The NC divergence is particularly interesting to me, of course. My gut (which has been way wrong before, so) says that Kay Hagan is going to win because the state GOP has managed to piss off Democrats at every available turn, but Wang's odds are much more favorable than I'd have expected. I sure hope he's right. Thom Tillis is a [expletive] jackwagon.
I don't know who is right, but I'm happy to side with the optimist.
I'm with Drum in every particular. Die by poll aggregation, live by poll aggregation.
1.
i hope Hagan wins, but her ads are soooooo lame. "I'm the most moderate Senator, and proud of it", "Not too far to the left, not too far to the right", zzzz.
It'd be awesome if Nate Silver's inevitable comeuppance comes from a democratic when when he's picking republicans. My impression from prior elections is that Wang is just better at this than Silver. In particular Silver's models are unnecessarily messy and his error bars are too big (that is he was underconfident in Obama winning last time). But this election is hard and on a 60/40 bet there's just a great chance of the wrong person winning.
I haven't been paying terribly close attention, but my generally pessimistic perception is that the Democrats won't win because not enough groups are pouring money into street-level GOTV efforts.
My record on political predictions is comically bad, though. So I'm just putting this here as another bit of evidence in that direction.
This topic reminds me of that guy with the "Unskewed Polls" website. That was great.
My unscientific gut feeling is that the Dems' hopes for holding on to the senate will depend on a) nothing really bad happening in the last six weeks before the election, and b) one or more of the R senate candidates coming down with a classic case of foot-in-mouth disease.
There's probably several people who are right now hooking up R candidates to electrodes and shocking them every time they say the word "rape."
9 -- Yeah, but that's going to vary quite a bit by state. I doubt Amanda Curtis is going to see much national money at all, but there's some real national interest in our state legislative races, and we have a shot at a tie or better in at least one house.
In other political news, it looks like PA is finally expanding Medicaid, sort of. The waiver that just got approved by the feds may still throw a number of current Medicaid recipients under the bus.
Don't worry. They cut transit funding. There is no bus.
not enough groups are pouring money into street-level GOTV effort
Ferguson PD and lobbies for deporting central american kids to the rescue on this matter, maybe.
8: How do you know that Silver was underconfident? IIRC, one of the factors he tries to incorporate in his model is a fudge factor for "the polls are systematically wrong in some way that is correlated across polls." In an election where the polls aren't systematically wrong, that's going to look like underconfidence relative to the polls. In an election like Reagan-Carter 1980, where the last public polls prior to the election were calling it a toss-up, that might look like sheer genius. (Or maybe not, since his fudge factor wouldn't know which way the polls were biased, so he would probably still call it a toss-up. Genius would show up if Reagan had been down by 3-4 points in the polls prior to the shift towards him that occurred over the final weekend, according to Carter's private polls.)
I haven't gotten much past keeping my fingers crossed that we'll lose McConnell at last, but for once it really does seem possible and that's an awfully good feeling.
Wang is not bullish on Grimes' chances (16%).
Cf. the thread on rape stats. The Senate Minority Leader, a conservative republican in Kentucky, is as likely to lose his seat as I am to get a 6 the next time I roll a standard die? I call those pretty good odds!
In an election like Reagan-Carter 1980, where the last public polls prior to the election were calling it a toss-up, that might look like sheer genius.
I thought polling had gotten so enormously better that reaching back this far for examples wasn't meaningful.
19-20: I know, right? This is as good as it gets! I'm not even joking!
17: I certainly don't know that Silver's underconfident. Definitely he's less confident than the other major modelers (Wang and Linzer) and I haven't seen an argument from Silver for his position that I buy. Might be prejudice on my end, as I assume that the academics are better qualified and I trust them more, but I certainly could be wrong.
21: Right, that's a big part of the dispute. On the one side you have "polls have gotten way better" and on the other side you have "maybe something's happened to make polls worse, just like things happened before to make them better." My feeling is that although there's some chance of an error like this, I think Silver is overstating it.
It doesn't require a poll aggregator to know that it basically comes down to the Democrats winning two of North Carolina, Arkansas, Alaska, and Louisiana, all of which are basically coin-flip races (I think they will.) The problem with all the Monte Carlo simulators is two-fold -- with four roughly fifty-fifty races, a couple bad polls (and there are some incredibly bad polling operations out there now) starts skewing things pretty badly, and some of the odds on the non-headline races are laughable; as I said, if I could find anyone giving me 70% odds on the Democrat in the Michigan Senate race, which is where the Post's Monte Carlo simulator had it a few weeks ago*, I'd put Jane's college fund on it**.
