Doubting the statistical analysis at 538

Politics & Elections
41

You need to free your thinking from the faux-objective punditocracy.

Clinton winning the nomination was predictable. Trump winning the nomination was predictable, in fact if you read the never-go-out-on-a-limb pundits since he actually did win, it was INEVITABLE.

Clinton sitting back and letting Trump act like an ape is predictable. That a majority of people don’t want an ape as President is predictable.

She has enough money and volunteers to identify and Uber every swing state undecided to the polls and buy them lunch afterwards.

42

Not that she would, mind you! That would be wrong. 'Course, if some wild-ass PAC were to do it, not much she can say about that.

43

I’m guessing the co-variance effect is modeling for an October surprise. If some candidate falls on their face for some reason or soundly wins a debate, probably all the states will move in the same direction.

44

Automatically adding an October surprise into your model would be completely ridiculous.

45

I think they just model in the possibility.

46

The model includes covariance because the things it is modeling are covariant.

47

asahi, I shouldn’t be pressuring you to be as confident as I am. I’ll stop.

48

The fact is that adjacent similar states are* not going to have wildly disparate results*, so if one of (say) three states goes for candidate A by x% more than the polls indicated, the probability of another state in that group having a substantially different change than the x% is low. You likely should look at groups of similar states as likely to show similar shifts in the end result - which means they are not independently random events. Right now, that would seem to be only about 4 groups of states.

Nate Silver makes this fairly clear in his projections.

49

Look at Rothschild’s estimates of state voting to see that Nevada and Iowa have much better pro-Trump numbers than they had a few weeks ago. If Trump wins any other key tipping states at all, he’ll win those two as well.

This makes it easy to see what the election hinges on. Much as before, Trump wins if he takes any one of Pennsylvania (12%), Virginia (10%), New Hampshire (10%) or Wisconsin (9%).

You can’t sum these four numbers to get a 41% Trump chance — that would assume maximum anti-correlation. But it’s easy to guess that, with correlation adjustments, the chances will add to about the 26% GOP White House chance that Rothschild currently quotes.

One can hope that those quoted state chances are way too pessimistic, but that wasn’t OP’s question. OP just wants to reconcile the relative odds. This is achieved when we note, as above, that Trump has fair chances in PA, VA, NH and WI and wins if he wins any one of those states.

50

Trump wins if he gets New Hampshire’s 4 electoral votes? How does that work?

51

I think what’s he’s implying is that:

If Trump wins ANY of PA, VA, NH, or WI, all 4 being lean Dem:
[ul]
[li]It means he’ll also win the rest of those 4[/li][li]Plus OH and/or FL[/li][/ul]

52

Maybe, but since other people said similar in this thread it kinda looked like he specifically did not say that. Regardless, it’s pretty banal to say “if there is a massive swing in polling across the board Trump may win”. The last few polls have her lead by 8-11 (!)% in NH an PA. Much lower in Wisconsin so I don’t see how swinging that state means all 4 go with it.

53

If Trump wins a state (NH) that he is only 10% to win, he still may be a longshot (say, 25% or thereabouts) to win a state ¶ that his prior odds are 11% to win. However, winning NH makes it very likely that he will win NV (26% prior) and Iowa (37% prior). He won’t be 100% to win those states, but let’s keep it simple and, anyway, I guess Silver’s covariance matrix is secret.

So NV and IA are the two swing states (along with OH 30%, FL 31%, NC 43%) that Trump is most likely to win. If he does get these five then the addition of any other swing state will give him the Electoral College.

Even itsy-bitsy New Hampshire? Yes — that state gives him a 269-269 electoral vote tie. The E.C. tie will almost certainly put the Republican in the White House (though possibly putting Tim Kaine into the V.P. slot.)

54

So what kind of swing in polling do you think that would take? Like 4-5% across all those states? How often has that happened at this point in the race?

55

Clinton leads by 3.4% with 60 days to go and a debate still to come. And you’re complaining that 75% is too low?

56

4.5% in 538’s polling-only model, but that’s a quibble. Go with the RCP one that overweights the Rasmussen and IBD ones. S’okay.

Not complaining, opining. Yes I am opining that 25% is an overly generous chance for Trump with only 60 days left and debates rarely change anyone’s mind. I take national number drops into the 3s as seriously as I took the jump up to the 8s. Both are noisy blips around the long term equilibrium of polling in what is in fact a fairly stable race with variability that is low as it has been since 1992. Moving a wee bit outside of the 5 to 7 range will happen in each direction, but more than 3ish either way from that long term stable point of Clinton +6 (and not outside that now) would be quickly followed by a regression to that mean.

