—i.e., virtually the same, suggesting there was no Hispanic surge toward Bush.
(Or, possibly, a hypothetical boost among Hispanics might have been balanced by a drop among non-Hispanics.)
"… if 44 percent is the wrong level for Bush's support among Hispanics, what is the right level? Of course, we'll never really know for sure, but I am persuaded, by playing with the numbers and making some reasonable assumptions to correct the anomalies in the NEP that it is somewhere around 39 percent."
That would be up four points over 2000's 35 percent, the same growth as in the non-Hispanic white vote (from 54 percent to 58 percent). It would confirm the general pattern that the Hispanic vote for Republicans rises and falls in the same cycles as the white vote—just consistently much more Democratic.
And, of course, as the Hispanic vote is swelled by immigration, narrowing the relative gap could still leave the GOP deeper in the hole in absolute numbers of votes.
Here at VDARE.com, we don't.