September 3, 2004

For Democrats looking for ways to spin the devastating Time poll.

I emailed the link to the Time poll to my son, John Cohen, and he wrote back what seemed to me to be a fine effort at putting the best face on it. So, with his permission, here it is:
Clearly Bush has a really good trajectory. Remember that in 2000 Nader got surprisingly few votes. There might be a lot of people defecting from Nader to Kerry at the last minute out of pragmatism. Nader has 3%; assuming (arbitrarily) that two-thirds of Nader voters will end up defecting to Kerry, then we should consider Kerry to be at 43% right now instead of 41, which would mean we should consider the split among likely voters (which is the headline of the story) to be Bush 51 / Kerry 43. The margin of error is 4%, so that's a statistical tie. (Even if you don't buy my calculation about Nader voters, it's still CLOSE to a statistical tie.) So I would consider this one poll obviously a great sign for Bush, but it's not proof that Bush is in the lead. Bush can only be considered to have a strong lead if other polls corroborate this one.

He adds:
The poll was done from Aug 31 to Sept 2. It will be interesting to see what the numbers are like after the convention. As I've said before, considering that Kerry got almost no bounce from the convention, I think it will be terrible news for Kerry if Bush gets a significant bounce out of his convention, since people's views of Bush are more solidified. If Bush doesn't get much of a bounce out of his convention, then you could say that both conventions were ineffective for reasons unrelated to the specific candidates or parties (the country is polarized, the conventions got relatively little coverage, etc.).

Yes, when Kerry got no bounce, the spin was: conventions don't really produce bounce anymore. So what can you say now? I predict: the Republicans did very bad things at the convention and thereby unfairly obtained the bounce that they got; if they had conducted an honorable convention, like the Democrats, there would have been no bounce.

UPDATE: An emailer observes:
I wanted to note this:  a 51/43 split with a margin of error of +-4 is not a statistical tie.  It does indicate that there is the possibility that there is a statistical tie in the actual populace, but there is no reason to think that the likelihood of this possibility is any different from the likelihood of a 55/39 split in the actual populace (which would render Kerry about as significant as I am in the coming election).  Moreover, neither of those extreme possibilities is as likely as a 51/43 split in the actual populace.  Similarly, rolling a sum of 2 on two dice is as likely as rolling a sum of 12 is, but neither is as likely as 7.  Obviously, the dice rolling is not correlated in the same way that the polling is, and while this correlation does support your son's comment, it does so only weakly.

I agree, definitely. But what I like about my son's comment is that it's amusingly in the vein of what would Susan Estrich say if she had to figure out something positive to say. I'll see if John comes back Estrichishly and re-spins.

ANOTHER UPDATE: If you want to shore up your knowledge of what "margin of error means," go here. And Matthew Yglesias argues about that here. I don't think you really need to invest a lot of time in studying all that, but if you're bummed out about the poll, it might cheer you up. Especially the first link. I'm pretty sure John is literally correct about the meaning of the term "statistical tie." At the same time, the emailer is right about likelihood of different results. But if you were trying to spin the bad news, you would avoid pointing that out. Indeed, I think Yglesias, who supports Kerry, was saying what he was saying because he was enjoying polls that showed Kerry ahead and he didn't want his pleasure spoiled by people wielding the "statistical tie" concept.

No comments: