[Guest post by Noam Scheiber:]
Much of Washington--this resident included--spent the morning clucking over the just-released Gallup poll showing Democrats down ten points in a generic contest with Republicans. This, Gallup helpfully informs us, is the largest GOP margin in the poll's entire history, dating back to 1942.
My first reaction was similar to Nate Silver's: It's highly unlikely that Democrats are actually down ten points--any given poll is likely to be an outlier if it, well, lies outside the range of recent data points. On the other hand, even if it's off by a few points, being down five or six at this stage can hardly be comforting for Democrats.
Anyway, Silver then noticed something I hadn't:
Gallup’s survey — and some other generic ballot polls — are still polling registered rather than likely voters, whereas its polls of likely voters are generally more reliable in midterm elections. At FiveThirtyEight, we’ve found that the gap between registered and likely voter polls this year is about 4 points in the Republicans’ favor — so a 10-point lead in a registered voter poll is the equivalent of about 14 points on a likely-voter basis. Thus, even if this particular Gallup survey was an outlier, it’s not unlikely that we’ll begin to see some 8-, 9- and 10-point leads for Republicans in this poll somewhat routinely once Gallup switches over to a likely voter model at some point after Labor Day — unless Democrats do something to get the momentum back.
Basically, the Gallup poll isn't yet reflecting the near-certain GOP turnout advantage. Once you factor that in, things start to look really grim for Democrats. (And I've been one of the few people around the office arguing that they can hold the House...).
Update: Ah, I see that Cohn picked up on a different portion of Silver's excellent analysis.