Where's the Generic Ballot At?

Are the Democrats going to take the House this November?  Good lord, I hope so.  But by all indications it's going to be a squeaker.  In a lot of ways, we have the wind at our back.  The Republicans aren't even contesting a bunch of safe Democratic seats, which means that we get to save our resources for other fights.  And upwards of 40 Republicans are retiring or trying for higher office.  Incumbency gives something like a 7.5% advantage, so that's actually a big deal.

Approach 1: The Generic Congressional Ballot

But the big number is, and has always been, the Generic Congressional Ballot (GCB).  This is a weird sort of polling question, and almost all pollsters ask it a bit differently.  Quinnipiac, for instance, asks:

If the election were today, would you want to see the Republican Party or the Democratic Party win control of the United States House of Representatives?
while Monmouth asks:

If the election for U.S. Congress was held today, would you vote for the Republican or Democratic Candidate in your District?
Those are actually very different questions. Arguably, the Monmouth wording is better, since it asks what someone is actually going to do, rather than what they want to happen.  For what it's worth, Quinnipiac has yielded results about 1 point better than Monmouth this cycle (but both are consistent within the noise). Overall, the average of the GCB from top pollsters has been remarkably stable during this cycle, fluctuating between a Dem advantage of about 6 to 12.  From my low-noise tracker:

We're near-ish to the bottom of the range, but currently on an upswing.  Fivethirtyeight.com produces a similar (but noisier) result.

But there's good reason to suppose that the GCB questions are fundamentally flawed. I've heard people snarkily comment that to Dems, every generic candidate is JFK, while to Republicans, every generic candidate is Reagan.

Take a look at some composite data of where the race stands now (first two lines only for the moment):

GCB (me)6.4
GCB (538)6.4
Specials (Uniform) 12.2
Specials (Model) 14.4
House Poll (Uniform) 12.6
House Poll (Model) 10.0
House Polls (Individual) 10.0

If every seat in the country swings by about 6 points, and does so uniformly, that's not enough to win the House.  At that level, Dems would win about 209 seats, roughly 8 shy of winning the speakership.

But there are reasons to suppose that the current Dem advantage is much, much better.

Approach 2: Special Elections

Democratic candidates have been killing it in special elections.  In part, that's because these are all open seats, which means that the incumbency doesn't come into play.  In part, though, it's because Dems are more energized and prepared to vote, something that's hard to predict accurately in the GCB question.  Special elections are a measure of what people do rather than what they say they'll do.

The plot at the top of the page shows how specials have turned out (D-R margin) compared to expectations (based on the 2016 election and the most recent election in that district).  Even without fancy numerical tools, you can see that there's a big shift, by about 12.2 points (see the table above).

FWIW, a 12 point uniform shift would give Dems 225 seats – easily enough to take the House.

But the swings aren't likely to be uniform.  As I noted in the earlier analysis, Dems tend to overperform more strongly in districts where the Republican lean is especially strong, kind of like a regression to the mean.  I also toyed with including the swing from 2012 to 2016 as a parameter, but found that it didn't provide any useful data.

Plugging this model in produces:

The post-hoc model is a very good fit to the data.  It also produces an an effective average shift (basically, the reddest district that we can expect to win) of about 14.4, enough to win 233 seats (flipping 38 of them).

Approach #3: Individual District Polls

There is one final approach that I think is better than either of the other two: use polling, but not in the generic sense.  I'm maintaining a database of polls of individual districts, and comparing those results with what we'd expect in a neutral year  By "expect," I start with the 2012 and 2016 presidential results and correct for the fact that nationally the dems won the popular vote in both years.  I then add the incumbency advantage and estimate from there. The trend is unambiguous.  Dems are significantly beating the neutral map by about 12.6 points, easily enough to take the House (if this model holds):

Using the more sophisticated linear  model, above, I find that the effective  swing is more like 10 points, just about exactly what is needed to hit the tipping point for the House.  There's a problem with this sample, however, as pollsters tend to focus on seats held by Republicans where the expected margins are Republican by 10 or 15 points. Sine the range of districts is relatively narrow, it's hard to constrain the outcomes as a function of historical performance.

There's one other possible way to look at this data, which is to confine it still further and look only at the seats where an incumbent is running, and compare the polling directly with their results to 2016.  In that case, we get a swing of just about exactly 10 points as well – but again I give you the same caveats about sampling effects above.

So are Dems currently poised to win the House?  Honestly, it depends on what dataset you find most persuasive.  Personally, I think they'll do it, but only if you get out there and contribute, volunteer,  and vote straight ticket and get everyone you know to do the same.