Ed Kilgore’s post on the November Congressional makes a lot of good points; the whole thing is worth a read.
But what struck me most was this:
much of the “overestimation” of Democratic strength in past generic polls has involved early tests with no “likely voter” screen. As we get closer to Election Day, the Gallup generic ballot is usually quite accurate (as shown some years ago by TDS contributor Alan Abramowitz of Emory). So it’s not a good idea to just mentally add a few points to Gallup’s number for the GOP and assume that’s close to reality.
This is worth more stress than Ed gives it. Likely voter screens try to figure out who will vote in November so as to poll only them. (Rasmussen, which uses a stringent screen, polls incessantly and is therefore overrpresented when sites like pollster.com or RealClearPolitics take poll averages.) But the well-known finding that Democrats had to be leading in the generic Congressional ballot to hold the house was based on past Gallup polls, which didn’t use a screen. Democrats needed a big edge to overcome their historically lower turnout in midterm elections. But likely voter screens, if they’re working as advertised, already control for lower Democratic turnout.
Pundits are routinely double-counting greater Republican enthusiasm.