Thursday, November 08, 2012

Wait 'til they hear about Santa

What's that, National Review Witchsmeller-General? Something about that thing you wrote the other day?

According to exit polls, the partisan breakdown on Election Day was 38 percent Democratic, 30 percent Republican, and 31 percent independent. That gave Democrats an 8-point advantage — the same they enjoyed in 2008. (In 2004, Republicans had a 5-point advantage in the Buckeye state.)

So, as someone who defended questioning the scrutinizing the polls’ partisan breakdown in their samples, I decided to look at the most recent Ohio polls, and see if they accurately called the partisan breakdown. Here’s what I found:

  • Public Policy Polling D+8 (43 percent D., 35 percent R., 22 percent I./O.)
  • Columbus Dispatch D+4 (40 percent D., 36 percent R., 21 percent I.)
  • SurveyUSA D+5 (39 percent D., 34 percent R., 25 percent I.)
  • Gravis Marketing D+8 (42 percent D., 34 percent R., 24 percent I./O.)
  • NBC/WSJ/Marist D+9 (38 percent D., 29 percent R., 32 percent I.)
I admit it: when I saw the NBC/WSJ/Marist partisan breakdown, I was fairly confident that turnout on election Day would not mirror that. But Marist was only off by a point.

Without any idea of what the standard errors* are in the exit polls, we can still draw a couple of conclusions. Self-report of "independent" identification appears to be consistently lower in preelection polls than in exit polls, and in most of those cases,** the difference looks likely to be significant. Self-report of Republican identification appears consistently higher in preelection polls, but only at the Republican-owned Dispatch (with the largest sample in the bunch) does the difference look like it'd be significant at 95% confidence. The best relationship appears to be between people who say they're Democrats before the election and people who say they're Democrats when they leave the voting booth. No, results don't bias the sample, because there is no such thing as time travel.

If you start with the intuitively sensible assumption that party identification isn't the same thing as voting intent, it's pretty easy to grasp the idea that party identification is more flexible. For one thing, it doesn't have to stop changing, whereas your voting intent has to turn into a fixed vote at some point before the polls close, or else you turn into a pumpkin.*** If you have to believe in a sinister global conspiracy slanting all the empirical data because that's the only way your world makes sense, on the other hand, I hope Santa is good to you this year.

OK, that's a nice start. For dessert, there's this gem uppage, from an unusually**** spittle-flecked tirade:

As it turned out, the polls were right all along, and the prolonged delusion that they were somehow “skewed” turned out to be a disastrous bedtime story.

"Prolonged self-inflicted delusion" would be more inclusive, but there you have it: If you'd rather lie about other people's work than look at the evidence, then you have few other people to blame for your choice in bedtime stories.

In the best-nommed-cold department, your attention is directed to the author's plaint from the previous day, "Crush Them." The author might sensibly conclude that if he wants the peasants to take up their pitchforks and follow him to the gates of Castle Husseinstein, he might want to stop crushing himself first.

I'll make espresso!

* For practice, calculate the standard errors for the preelection polls from the data here. Which two of those could you multiply by 1.96 to get the most appropriate "margin of error"? And remember, friends don't let friends say "within the margin of error."
** I don't know why the writer omits the Rasmussen survey that was in the field Nov. 4; maybe it's bad salesmanship to mention Paris today at the National Review.
*** And you probably deserve it.

**** Sorry, I keep thinking Bill Buckley is still enforcing manners at the place.

Labels: ,


Post a Comment

<< Home