WAR SUPPORT
Kevin Drum posts some interesting survey data on public sentiment over the war.
Not surprisingly, given the steady drumbeat of depressing news (except for Uday and Qusay being “otherwise dealt with”), support is down now compared to the euphoria of May. Indeed, I’m surprised by how little real variation there is in the results.
Kevin qualifies the methodology, but there are some reputable researchers associated with it. But, even if we accept the methodology, there is “a margin of error of +/- 3 to 3.5% (depending on which question asked),” according to page 3 of the PDF.
Let’s assume it’s just three on the support question. Those who still think the war was the right thing to do are then better expressed as a range:
7/03: 42 to 48
6/03: 43 to 49
5/03: 50 to 56
Theoretically, this is just a 2% drop since May (from 50 to 48). If we take it at 3.5%, this is technically a statistical virtual dead heat.
Update (0847 7-31): Minor correction in the last sentence as per John Lemon’s comments. And I don’t want to be misleading: My guess is that enthusiasm for the war has in fact declined; I’m just saying that as a matter of mathematics, the poll here doesn’t necessarily demonstrate much of a drop.
And the day Saddam dies it jumps 10+ points.
EOD
It’s true that there’s a statistical range of error, and in fact the range is actually quite a bit bigger than 3% given the real world constraints of polling.
However….comparing the bottom end of May to the top end of July is definitely not kosher. It’s certainly *possible* that there’s been no (or little) change from May to July, but it’s not very likely.
Kevin,
Most polling houses artificially inflate the margin of error for caution, thus the +/- 3% is a conservative range estimate. They make certain assumptions about the underlying population that helps to expand this (e.g., using a p – .5 for dichotomous questions rather than the actual sample result). Plus they bump it up a little bit as well.
James,
Ooooooh, you might want to remove your last sentence. You perform statistical tests on the sample means, not on the upper and lower boundaries of the confidence intervals. Just eyeballing these numbers I can tell you the result is statistically significant at the .05 level (if not .01), so this is not a “statistical dead heat.” And the drop is moderately substantive, meaning that there has been a real shift in public opinion that could affect policy or elections. Bush is losing support on the war, but so what — this should not be unexpected for the reasons that you mention.
If you want a better comparison or how public opinion is holding up or not holding up, compare this to the polling numbers for Kosovo, Somalia and the first Gulf War. Granted the former two were not as salient as this war was (is), but the numbers could be telling.
Hey, it looks like some fuzzy statistics are happening over here, James. I accept your interesting point about the poll results, but you are comparing the wrong numbers between May and July!
You should be comparing the 50% (lower bound) in May to the 42% (lower bound) in July, not 50% to the 48% (upper bound) in July.
A change from 50% to 42% seems like an actual decline in support for the war to me, even with the margin of error considered. Then again, I’m not a big fan of relying on polls for accuracy.
Kevin: Yep, you’re right. I don’t mean to imply otherwise. The center of the range is the best guess, but anywhere on the range could be the actual result.
John: You’re right, too. I’m using journalist-speak rather than stats talk. I’m trying to convey that the results could be virtually the same because of the ranges.
Anthony: No.
—