Earth to Gallup!

By Andrew Sullivan, The Daily Beast

28 October 12

here are two snapshots of the polling right now. One of them comes from Rasmussen and Gallup, which show a clear lead for Romney; the other comes from all the other polling organizations that show a slight Obama lead. Rasmussen, a firm headed by a committed partisan Republican, whose polls are the basis of countless polling quotes from the Republican blogosphere, has nonetheless been removed from Fox News' polling unit, which tells you something. Even Fox News doesn't trust him. And they shouldn't.

Gallup, though, is a different case. There's no indication of editorial partisan bias - or an obvious strategy to provide polls for purely propaganda purposes for one party. Just a party identification sample that produces different results than every other polling organization. Nate Silver noted:

[Gallup's] results are deeply inconsistent with the results that other polling firms are showing in the presidential race, and the Gallup poll has a history of performing very poorly when that is the case.

To wit: in 2008, Gallup missed the margin of victory for Obama by six points. That's a huge discrepancy. And that time it was in Obama's favor: they predicted a landslide double-digit 11 point victory, when it was only 7 percent. But they were dead-on in 2004; and now they are claiming that the demographic make-up of this year's electorate is almost exactly the same as they measured 2008. So with the same turn-out in terms of race, age, gender etc. they are predicting this time a Romney victory over Obama of five points. From 2008 to 2012, Gallup believes that Obama has lost 17 points to the Republican challenger.

Gallup weights its sample for measures party identification, but doesn't put much emphasis on it in their model. They say their model for 2012, with the same demographics as 2008, is 36 percent Republican; 35 percent Democrat; and 29 percent Independent. The poll of 705 other polls shows party identification as 29 percent Republican; 36 percent Democrat; and 31 percent Independent. What has happened since the summer is a sharp drop in the "Independent" category - giving gains to Democrats and Republicans pretty evenly, with the Democrats gaining a tiny bit more.

Here's the Gallup/Rasmussen analysis of the race since September, if you just use those two polling outfits:

Here's the same graph in the same period for all the other polling organizations combined:

In the model with the least smoothing, Gallup and Rasmussen have shown a clear Romney lead since the beginning of September. Obama has never led the polls on that graph since September 1:

In all the other polls, on the same unsmoothed graph, Romney was only ahead from October 6 - 13; and then briefly on October 23.

So you either believe that Romney has held the national lead 100 percent of the time since September 1; or you believe that Obama has had the lead for 86 percent of the time since September 1. Obviously, the two models cannot both be true.

They are in completely different universes. Well, as Larkin put it, we shall find out.