Trump could win the election in a nowcast by FiveThirtyEight

[RickJay]

Well, sometimes it is and sometimes it isn’t. In some states you can draw some solid observations from it and in some you really can’t beyond “X people have voted already.”

The results in Nevada are pretty well added up and contain a lot of info, and the results look very blue. So that’s pretty good information. The results in North Carolina are hard to gauge; far more Democrats voted early than Republicans. But that was also true in 2012, and Romney squeaked out a win there; the evidence would suggest Democrats are simply likelier to vote early, possibly because North Carolina has taken steps to ensure it’s harder for black people, especially, to vote. So Election Day will see more Republicans show up. I would ascribe almost no value to early voting info in NC, even though it’s more than half the votes that will be cast.

Here are the current figures:

http://dl.ncsbe.gov/ENRS/absentee11xx08xx2016_Stats.pdf

Unless NC turnout is 100%, which it will not be, probably two thirds of all votes have been cast. Registered Democrats hold a huge lead to date, but as you can see, there are many registered Republicans waiting to vote, and the number of independents is huge.

[Riemann]

@CarnalK

Early voting isn’t very predictive, maybe that’s why he doesn’t incorporate it.

Ok, so my assumption that it contributes useful information is just wrong?

[Tzigone]

In regards to poll aggregating sites - I saw someone (I don’t recall who, if it was a news article or message board poster, etc.) express concern that with higher accuracy from poll aggregating sites, fewer/smaller polls were being conducted. That might, in turn, effect the quality of poll-aggregating sites. Is there is any basis in that - has the number of polls or size of polls (or number of state-specific polls) gone done since 2004? Anyone know?

[RickJay]

To add to my comment, the NY Times has looked at early voting in North Carolina and concluded Clinton will win the state easily. Six points. I believe they are insane.

[Honey]

@RickJay

To add to my comment, the NY Times has looked at early voting in North Carolina and concluded Clinton will win the state easily. Six points. I believe they are insane.

I wish that were true, but 6 points? What are they basing that on?

[CarnalK]

@Riemann

Ok, so my assumption that it contributes useful information is just wrong?

Well it can and a bunch of interesting stuff is happening this year but historically it isn’t a very good predictor so it’s hardly surprising he doesn’t want to be making a bunch of from the hip adjustments to his model to incorporate it. I posted these articles in the “stretch run” thread:

@CarnalK

I haven’t seen anything that says early voting is predictive.
538: Don’t read too much into early voting

RCP: Early Voting a Poor Predictor of Final Results

[RTFirefly]

@Riemann

1/1.21 = 82.6% if you want to lay her.

Well, she’s only six years older than me, and looking good for her age.

[CarnalK]

And speaking of North Carolina, from that 538 article:

Indeed, if you were relying on early votes to forecast what was going to happen on Election Day in 2012, you’d get wildly misleading results. Democrats maintained substantial leads among early voters in North Carolina, Louisiana and West Virginia, and were trailing by a relatively narrow margin in Oklahoma, but still lost those states when all the votes were counted. Republicans won early voters in Pennsylvania and Colorado but lost the final tallies there.

[RTFirefly]

@RickJay

AS of 35 minutes prior to this it’s 65 and change.

Anyway, Silver’s model is wrong now, of that I’m certain.

Well, it’s not wrong. It works for what it does, it’s just a model on a computer. But it doesn’t incorporate early voting information, which at this point is very clearly favourable for Clinton. The state we know most about is Nevada, which 538 has as a toss up but is at worst 95% likely to go Clinton; that does change the forecast if you figure that in.

Also, I’ve been clicking on the ‘updates’ button to see what’s new, and aside from low-rated or unrated pollsters, it looks pretty good. I’ll do state/US, pollster, Clinton lead, 538-adjusted Clinton lead:

NC, Siena, tie, -1.
FL, Q-poll, +1, +2.
NC, Q-poll, +2, +3.
US, CBS/NYT, +4, +4.
US, ABC/WaPo, +4, +3.
US, IBD/TIPP, -2, tie.
US, Selzer, +3, +4.

It’s hard for me to believe that if Clinton wins nationally by >3%, that she’s not going to bring enough states with her to get her over 270. The problem of her gaining votes in places like Texas, where she’ll just make it a closer loss, or California, where she’s winning going away anyway, is absolutely real. But Arizona is going to correlate somewhat with those. A Trump path to victory with that sort of popular-vote margin for Hillary involves winning a ton of really close states, from AZ and NV to FL and NC and NH.

It could happen, but it sure seems like a longshot.

[JohnT]

@CarnalK

Early voting isn’t very predictive, maybe that’s why he doesn’t incorporate it.

I don’t know about this. Listening to his podcast, Silver has bet a LOT on the idea of demographic-based modeling. It’s one of his arguments as to why “unskewing” polls for political party is a flawed strategy - party affiliation changes upon choice, whereas age, gender, race, are independent of decisions made by the voter.

