I keep reading national polls pitting Obama vs. McCain and Clinton vs. McCain, and it strikes me that these are being drawn from the same survey samples. That is, Gallup or WaPo or Rasmussen or whoever asks a selection of people, “Who would you vote for if the election were held today and the choices were X and Y?”
The problem is, the composition of the electorate will be different in November depending on whether Obama or Clinton is the Democratic nominee. It seems pretty clear that young people and African-Americans would make up a larger slice of the electorate in November if Obama is the candidate than if Hillary is. Yet I don’t believe any of the polls we’re seeing reflect this.
Is this right? Or are pollsters adjusting their samples to account for the somewhat different electorates the two possible general election matchups would generate?
This seems pretty important. Recent polls mostly show either Obama or Clinton slightly ahead of McCain in the national horserace. But unless pollsters are adjusting their samples, i.e., the 1000 or so likely voters surveyed for Obama-McCain are not precisely the same 1000 likely voters surveyed for Clinton-McCain, at least one of these snapshots will be based on an inappropriate sample.
Any thoughts on this?
Are you familiar with this website?
<
p>This anonymous statistician does some great analyses.
<
p>In the upper right of the blog’s pages are some pie charts comparing the Obama/McCain match-up with Clinton/McCain.
<
p>Clinton comes out looking slightly better, and my hunch is that no adjustment was made for the factor you mention.
<
p>I don’t have time to dig into this, but maybe someone out there in BMG-land knows the answer or can find out…
Krugman on his blog (could dig this up if you want) pointed out two things:
Reinforcing your point, I read that cell phone only households tend to break much more for Obama than land line households. So, yes, there is some evidence that polling methodology might undercount Obama’s support. On the other hand, there is a well-known phenomenon whereby people (or maybe it’s just white people, don’t know) will report more support of an Afro-American candidate to a pollster than their actual voting reflects.
…to be accurate, even as a very early snapshot, you need to approximate the electorate as it will, in fact, turn out. If Obama’s the nominee, the people who show up in November are likely to be somewhat younger, and have a higher percentage of African-Americans, than if Hillary is nominated. I have heard it said, though have no way of knowing, that Hillary would generate more evangelicals turning out to vote against her than Obama would. I’m sure there will be other differences too.
<
p>My point (and it’s really a question) is a methodological one: pollsters probably need to rethink how they construct their samples based on the particularities of this year’s contest, not simply tinker with their 2006, 2004, and 2000 turnout models. Failure to do this makes horserace numbers vs. McCain suspect, and makes comparative Democratic horserace numbers (how Hillary and Barack each do against McCain) misleading.
<
p>Disclosure: I am an Obama supporter, and believe improved drawing of samples would show him beating McCain by more than current polls do, and by more than Clinton beats McCain as well.
remember media encompasses polling organizations also and these organizations support the very same status quo establishment, keep the population stupid memes.
http://www.projectcensored.org
One of the places promoting the blacklisted “news” topics.
To me, the incredible growing inaccuracy of polls has been driven by every two-bit network conglomerate – FSNMSNBCNN-M,O,U,S,E – announcing that they have polls.
<
p>We HAVE pollsters – Rasmussen, Zogby, Suffolk, ARG, etc. We know their track record and methodology and can make our own internal mental corrections based on past accuracy. Now, EVERYBODY’S a pollster!
<
p>Hell, if the price is right, I can pay to have a question added to State House News, and trumpet the results as from a ‘respected Boston polling organization’.
<
p>Publius – you’re looking for a comfort food push poll in service of Obama-Magic, and if you ask the right folks, they’ll give it to you. Won’t mean much in November, but it might make you feel beter now.
Hell, if the price is right, I can pay to have a question added to State House News, and trumpet the results as from a ‘respected Boston polling organization’.
<
p>Tell me more…
What does Ernie charge to add a question to his polls?
I disclose that I’m an Obama supporter and you accuse me of looking for a push poll? What’s that about?
<
p>My argument was essentially the following:
<
p>(1) if you’re going to do a scientific poll, you need to start with a sample that you believe approximates the eventual electorate;
<
p>(2) if pollsters are using the same samples to trial-run McCain-Obama and McCain-Clinton they are making a mistake, since those two matchups are likely to yield somewhat different electorates; and
<
p>(3) that if the pollsters are basing their samples on earlier year’s presidential elections then they are likely to be undersampling young and black voters if we end up with a McCain-Obama matchup.
<
p>Peter, if I’m missing something you’d be among the first to spot it. Methinks you don’t like the implications for your favored candidate of a younger electorate with more African Americans in it. Who’s looking for comfort food??