Oversampling Just Isn’t that Hard to Understand People
Some of the dimmer lights in the blogosphere aren’t quite understanding the concept of oversampling for subgroups as NBC/WSJ did for the African-American sample in their latest poll.
It’s fairly simple. There is a general sample of 700 respondents with 11% African-American and 75% White respondents. Looking at their subsamples, there are 520 white respondents which is approximately 75% of the whole sample. If you read what the WSJ and Chuck Todd say is that they added 100 African-American respondents to the crosstabs–or the breakdowns by race. This means that in the general sample there are 77 African-Americans and in the smaller African-American sample there are 177 African Americans.
Before trying to discredit the poll or acting all outraged, do the math. All one needs is a basic understanding of percentages.
Furthermore, Taylor Marsh is very upset that the poll includes Republicans. I kid you not. She might read the poll results with questions from the article she linked to and notice that it only includes Democrats and likely Democratic primary voters if they identify themselves that way. But shiiiiiiiiittttt, we’d hate to read the damn thing and know something about what we are talking about.
March 27th, 2008 at
Man oh man. The second I read that Chuck Todd piece, I knew, KNEW the oversampling would be the talk of the tubes.
Morons.
March 27th, 2008 at
but … math is HARD. And percentages are so … old fashioned; grade school kids would rather learn algebra.
At least that’s what my high school math teacher friends tell me they encounter when freshmen show up knowing basic algebra but unable to understand fractions/percentages.
March 27th, 2008 at
So…why did they add the hundred in there? I don’t get it. This is my understanding, tell me if I’m wrong:
The 77 black respondents were kept in the “topline” of the poll, but in order to get a better understanding of what the black population in general thinks, they “re-included” the 100 in the crosstabs, because otherwise the sample would have been too small?
In any case: who cares? It’s a poll, not an election. People’s obsession with polls, particularly so far away from an election and in the middle of a primary fight, are beyond meaningless. The horse race stuff has spiraled out of control.
Wasn’t the margin of error over 7% anyway?
March 28th, 2008 at
Yes, but the margin of error would have been near 12.
It’s a fairly simple concept that is used quite frequently for small subsamples. It simply improves the validity of the findings.
I actually think this is a far more interesting type of polling because it does look closely at a subgroup and their attitudes instead of just giving you a quick yes or no–I wish we did more of it because I think it teaches us a lot more about subgroups than general polling.
March 28th, 2008 at
==At least that’s what my high school math teacher friends tell me they encounter when freshmen show up knowing basic algebra but unable to understand fractions/percentages.
Much more analytical students, but not so much good at basics that you need to make good analysis.