Sam Wang just tweeted that in retrospect, he might have put different polling error in his model, leading to probability 91-93% instead of 99%+…
???
I don’t even see how a ceiling of 322 adds up given a look at the electoral vote map. That isn’t even a likely number comprised of a likely combination of states.
Clinton’s CEILING is quite obviously higher than 322. If the polls underestimate her by 2-3 points and Latino turnout is high, as we’ve seen evidence of, her ceiling - the best that is reasonably possible - is probably 359; that would be all battleground states, plus Arizona, plus NE-2.
Her FLOOR is clearly lower than 278. If everything goes wrong, she would lose the election, which is less than 270.
No contradiction to my read.
His point is that the varying the states together is simple and effective. Any random amount would be fine but keeping it simple he sets it at 100% covariance
I don’t read it that way, but you’re more familiar with PEC than I am.
In my view, there is a clear reason to prefer a model that thinks OH is more like PA than CA, or that ME-2 is more like ME-1 than FL.
But like I said, we’ll probably have a chance to compare the models since they have differing calls with respect to some states/districts.
Yes, as stated, RCP was significantly off. But they are not a true aggregator. Silver, Wang, HuffPo … none were off by that much, and none are calling the popular vote as close.
538 was off by 1.4. He currently predicts a 2.9 point win. But insists on large error bars. Still he’d need an error twice as large as his previous biggest.
Wang was off by 1.7. His meta-margin is currently 2.7 points (not quite the same as the popular vote margin but for our purposes good enough). He uses smaller error bars.
HuffPo was off by more: 2.4. Not quite RCP bad and their call is currently a 5.2 point win, so they would also have to off by more than twice as much as they were last time.
You’re forgetting that Maine splits its electors - try again.
Yes, she could win Ohio, Iowa, Georgia, and Arizona, but none of those are very likely. Ohio is somewhat possible given the fact that she has a serious advantage on the ground in terms of organization. But here’s the deal: Trump has already won races against an entire republican field with little or no organization. So organization isn’t everything. Yes, a general election is different from a primary, but that difference isn’t going to help Hillary win red states, and Ohio and Iowa are becoming redder and redder.
Well of course she could lose. What I’m saying is that based on what I am seeing now, I don’t see her going lower than 278. I’m bullish on Nevada and New Hampshire. Now if she loses either of those states then it’s either 274 or 272 depending on which one she loses. If she loses both, she could actually lose the race altogether.
In other words, he was really wrong after all.
Told ya.
Imagine that Silver and Wang freeze their predictions here and the result is 2.8. Would Silver score as strongly as Wang?
Look at it another way - the point of probabilities is to help decide risks, in one explicit manner making bets, but also for investors who view a Trump win as something that might tank certain investments. Having a good estimate of how big that risk is guides rational decisions.
The betting market reads Silver … and Wang, and The UpShot, and all the rest. They’ve discounted Silver’s assessment this time. Predictwise puts it at Clinton 86%, closer to The Upshot’s 84%, Lizner’s 90%, even closer to Wang’s crazy 99% (thanks squidfood, that seems more realistic), than to Silver’s 65%.
BTW PollyVote, which also has a great track record (outperformed 538 in 2012 for example), calls it C+5.4.
Remember we also have firm EV predictions. Wang is sure Trump won’t reach 240, Silver’s forecast has Trump above that and his no toss up forecast is 272-266. If that’s the end result we might lose some Dopers to heart failure overnight.
Wang might end up being right in that Trump won’t get to 240. If Clinton wins Florida, then he’s probably right. Right now Florida is too close to call so if you’re gonna make a call why not wait until the last batch of polls are released on Monday.
Clinton will not lose an EV in Maine if she wins that many electoral votes.
Well, although 538 still persistently has Trump up around 35%, I’m somewhat relieved to see that in the betting markets Trump has drifted significantly. He topped out at implied odds of 30% when he rallied last week, but has dropped since then, now at 22%.
Interesting article about why Silver’s calculations lean way more to Trump than the others.
Nate Silver Is Unskewing Polls -- All Of Them -- In Trump's Direction | HuffPost Latest News?
Silver responded with a furious tweetstorm saying the article is nonsense. Because reasons. I don’t know who to believe now. Just hope Hillary wins.
well, if Silver’s wrong it would be the first time. Wang was wrong 2 years ago.
I mean, Silver claims he’s tested everything and his data supports his results … but if it was all based purely on hard evidence, wouldn’t everyone have the same results? Does Silver have some source of hard data that the rest of the field doesn’t? Silver says it’s all math - I mean, he’s not the only guy who can do math. So why is his math right and everyone else’s math wrong?
I have no idea. I’m not really qualified to discuss the math. I’m curious how others would answer those questions.
Everyone uses their own proprietary algorithms. Silver is pretty transparent about how 538’s is constructed and why it does what it does. I’m sure Wang’s is transparent too, but he doesn’t explain it well to laymen. Damn professors.
But they all rely on a few different assumptions. Silver’s is the only one I know of taking into account the uncertainty of undecided voters and third party voters, whereas the other models just sort of hand-wave it away and assume undecideds will split evenly. Silver says he’s using the same simulation model that predicted a 90% chance of an Obama win at this point in 2012.
Huffington Post wrote a pretty ignorant article about how Silver “skews” the state polls. He does that because the state polling is not as good as it was in 2012, so he has to figure out how the state polls change as the national polls move. Obviously if Clinton drops 3 points nationally, then a poll showing her 8 points ahead in Michigan last week isn’t telling us much useful. So Silver’s model extrapolates what a 3 point drop in Clinton’s support does to her numbers in Michigan.
Now it’s true that Silver is “unskewing” the state polls to show Trump ahead in places where he isn’t. Wang is taking the polls at face value. Silver is adjusting them based on the trendline. So someone’s going to have egg on his face in a few days.
I think Silver and Wang will have to settle this the old fashioned way: Epic Rap Battles of History.
That piece by HuffPo was a bunch of Huff Poo.
Nate Silver is saying Hillary is the favorite, and there’s no doubt in anyone’s mind that she is, but he’s spot on about the fact that the other aggregators are putting down a lot more certainty into their results than they should. I’m struggling to find out where the evidence to support certitude is when state polling is scant and relies on instruments like Google Consumer Surveys and Survey Monkey, with polls showing either candidate up by double digits. Sorry, but you can’t call that certainty. And on the other hand, in Florida, a slew of recent polling produces a tie or +1 advantage for either Trump or Clinton - all except one poll. Where’s the certainty? Yes there’s early voting and we can draw some conclusions from that but those are far from final.