Math Is More Fun If You Only Count What You Like
It’s time to start reading Nate Silver really closely if you haven’t been already. You’ll recall that he pegged the 2008 election within one vote or something crazy like that. Today, he is especially interesting to me as he sheds light on something I’ve been wondering about.
As we know, Democrats and Republicans fixate on different polls. Republicans trumpet Rasmussen, which always leans in their direction. Democrats prefer, uh, most anything that isn’t Rasmussen. When all these polls come out showing big Obama leads, conservative pundits are on their twitters within minutes tweeting “sample is D+8. In Virginia??? REALLY?????” and so on.
Often they have a point. This new Marquette University poll of Wisconsin, for example, showing Obama up 54-40 looks odd to me. Conservatives note that the Democratic sample in the poll is +11 (that is, 11 percent more Democrats than Republicans), whereas previous MU polling has been more like D+2, D+4. That’s a fair point.
I’m not of course charging that they’re cooking. Usually pollsters just ask respondents “Do you consider yourself” a D or an R or an I, and they use those results and then do some weighting. It’s all a bit technical for your servant here. Now, it’s true that there simply are more Democrats than Republicans, and more Democrats will vote than Republicans on Nov. 6; in 2008, it was D+7, according to the exit polls, 39 to 32 with 29 percent independents. But I would concede as a general matter that some polls have had what seems to be a high D count. But there’s another side to the coin, and here’s where Nate comes in.
Some polls have figured out how to call people with only cell phones, and some haven’t. Generally, it’s the robo-callers, who don’t use live humans to do the interviews, that don’t bother with cells. Rasmussen is chief among these. Many other firms do call mobiles.
And unsurprisingly, Obama does much better in polls that call cell phones (younger people, more urban people, etc.). So Nate ran all the numbers in his usual thorough way (the link is up there in my first graf) and devised “Obama Win Probabilities” in key states in polls that don’t call cells and polls that do. Some results are amazing. LL below means polls that call only land lines, CI means cells included.
Ohio: Obama win probability is LL 57.3 percent, but CI 78.9 percent.
Wisconsin: LL 66.7, CI 90.0
Virginia (look at this one!): LL 42.2, CI 80.5
Florida (and this one too): LL 35.3, CI 67.8.
And so on. Of the 11 states he ran, there was only one where the second percentage went down. In Colorado, it was LL 59.6, CI 59.1. What does that say about Colorado, I wonder? That young people out there are more libertarian?
In any case, this obviously makes one wonder if Obama’s lead isn’t even bigger in some places than we think. And the next time I see a conservative whine-tweeting about how unfair and bias X poll is, I’m gonna remember all this.