Review of The Signal and the Noise
In Science magazine, Ben Campbell and I have a review of Nate Silver’s book, The Signal and the Noise. Briefly…it was good for people wh...
Senate: 48 Dem | 52 Rep (range: 47-52)
Control: R+2.9% from toss-up
Generic polling: Tie 0.0%
Control: Tie 0.0%
Harris: 265 EV (239-292, R+0.3% from toss-up)
Moneyball states: President NV PA NC
Click any tracker for analytics and data
Michael Mosettig explains it well. I cannot improve on what he says. Go read that. Then follow live returns here and projections here.
Current projections by electionforecast.co.uk indicate considerable losses by the two ruling parties, the Conservatives (281 seats) and the Liberal Democrats (27 seats). With a current forecast total of 308 seats, these two parties together seem likely to end up below the 323* seats necessary to get a majority in Parliament by themselves. But the uncertainties are rather considerable.
The projections have high uncertainty for two big reasons:
The other total to watch is Labour + Scottish National Party (currently projected at 266 + 51 = 317). The remaining 20 or so seats would be split among the U.K. Independence Party, DUP, Plaid Cymru, and various others. These minor parties may end up being important after the election, if the likely Conservative+ and Labour+ coalitions can’t get to 323.
If all of the above is true, Labour seems to have a slight edge in forming a ruling coalition. However, the Conservatives could well find a way to remain on top, especially if they end up with the most seats as an individual party – at which point they may assert bragging rights to form a government.
*One minor party, Sinn Fein, may end up with about 5 M.P.’s, but their policy is one of abstention. Thus only 323 out of 630 votes appear to be necessary to form a government.
Probably worth noting that electionforecast.co.uk itself notes that its predictions are far less reliable than U.S. efforts such as PEC and Linzer. There are good reasons for that (e.g. sparse and tiny-sample polling for individual constituencies), which are elucidated on their site. But it means that they expect to be off by as much as 50 seats either way for the two major parties. That’s a huge swing in an election as tight as this one.
For what it’s worth (probably not more than the above) the latest Guardian-sponsored polling puts Labour and Conservaties both at 270 seats come Friday morning. Wouldn’t that make things interesting. Well, we’ll soon have the proof of the pudding.
Good point about the uncertainties.
I correct myself on one point. The Guardian bases its ‘it’s going to be a tie’ prediction on this: “a weighted average of all constituency level polls, national surveys and polling in the regions, sees both the Conservatives and Labour winning 273 seats . . . .”
The pollsters are covering themselves in caveats. Pollsters function well when the political climate is in equilibrium, changes in voting (who votes, how they vote) are adiabatic.
This election is the textbook definition of non-adiabatic. See Scotland.
The SNP’s stated most important goal was to lock the Tories out of a second term (a worthy goal!), but in bitter irony that gave Cameron the winning card of ‘don’t let Labour try to govern while on an SNP leash.’ Miliband could simply find no way not to be pressed up against that SNP third-rail effect.
Looks like nat’l % close to polls for majors but seats vary widely from predictions.
Yet another election where conservatives outperformed the consensus of polls. In the case of the Israeli elections the narrative was scare mongering in the last few days before election. There may be a similar spin here. It will be just as wrong.
There is a systematic problem with likely voter models. Reality seems to have a rightward skew compared to predictions.
Now that is an interesting take: US midterms, Israeli elections, and UK general elections. On the other side we have Democratic outperformance in 2012. At what point is there a built-in problem in opinion polls?
I agree, likely-voter models need a hard look.
Regarding “jerseyvotes”:
The SNP gained 50 seats (they now have 56) with a vote total of ~1.5M. The UKIP gained one seat (they now have 1) with a vote total of 3.8M
A “scotsvote” is worth much more than an “englishvote”
In the US we’ve seen that rightward skew relative to polls in midterm elections, but not in recent presidential elections. What’s the difference, and does it apply to elections elsewhere? Lower-turnout elections would require better modeling of likely voters, but was the UK election low-turnout?
Apparently it was the highest turnout since 1997. So that is a source of polling error.
Another possibility is the undecided-voter break. The Scotland referendum did worse than expected, seemingly because of a last-minute break.
Twitter used to forecast UK elections:
http://www2.warwick.ac.uk/newsandevents/pressreleases/computer_scientists_use/
Sorry if it’s old news.
–bks
Once again, the betting prediction sites did better than the pollsters. Karl Rove would kill for this kind of math.