Then look at states that matter and assume strong if not complete covariance. Move the national number up that … we are calling it 3.4%? … to a tie. Historically moving that much from this point is … uncommon. NH is still over Clinton +5, PA is still Clinton +2 to 3, VA is still +2 to 3, NV still Clinton, OH and FL are toss-ups but give them to Trump, NC to Trump probably, others all pretty much stay as is. Give Trump the close NC, both OH and FL in toss-ups, and NV which would be leaning to Clinton … and it is still Clinton 273 to 265. Which is his other state going to be? He has to also overperform, beyond the unlikely national shift, in NH (for a tie) or PA (for the win).

Yes, I know Silver, in his “there is too a horse race” narrative, emphasizes that she is not, on average, that much stronger in the key swing/tipping point states than she is in the national polls, but she is in enough of them. Specifically she is significantly stronger in PA , NH, and VA than she is nationally. He could theoretically run the table with the others (FL, OH, NC, and NV) and still would lose.

Again, no complacency. This needs to not be just a win but a historic repudiation of what Trump is selling and represents. And if believing it is close gets you addie to vote Clinton over wasting your vote on Johnson, sure it is close. But yes I think a nearly 28% chance of a Trump victory is an over-estimation, even if Wang’s 95% Clinton win is high.

57

Okay, so what is the mean we’re supposed to be reverting to? As the race has tightened, that’s been a claim of a lot of Clinton rooters. Okay, so where is the race supposed to settle? Where it is now? At +2? +1?

As for a historic repudiation, that ship has sailed by your own argument. if Trump can’t make up 3 or 4, Clinton can’t gain 3 or 4. I think this race still has some volatility in it personally, so Clinton certainly could win by 10 points or more, but as of right now it looks like it’ll end tighter than Obama/McCain, and possibly tighter than Obama/Romney.

58

I think polls ending up Clinton +10 is just as unlikely as Trump ending up polling in any positive number.

Again, as stated, I see the long term mean is that Clinton +5 to 7 in the two-way race polling numbers, or call it Clinton +6 +/- 3 (historic range for polls to vary from their long term mean since 1992). Maybe even only +5 with the same +/- reviewing the graphs now.

I would be shocked if the race ends up polling outside of the +/-3 of Clinton +5 for any significant length of time.

You can think that race has been volatile but the fact is that by historic standards your thoughts would be wrong. Overall it has been pretty stable with the only turbulence being the expected turbulence around the conventions, with a smaller and briefer GOP bump and a bigger and longer democratic bump before settling back in. Yes it could change but there is no reason to think it will.

Now I am open to being surprised on election day with a result not exactly as polls would call it, by turnout factors, and again, my WAG is a better than polling prediction for Clinton, but that’s hard to call. I want the blow-out and have small hope that is could occur. But rationally it is very unlikely.

59

BetFair has GOP White House at 28%; other prediction markets are even higher than that. Nate Silver also shows 28%. (Rothschild shows 24% — I don’t know where he gets that; maybe he just feels the higher numbers are ridiculous.) Are these estimates reasonably accurate or are they are way off? If you’re convinced they’re way off, you owe it to yourself to arrange a large wager at BetFair, or some other sportsbook. Even after paying round-trip airfare to London to place your bet, your expected profit (even risk-adjusted) from betting on Hillary will far far outweigh your gains on the much-touted Vanguard Index fund.

And don’t kid yourself. 28% is Huuuge. Out of 435 races for the House of Reps there are only 25 where the underdog’s chances are 28% or more. It’s because 28% is huge that for every $10,000 you’re willing to risk on Hillary, you can walk home with almost $4000 profit.

Or, you can accept that the 28% figure is reasonable and ask Why? There are various reasons why polls might be inaccurate. Some pollees may be ashamed to admit they’re going to pull the Trump lever when alone — it would be like admitting they’re the one who laid a juicy fart at the important meeting. “Likely” voters might be upside-down with the usual bunch staying home exasperated, while red-neck yokels proudly cast the first vote of their lives. Some will have a brainstorm when they see Gary Johnson’s name on the ballot. And have the burned Bernie-Brats come to their senses yet?

And what about October surprises? We talk about this possibility, but it usually doesn’t happen. However Russia and China both strongly want Trump to win — Trump in the White House would be like winning a major war for them or a windfall of a trillion dollars in their calculus. We already know that Russia is happy to put its computer crackers to work helping Trump. Expect more of the same and worse. Putin can easily force a big flare-up in Syria during October. China (perhaps with assistance from North Korea) will escalate a trade or territory dispute.

Our adversaries didn’t have a strong opinion on whether Obama, McCain or Romney should lead the U.S. Foreign military and trade policies would not vary much. But that’s not true of this cycle. China is delighted to know that Trump will surely scuttle TPP. Putin may be able to increase Assad’s power in Syria, gobble up the rest of Ukraine and look for more. Both countries will be delighted to see a rudderless U.S.A. led by an incompetent fool.

I certainly hope and pray Hillary can turn her 72% chance into victory. But it’s far far from a “done deal.”

60

I don’t think a 28% of a Trump win is reasonable but you’ll have to forgive me that I won’t prove it to you by flying to England and placing a $20 000 bet.