538 Model Megathread

[GeorgiaModerate]

Clinton's now at 63.3% in polls-only, 67.7 in polls-plus, and 82.2 in nowcast.

[GeorgiaModerate]

Nate Silver tweeted a nice summary of the difference between 538's models and some of the other well-known ones.  https://twitter.com/NateSilver538/status/770613589992280065

[GeorgiaModerate]

The race certainly appears to have tightened a bit, but it's closer to a ripple than a surge.  Trump would need to actually lead in a good number of national and swing state polls to move above 50% in the 538 models.

[GeorgiaModerate]

he's not wrong. Reagan was hapless and completely unqualified for the job.

But the Reagan example proves why we shouldn't elect Trump - Reagan was a terrible President.

Reagan is the best president post-Eisenhower. Enough said.

Still, he wasn't qualified for the job. Neither was Obama. Didn't stop either.

Both Reagan and Obama had some experience in government; Trump has zero.  More importantly, both had the temperament to be President.  Trump, by all indications, does not.

[GeorgiaModerate]

An intresting article about a different models. I didn't know that states are more correlated in 538's model than in others. It explains a lot

http://www.nytimes.com/2016/09/15/upshot/as-clinton-trump-race-tightens-heres-how-forecast-models-differ.html

Great article.  Thanks for posting.

[GeorgiaModerate]

I'm not sure what polls are responsible, but 538's model has Trump still surging.

This has been puzzling me too, since the recent Trump surge seems to have peaked, but his chances keep getting better on 538.  Meanwhile, the Upshot model (which is generally slower to move) has tracked the poll movement pretty closely.  During the surge, Clinton's win probability in Upshot dropped from near 80% to 73%, but it's recovered in the past couple of days to 75%.

I'm starting to wonder if Nate Silver has tried to do too much in his models, and they're becoming less useful as a result.

[GeorgiaModerate]

Another huge difference is that Silver's model assumes that the states move together and do not act totally independent of each other. If Trump gains in North Carolina, the model assumes that he also makes some gains in Virginia, unless/until there's polling evidence to contradict that. Not the case with other models.

Yes, which is something that always bothered me. Like... does a bad IA poll impact MO, IL, MN?

Really? That's one of the model's main benefits to me. Assuming the states move independently makes no sense.

I agree to a point, although assigning the proper amount of correlation between the states will be tricky; for example, MS and AL are going to be more alike in their movements than MD and WV, to pick another pair of neighbors.  But this also has the potential to overstate the effect of outliers or other noise.

[GeorgiaModerate]

Clinton is back above 60% in all three models this morning.

[GeorgiaModerate]

538's polls-only forecast gives Trump a higher chance of winning ME-02 than NE-02. How accurate is this?

Recent polling in ME-02 shows Trump definitely ahead there.  There isn't any recent data from NE-02, so this is more uncertain.

[GeorgiaModerate]

One thing to remember about the 538 models is that they have a "wider" graph of probabilities than some other models (i.e., although they show a greater chance of Trump winning than some, they also have a greater chance of a big Clinton win.)  Nate tweeted about this (https://twitter.com/NateSilver538/status/780050423470100480):

You must be logged in to read this quote.

[GeorgiaModerate]

Nowcast has now flipped back to Clinton, 52.4-47.6.  She's at 55.0 in polls-only and 54.2 in polls-plus.

[GeorgiaModerate]

North Carolina is now blue in the polls-only model. I remember some states swinging around a bit in 2012, but the amount even his polls-plus model has swung around is pretty insane.

There's been a lot more uncertainty in this election than there was in 2012, when the trend was pretty obvious (except to poll unskewers) for most of the election, except right after Romney's victory in the first debate.  The 538 models are bullish on uncertainty (to quote Nate) so they do react -- perhaps overreact -- to fluctuations in polling. 

[GeorgiaModerate]

Is North Carolina really going to vote to the left of Ohio and Iowa?

NC to the left of Iowa looks very likely.  I'm skeptical about it being left of Ohio.

[GeorgiaModerate]

Does Trump have a path without PA, VA and NC?

It would be very tough (if he's already lost PA/VA/NC, winning all the other swing states would be like drawing to an inside straight flush) but this would get him to 270 exactly:

[GeorgiaModerate]

Clinton is now above 70% in all three models (over 80% in nowcast).  It's the first time she's been above 70% in polls-plus since September 1.

[GeorgiaModerate]

Iowa has flipped to blue on polls-only. 

[GeorgiaModerate]

Alaska is at 30% on Polls Only - higher than GA

Probably because no one has polled GA since Hillary has had her comeback.

No, the states should move around based on national polls. Alaska is light pink due to its high uncertainty; the actual predicted margin is still more Republican than Georgia.