So if you have states, such as NC and NV, releasing official early voting statistics which correspond to racial or gender demographics, it would seem worthwhile to explore how to incorporate this data into the model. In TX, we have counties in which more Hispanics early-voted than voted in 2012. Odd that it’s argued that this doesn’t matter.

[CarnalK]

Because you can’t be sure that high turnout of a certain demographic in early voting will mean higher turnout overall. Some of the hispanic turnout numbers are great and I would be surprised if that doesn’t mean good numbers at the end but I certainly understand him not wanting to start fiddling with his coding on some flashy early voting news.

[RickJay]

@RTFirefly

It’s hard for me to believe that if Clinton wins nationally by >3%, that she’s not going to bring enough states with her to get her over 270. The problem of her gaining votes in places like Texas, where she’ll just make it a closer loss, or California, where she’s winning going away anyway, is absolutely real. But Arizona is going to correlate somewhat with those.

Your last sentence contradicts what comes before it.

Obviously, it is possible for the popular vote and EV count to not match. But those scenarios all lie within an election where the popular vote is VERY close, as was the case in 2000, when the popular vote was half a percent apart.

If the popular vote is 3 percent apart the odds of a PV/EV conflict are miniscule. “Longshot” doesn’t do it justice; it’s a Lloyd Christmas scenario. California will not run up the vote alone, that just doesn’t happen. If Clinton wins the popular vote by three percent she wins the election 99.95% of the time. Even at 2 percent it’s probably 98-99 percent likely a clear victory.

[RickJay]

@CarnalK

Because you can’t be sure that high turnout of a certain demographic in early voting will mean higher turnout overall. Some of the hispanic turnout numbers are great and I would be surprised if that doesn’t mean good numbers at the end but I certainly understand him not wanting to start fiddling with his coding on some flashy early voting news.

Especially since Silver is very insistent on his model’s moving parts being based on prior evidence. He doesn’t just say “well, this looks interesting” and add it in; he tests it against what it mean in 2012, 2008, 2004, 2000…

Early voting is more popular than it used to be so comparing it against past elections is usually hard to do.

[RTFirefly]

@RickJay

Your last sentence contradicts what comes before it.

Obviously, it is possible for the popular vote and EV count to not match. But those scenarios all lie within an election where the popular vote is VERY close, as was the case in 2000, when the popular vote was half a percent apart.

If the popular vote is 3 percent apart the odds of a PV/EV conflict are miniscule. “Longshot” doesn’t do it justice; it’s a Lloyd Christmas scenario. California will not run up the vote alone, that just doesn’t happen. If Clinton wins the popular vote by three percent she wins the election 99.95% of the time. Even at 2 percent it’s probably 98-99 percent likely a clear victory.

I don’t see the contradiction. Not every state correlates well with every other state. Clinton’s outperforming the last few Dem candidates in much of the Sun Belt, but underperforming them in much of the Rust Belt. And obviously she’s benefiting from a Latino population that’s taking this election much more personally than the last few. (Can’t imagine why.) But black turnout for her may well fall short of what it was for Obama, especially in places like NC where it was at epochal levels in 2012.

So what’s happening in CA and TX isn’t necessarily going to mean much in NC, let alone Ohio and Iowa. But it’s hard to see trends in CA and TX not also happening in AZ and NV.

[JohnT]

@CarnalK

Because you can’t be sure that high turnout of a certain demographic in early voting will mean higher turnout overall. Some of the hispanic turnout numbers are great and I would be surprised if that doesn’t mean good numbers at the end but I certainly understand him not wanting to start fiddling with his coding on some flashy early voting news.

You’re exactly right about this - it would do no good to mess with the model now. But it needs to be reviewed how EV numbers/demographics can be incorporated into the model because this is a data point that is purposely ignored. Yes, there may be good reasons why it wasn’t done this year - but that doesn’t necessarily mean the issue should not be revisited.

[RTFirefly]

@CarnalK

I certainly understand him not wanting to start fiddling with his coding on some flashy early voting news.

This. At this point, you’ve got to go with the model you’ve been using all along. If there are things it should have picked up on but didn’t, you fix it afterwards, and hopefully it does a better job next time. But maybe it’s seeing reality more or less correctly. We’re about 36 hours from having answers.

I think I’ve already said that FL, NC, and NV will be the proof of the pudding either for Nate Silver or the more optimistic group of aggregators/modelers, most notably Sam Wang, since Silver has Hillary at under 50% in all of them, while the others see all three of these states as strongly favoring Hillary.

[Dinsdale]

Fucking 538 has HC steadily gaining throughout today.

Sure hope he is correct.

[GIGObuster]

Yeah, and right now both North Carolina and Florida are getting a shade of blue too.

[Dinsdale]

Of course, Dem Senate odds are dropping as Hillary’s rise…

[CarnalK]

@Dinsdale

Of course, Dem Senate odds are dropping as Hillary’s rise…

On the Upshot, Dem Senate odds are going up and are almost the reverse of 538’s numbers.

2016 Senate Election Forecast (Published 2016)