* "Although national conditions and the state's partisanship tilt Michigan toward the Democrats, basic information about the candidates' political experience actually tilted our April forecast to the GOP." What this says to me is that there is something wrong with John Sides' formula.
** They now have Michigan at 99% Democratic retention, so I guess the race has gone from competitive to a lockup in a month and there's nothing wrong with their simulator, nope nosiree. Also, the 65% odds on a Republican win in Iowa are a joke, so that's how you get to a 60% likelihood of a Republican takeover in your predictor even giving Kay Hagan a 99% chance of a hold, which is also a joke.
Sorry, 90%. They're saying North Carolina is an uncompetitive a race as New Jersey. I think Hagan is a likely winner, but doesn't that sort of counterintuitive result make you suspect that you r doin it rong?
I'm on the side of the people with lower confidence in their own beliefs.
I'm probably missing something, but can't they look at the past and see how accurate purely poll-based versus polls-and-fundamentals-based models would have been in past years with the data available at this time of year? Are there not enough years with sufficient data for that exercise to be conclusive?
21, 24: I don't know that any advance in polling techniques could overcome "a massive number of previously undecided voters make up their minds for Reagan over the final weekend," short of flash polls Sunday night/Monday morning. Most of the major polls are still out in the field over several days. Silver surely doesn't have the probability of systematic error exactly right, but it is almost certainly greater than zero (which is what anyone without a fudge factor is implicitly assuming).
To take two examples that he's written about that were particularly relevant to the 2008 election (and to a lesser extent, to 2012):
A. As of 2008, many major pollsters were not surveying people who did not have a landline. If the political opinions of non-landline users had differed significantly from landline users, this could have resulted in a repeat of the Literary Digest fiasco from 1936. (How's that for reaching back several years for an example?)
B. As of 2008, the "likely voter" models that most pollsters used did not take into account the possibility that the person surveyed had already voted in early voting. Using historical patterns to estimate that there is only (say) a 40% probability that the voter you are talking to will actually turn out to vote seems obviously wrong if they have already voted.
23 Might be prejudice on my end, as I assume that the academics are better qualified and I trust them more, but I certainly could be wrong.
Silver's book and especially his take on climate change didn't really convince me that he's very trustworthy, but it seems like the issues there are orthogonal to the ones that separate him from the other modelers.
I'm not sure why, but Wang strikes me as having a more obvious Democratic bias in his judgement, so I'm skeptical that the Dems will hold the Senate.
My friend Andrew Thomas looked at the precision question back in 2012, and found that Silver was pretty damn well calibrated. At least, he was for the state-by-state presidential results; I don't think Andrew ever looked at other races.
32: jeez, that is pretty fucking good
Yeah. Looking back at his final data, his uncertainty dropped a lot in the last week, and I think I'd forgotten that and was thinking more about his numbers a week before the election which were even more conservative.
9: my generally pessimistic perception is that the Democrats won't win because not enough groups are pouring money into street-level GOTV efforts
A related point -- which I don't see addressed in Wang's piece -- is the extent to which the currently existing polls on which PEC's and DKos's current projections are based focus on registered voters or likely voters. Ed Kilgore had a piece recently pointing out that we're only now getting close enough to the election to be able to sensibly poll just likely voters (the only ones who actually matter).
I don't doubt that Wang is aware of this, of course.
Another item of interest, in terms of particular races and their impact. Martin Longman has a fairly fascinating piece on the state of affairs in the Kansas Senate race. It's a bit into the weeds, but short version:
Kansas has already had its primary, and the Dem and Republican candidates have been chosen. Turns out there's a third-party candidate, an independent, Greg Orman, who polls better against the Republican than the Democratic candidate does, by quite a bit. Should the Dem, Taylor, oh, drop out of the race, say, Orman could win. He might could caucus with the Dems (he's a former Dem) if so. This would effectively turn Kansas blue, sort of. But he might not caucus with the Dems, hard to say. Longman thinks that Harry Reid and other ranking Democrats should be having discussions with the Orman and Taylor camps in Kansas.
In reviewing the Longman piece I see now that it links to Sam Wang on this matter. So there you go: Wang's got his hat on straight.
Anyone whose statistical model has "secret sauce" in it should be laughed at cruelly until they stop saying that.
Two all beef patties, special sauce, lettuce, cheese, pickle, onion in a special regression.
Hey, that "secret sauce" could be totally reasonable and appropriate stuff. I mean, you don't know because it's secret.