They move because of the national polls, but their partisan order (think of that snake they have) are based on state polls.

And there have been several recent AK polls that showed it close. 

[GeorgiaModerate]

There is a way to check to see if he's right or not; if events that he says has x% chance of happening actually do end up happening x% of the time.

His predictions have ended up being quite accurate in that regard.

This is not measurable.  It's impossible to verify whether (for example) a 92% chance of Obama winning in 2012 was accurate.  (I think the final number was close to that.)  We have a sample of ONE.  To get close to checking whether 92% was accurate, you'd need to rerun the 2012 election dozens of times, which is obviously impossible.

[GeorgiaModerate]

There is a way to check to see if he's right or not; if events that he says has x% chance of happening actually do end up happening x% of the time.

His predictions have ended up being quite accurate in that regard.

This is not measurable.  It's impossible to verify whether (for example) a 92% chance of Obama winning in 2012 was accurate.  (I think the final number was close to that.)  We have a sample of ONE.  To get close to checking whether 92% was accurate, you'd need to rerun the 2012 election dozens of times, which is obviously impossible.

"Checking" whether the probability of a single event was "correct" or not is nonsensical.  The point is that if you run the same model on many different elections, then you should be able to check if the model is any good by seeing if 92% favorites do indeed win 92% of the time.  The problem is, there are only so many presidential elections to look at since the advent of polling, so you're checking with small number statistics.  You can overcome that problem by looking at each individual state separately, but the states are correlated with each other, so it's a bit messy.

I assume you can also do it for Senate races, which are presumably going to be less correlated.

All right, that's a fair point as far as it goes; but as you noted, the number of Presidential elections since such models existed is far too small a sample to be useful.

[GeorgiaModerate]

All right, that's a fair point as far as it goes; but as you noted, the number of Presidential elections since such models existed is far too small a sample to be useful.

The election doesn't have to have happened "since such models existed".  You can go back to elections from long before your model existed, and apply the model and see how it performs, as long as you have the polling data from those old elections.  But you're still stuck using elections since *polling* existed.  And more than that, elections for which large numbers of state polls existed.  And I guess that takes you back a few decades, but not more than that.  So yes, a limited sample.

To be clear, going back to past elections to see how the polls predict the final outcome is how the models were constructed in the first place.  However, since you're again stuck with a limited sample, you have the potential for over-fitting.

*concedes gracefully*

[GeorgiaModerate]

Decided to make a map of all the states that have never dipped below an 80% chance of the leading candidate winning for more than a week in the polls-plus model:

Clinton — 176
Trump — 146
tossup* — 216

*Edit: I do not think the grey are tossup (that's just what it says in the calculator), I think they are the not-titanium-[party here] states.

Edit #2: I used the polls-plus model for this map.

What's interesting about this map is that Trump has zero chance of winning any of the red states, but some of the blue states are close enough that Clinton has a non-trivial chance in them: SC, TX, UT, IN, MO.

[GeorgiaModerate]

The polls-only forecast gives McMullin a 3.3% chance of winning Utah.

It's reasonable based on polls, but I think that underestimates him. 3-way races are a lot more volatile.

This, plus I don't think it accounts for his recent momentum in Utah.

[GeorgiaModerate]

One thing that always stood out to me as weird in the 538 is their scenarios stuff at the end. Currently in the Nowcast the models says:

Clinton wins at least one state Mitt Romney won in 2012           86.4%
Trump wins at least one state President Obama won in 2012   54.1%

This seems really weird to me. It implies that they assign at least a 40% probability of Trump winning an Obama state while Clinton wins a Romney state. Generally these numbers have always struck me as too high throughout the campaign.

Well, they're not independent events, so the odds of both happening are not quite that straightforward.  But it does seem like something that could easily happen, e.g. Trump winning Iowa while Clinton wins NC.

[GeorgiaModerate]

Most of the adjustments do come from trend lines. That isn't unreasonable in itself but I think the problem is that these trend lines are largely derived from all the dumb trackers. And that's where they gain undue influence. The last few days the model has been insane, IMO. Most of the win probability adjustments have been going the wrong way.

Lol, no. If there were no Trump's surge (temporary/artificial or not), we would not see polls from a reputable pollsters showing tie races in CO/NH/NV/FL.

Just for 2 weeks ago, Trump was trailing in most state polls by 3-7% points.

There has clearly been a Trump surge over the past two weeks.  It's starting to look like it reached a peak and has perhaps ebbed a bit.  From Nate Silver:

You must be logged in to read this quote.

[GeorgiaModerate]

538 addresses the elephant in the casino.

This makes me picture an elephant pulling a slot machine arm with its trunk!