Perhaps in AK, LA, and AR, there’s a 60% chance of the Republicans winning the election in that state; that means there’s only a 21.6% chance of winning all three, and if the Democrats win one, it’s 50-50, which becomes 51-50 Democrats because of Biden.
THat’s not how national elections work though. If you have six 50-50 races in a given year, it’s not the equivalent of six coin flips. What actually happens is that on election day, turnout or the prevailing winds of voter sentiment can be a point or two in one direction or the other and one party wins most or all of those races. That’s why Silver has, or had few days ago, a 25% chance of Republicans winning 53 or more seats. If national sentiment moves as little as 2 points in favor of Republicans by election day, they sweep the 50-50 races and perhaps take a couple of the ones that looked barely out of reach, like NH and MI. Likewise, if things go 2 points towards Democrats, they only lose the three seats and probably take either KY or GA. Which I guess means they also get KS, which turns their 3 seat “minimum” loss into break even or -1.
I do find it interesting, and telling much (none of it good) about Silver, that Silver has gone out of his way to disrespect, well trashtalk, the lesser known Wang.
He then, it seems to me, misrepresents some information in his attempted takedown. As responded to by Wang:
Personally I think that is … polite. It gives credit to the possibility that Silver does not know what he is saying is factually untrue.
To me it reads that Silver is butthurt that Wang had predicted the winners with high probabilities in 2012 in North Dakota and Montana while 538 had called those races wrong with equally high probability. Or maybe feeling insecure in his media darling status.
Wang, to his credit, is taking the high road. His take is to highlight the difference between models based exclusively on polling (in my mind the equal of an index fund) with those who use varying degree of “fundamentals” to adjust the polling data, peeling it off some as election day gets closer (in comparison stockpicker funds).
Again, to my read both have been saying that this is pretty much a toss up and both have very good records, Wang’s arguably a tad better, albeit also imperfect. And to my read when two models perform not too far apart from each other the simpler model is to be prefered.
Wang stood alone saying that reports of the death of the Democratic Senate majority were more than somewhat exaggerated … or at least very premature. Now Silver and the rest of the fundamentals added herd have had to move toward Wang’s stable election day forecast. It seems like that bugs him.
adahar honestly I am not so sure how tightly correlated or independent AK, LA, and AR are with/from each other. At this point I suspect most prevailing winds will be very local in origin and minimally correlated. Minimally I see no reason to believe that there are any likely events that can occur to move sentiment by fairly evenly in any direction across the various in-play states at Senate levels.
What is interesting is that Wang uses a much simpler model than Silver. It’ll be interesting to see who does better this year. Add in Rothenberg’s prediction based on experience and call it a three-way rumble of prognosticators.
First of all, Wang could use the publicity. Secondly, it’s not just trash talk. Thirdly, this is how science progresses. Fourthly, it’s not just trash talk: [INDENT] In 2010, for example, Wang’s model made Sharron Angle the favorite in Nevada against Harry Reid; it estimated she was 2 points ahead in the polls, but with a standard error of just 0.5 points. If we drew a graphic based on Wang’s forecast like the ones we drew above,35 it would have Angle winning the race 99.997 percent of the time, meaning that Reid’s victory was about a 30,000-to-1 long shot. To be clear, the FiveThirtyEight model had Angle favored also, but it provided for much more uncertainty. Reid’s win came as a 5-to-1 underdog in our model instead of a 30,000-to-1 underdog in Wang’s; those are very different forecasts. [/INDENT] Clicking the link, here’s what Wang wrote on November 1, 2010:
“NV: Angle (R) over Reid (D) by 2.0 +/- 0.5%.”
Wang’s claims about current/non-current don’t apply that late in the game. The above prediction suggests some serious mis-estimation of the error term - that was an 8 standard deviation advantage that the defeated Angle allegedly had!
Look: Maybe Wang has improved his error estimation. But set that aside. Getting these models right takes time and is part art, part science. It wouldn’t surprise me if Nate is over-weighting fundamentals. But I predict that fundamentals will still be a discussed issue in, say, 2030. Whether they will be set aside in September 2030 is something that we’ll have to wait and see.
Interesting times!
Well, it’s really too early to say. We’re discussing maybe a week of trending. But if these trends continue through, say, Sep 27 then maybe we’ll have some interesting evidence for the way the world works.
I reject that analogy. Economic performance during the election year (plus incumbency) is a powerful predictor of Presidential vote share. Cite. It’s just that polls reflect all of that by the time October rolls around. Ok, ok one problem is that we only have wobbly forecasts of GDP in March of the election year. But I’m not arguing relative predictive power. I’m arguing that underlying fundamentals are demonstrably and statistically important. It’s not just stock picking. Stock picking is what they do on the Sunday talk shows when they blovate on the effects of a politician’s gaffe on the election. That has no real evidential basis: it’s just noise.
