It also mentions Scott Brown and Mitch McConnell.
You must have missed it in your haste to post.
It mentions them only by *speculating *that they *might *do the same, but they *haven’t *yet.
Looks like you’re even hastier than me, hmm? Basic rule of the Dope, you gotta read your own links before posting them, or shit like this happens to ya.
It speculates that they might do so, and then follows up with the actual videos that Brown and McConnell ran.
It sounds like you’re reluctant to read the source and insist on denying that it contains what it contains. But anyone else can read it and watch the videos at that link. So I’m not sure how much that will get you.
Amazing.
You seem to be constantly surprised and amazed on the Dope. Is that what keeps you coming back?
I was kind of surprised that you felt only political junkies would care about that kind of stuff. Tying Senate Democrats to Obama is exactly what Republicans have wanted to do, and in an election that is historically a referendum on the incumbent, is precisely what the average voter is paying attention to. Obama basically just confirmed what Republicans have been saying and Democratic candidates have been desperately denying. It’s awesome 30-second ad fodder, right up there with McCain saying “the fundamentals of the economy are sound.”
Even if Wang turns out to be right, Silver pointed out the major flaws in his model. Polls in June are meaningless today. Silver pointed out his other problem a couple of months ago: he was relying on a “polls only” forecast when there was a paucity of good polling. GIGO.
Actually that’s the key question. Wang’s premise is that “current conditions” only become more meaningful than longer term trends (June through September) in the last several weeks and otherwise represent more noise than signal and that even a few decent polls outclass fundamentals for predictive value at even an early stage. Is he right or wrong? I see no reason to strongly conclude one or the other based on actual evidence.
As for the nerd-off there are a couple of metrics to consider using:
-
How well did early predictions hold up to conditions immediately before the election? What we see there is that so far both Wang’s and Silvers models have moved by roughly the same amounts, just from different sides both to the middle. Early added fundamentals seemed to result in early predictons as far off from any consensus of later predictions as the simpler polls only approach even when fairly few polls were extant.
-
Over many individual elections how many races do they call right and how do their hit rates match with what their error bars would predict? Here I believe Wang has done marginally better on the first part - in 2010 senate both missed Colorado and Nevada and in 2012 Wang missed none which Silver did not accomplish. For the second part Wang on Colorado and Nevada had error bars way too narrow and Silver’s error bars have seemed to be a bit too wide overall. Again, it will be interesting to see how their performance on the Senate races match up in that metric this time.
Bottom line remains that there is no evidence that Silver’s extra machinations bring any value to the table over a more basic meta-analysis of statewide polling data. And Wang’s premise that June to September polling long term is at least as predictive of conditions in November as short term polling conditions and as polling plus “fundamentals” until the last month before the election (when short term polling becomes more predictive) seems supported by the fact that all models have changed by about the same amount.
Once again, if a simpler model performs as well as a more complex one, the simpler model is to be preferred. Silver has morphed from a nerd taking on the pundits into an insecure media brand. He may have once been David taking on Goliath but now he comes off more as the later King David who had Uriah murdered after he got Uriah’s wife Bathsheba knocked up. Celebrity and power has not done him well.
Well, let’s look at the 2016 polls. We’ve got a thread already that assumes a big Democratic landslide because Clinton leads all her likely opponents by a pretty big margin. However, the “fundamentals” heading into 2016 look pretty bad for Democrats. Historically, a party is only rewarded with three straight terms in the White House if the incumbent is going into the election with a 60+% approval rating. Another factor is name recognition. There are Clinton supporters in those polls who pick her, there are Clinton haters in those polls who just pick whoever she is not, and then there are the on the fence people who either choose Clinton because they know here and not her opponent, or people who choose undecided because they don’t feel they know enough to make a choice.
Should we assume that Clinton’s poll numbers are the likely result in 2016, or should we take into account the historical factors that make it unlikely that she’d win by that much?
Obviously fundamentals are not very useful two months before an election, but two YEARS or, even six months out, they probably tell us more about a race than the polling. For example, as recently as four months ago all the Democratic incumbents were leading in their races. But Silver, who has studied election trends, would have told you that polling when these incumbents don’t even have an opponent yet is of limited use. Once they got opponents, most of those incumbents fell behind. Hagan has been an exception, but Udall, Pryor, Landrieu, and Begich all now trail and Shaheen is also looking threatened.
Wang’s numbers are converging towards the consensus, not visa versa. As of today it doesn’t look good for the Princeton approach. I was prepared to accept Wang’s version of the world a month ago. But the evidence isn’t pointing in his direction.
Nobody is an idiot here: frankly I believe the alleged Nate/Sam spat is overblown and a product of gossipy reporting. On the contrary, putting forth reasonable hypotheses and comparing them with factual developments is a good thing. As is putting one’s money where one’s mouth is.
Let’s keep our comparisons clear.
Wang compared to other polling onlys consensus. The difference in technique being his belief that longer term data is more predictive than short term fluctuations (the issue in any meta-analysis is what studies to include) and others using shorter term data only. There Wang has moved a bit more than them.
Silver compared to the non-Wang polling onlys. The difference being Silver added fundamental. There Silver has moved a bit more than them.
The other polling onlys haven’t moved much and Silver and Wang have both moved by something like 10 to 15% in their odds of Senate control, with Silver now saying the almost exact same thing as the polling onlys have been saying all along.
Some of that remaining difference explainable by how Orman is handled differently by the different sites.
So I am left looking for evidence that Silvers extra machinations add very much in predictive value, do more than fill up blog space much as talking heads pontificate.
And I see little evidence as well for Wang’s beleif that longer term is better than ishorter term polls either. N of one for each so no conclusions at this point.
The evidence however does not point to either of them as any better than other simple short term polling meta-analysis approaches earlier on.
There is powerful evidence that fundamentals matter. Fair first published his fundamentals model in 1980. He revisits it every 4 years. He has tweaked it only once: we don’t have an overfitting problem.
However, according to Nate Silver fundamentals are fully reflected in polling late in the game. The question is how fast should you depreciate the fundamental aspect. I haven’t read any papers on that and frankly it’s probably a matter of judgment. But Sam Wang’s erratic record at least provides a little insight. Ultimately though, I’d like a third party to review these various attempts and set up a scoring system moving forward.
The incentives break both ways incidentally. If fundamentals matter, that might imply that you need a small staff to maintain the model: a part-time amateur like Sam Wang would be shut out. While I trust his statistical knowledge, Wang isn’t a political scientist and AFAIK hasn’t combed through that literature. Nate isn’t a political scientist either, though he has apparently read the relevant academic papers. This doesn’t mean Nate or Sam are correct: I’m just saying they both have incentives for their stances. And Wang’s stance remains an outlying one, which btw I’m not quite dismissing.
Both analysts are worth reading. And the current difference between a 51% probability and a 58.5% take simply isn’t very large. Furthermore, horserace commentary by the statistically illiterate should be ignored.
Their predictions are going to change as the elections near and by the end they are both probably going to be right, arguing about who is right at this point is meaningless because there is not going to be an election right now to settle it.
It does matter. Wang’s model was showing pretty solid odds of Democrats maintaining control. Now, he’s moving more towards the other models. Has the election really changed since March? No. Nothing surprising or unpredictable has occurred. Silver’s model has been pretty steady as a result since he started it. Wang, on the other hand, is making a much bigger shift despite no major changes occurring in the national environment. That’s because his model is only worth anything starting in mid-September.
The problem with Silver is that he can never be “wrong.” If he predicts a 60% chance of a GOP takeover of the Senate and it doesn’t happen, well, there’s your 40% at work. The model was still right. If the GOP does take control of the Senate, then he’s Mr. Magoo the predictor of the future.
Huh? He’s bounced between 65 and 53%. Yes that is less than Wang’s move from 70 to 51% but seriously not so significantly so. A movement of 19% bothers you lots but 12% is “steady”?
In point of fact polls HAVE moved since then. That said Silver’s movement is because he peels off some of the fundamentals that he based his early prognostication on as the election gets closer (moving more to polls only) and Wang’s movement because he switches to shorter term polling data as more predictive as the election gets closer.
Actually best I can see the winners here by the least change metric have been the other polling based prognisticators who have between the two of them throughout.
I say maybe. But I don’t think it changed in September as much as Wang’s model implied. Moving from 30% to 51% in a couple of weeks without major events doesn’t sound right, especially when 538 bopped around within a 5 point range at the same time.
The problem with weather forecasters is that they can never be “wrong”. If they predict a 60% chance of ran and it doesn’t happen, well, there’s your 40% at work.
It’s not Nate. It’s the nature of probability. Both Wang and Silver in fact can be evaluated rigorously because they make specific forecasts and lots of them. There is a science to that and it’s called -wait for it- forecast evaluation.
Wang’s movement isn’t done yet. Unless things change, and they aren’t likely to, Wang will be moving another 10-15%.
The polls have changed, but the national environment has not. The change in the polls was entirely predicted by Silver’s model, which is why he’s never wavered from Republicans having at least a 53% chance of taking the Senate.
Oh that’s silly. Wang has “never wavered from” the Democrats having at least a 51% of keeping the Senate leadership. Both Silver and Wang moved towards 50/50 by fairly comparable amounts. If Silver predicted the changes in the polls he’d have been saying this beore instead of 65%. He did not. He like Wang stated the obvious: lots can and likely will change before elction day; the situation is fluid.
They are both now within 10% of 50/50. HuffPo is also within 10% of 50/50. NYT too. The Washington Post is the current outlier at 78% for the GOP as they are the most still fundamentals heavy of the batch.
BTW, this chart from Chris Cilizza? Absolutely hilarious:
Interesting question brought up in another forum: assuming the Republicans do grab the Senate this year: what does that mean for Senate control in 2016, when a bunch more seats come up for grabs? How much of a cushion will that offer them when those elections come around?