But statistically measuring the effects of incumbency and the last election’s voting patterns is doing something very different.
Oh I certainly would not claim that Wang’s model was or is perfect. Indeed even he acknowledges that his model’s MOE assumed normal distribution and that the real world seems to behave more with t-distributions (longer tails). Yes, if the polls systematically miss something (which sometimes happens) then a polls only approach will be faulty.* Of course Silver got the Nevada race wrong too. He is lording over that he had merely stated the Reid was likely to lose with a 5 to 1 level of confidence and the polls only method Wang used then (since changed) gave a standard of error number that translated into an absurd level of confidence. Weak sauce that Silver, while wrong with great confidence, was less dramatically wrong in that one race, and no endorsement for all the special sauces he adds. A valid reason for the polls only method to recognize that the election world is not Gaussian, yes. Probably in addition to modeling with t-distributions one should realistically state that no prognostication can exceed some upper limit of confidence given the possibilities of unknowns. Only experience will end up telling how often those unknowns occur and conservatively topping off a model at say 95% to 98% seems reasonable to me.
Every stockpicker I know and have read of will quote you reasons why their methods historically are valid, or at least have been.
Whatever you are arguing, the question is if (once some reasonable mass of polls are available) the simple polls-only model performs as well, if not better, than trying to improve upon what it tells you with various contortions. If after all the hosts of contortions and exposition your model ends up no better than the simpler one then why bother with the contortions … the only reason is that the simple model does not story tell so well and as pundits need poof to fill up the hour, bloggers need something to bloviate about.
*Anybody who ever collects any data should know that your first hypothesis to explain a very unexpected result is that there is problem with the measurement, be it a poll that says the result is known to 99.999% (or whatever) confidence, or faster than light neutrons.
Colorado now a true toss up:
http://www.realclearpolitics.com/epolls/2014/senate/co/colorado_senate_gardner_vs_udall-3845.html
The bigger impact news is that Chad Taylor is now off the Kansas ballot. That substantially lowers the GOP’s chances of holding Kansas. Of course if Orman caucuses GOP then they are golden.
Newest Alaska poll, FWIW, has Begich up by 5. Trendline of HuffPo running up by 1.4%, by RCP down by 1.3%, not listing the most recent poll.
Wang has promised to eat a bug if Brown beats Shaheen.
Republicans may have found a smoking gun against Udall in Colorado. he promised when running in 2008 that he did not support a “government solution to health care”. Then he not only voted for ACA, he supported a public option.
Whether or not one believes his position was the correct one, it’ll be easy to portray him as a faux moderate and a very dishonest one at that.
Add voting to authorize the war, along with immigration, as another thing the Democrats are too cowardly to do until after the election:
You know, if they don’t want to lead, we can oblige them.
LOL. This is satire right?
It’s pretty easy to get out of this – he can say he meant a government takeover, or something, like single-payer.
Problem is, that still pegs him as too liberal for Colorado. He tailored his message in 2008 for a purple state electorate and he’s still trying to do that. Gardner, unlike many Republicans, is doing a pretty good job of sounding moderate and reasonable. Given Udall’s gaffes lately, I think Gardner might topple him.
Last two polls put Gardner ahead. Quinnipiac has him up by 8, but I’ll believe that when I see more polls showing that. Quinnipiac says it’s their likely voter model, before they were using registered voter models, but they must be predicting a massive turnout discrepancy to get some of the results they are getting in favor of the GOP lately. They’ve also got Hickenlooper down by 10(!) and Ernst up by 6 in Iowa.
I’ve learned my lesson about the polls as a whole being skewed, but an individual pollster certainly can be. Quinnipiac might be doing something wrong.
Considering Obama’s margin of victory (5 points – not huge, but solid) in Colorado in 2012, it’s silly to say he’s “too liberal for Colorado”. The race is close, but you’re getting way ahead of yourself if you’re confident Udall will lose.
I am not confident, I just slightly favor Gardner at this point. Udall’s shooting himself in the foot too much and Gardner’s running one of the best campaigns on the GOP side in the country.
As for whether or not Colorado is a purple state, Udall felt the need to use the “I don’t support government health care” talking point as opposed to the “I support universal health care” talking point we saw in bluer states.
It’s a purple state, it’s just really silly to say Udall is too liberal for Colorado. Not even close.
Well, at least more liberal than he wanted to portray himself as then.
And on NPR this morning, GOP Congresscritters were freely admitting that the reason they weren’t shutting down the government this October* is that there’s an election coming up.
So you might want to amend your point from “Democratic politicians are afraid to do anything controversial with an election coming up” to simply “politicians are afraid to do anything controversial with an election coming up.”
*Note the key word ‘October.’ They’re fully funding the government through…December 11, rather than for the entirety of FY 2015. Apparently they’re preserving the option to shut down the government when the next election is a nice, safe 23 months away.