538 Model Megathread

[Beefalow and the Consumer]

You must be logged in to read this quote.

To be fair and balanced, link to NYT article which sources one garbage QU poll.

Good jorb.

[Beefalow and the Consumer]

It's a lot more valid than his previous prediction that Trump has almost no chance in the primary. That prediction turned out to be bad because all of the primary polls went against it. Meanwhile all of the polls are on Clinton's side in the general.

It's a pretty good prediction based on where the polls have been, plus the amount of time that is left.  If it were September and we had poll numbers like these, I'd put the odds of a Clinton victory at 95%.  October: 99%.  Eve of the election: 99.9%.

[Beefalow and the Consumer]

ARIZONA!!

[Beefalow and the Consumer]

Looking further down:

You must be logged in to read this quote.

Wait...

Whaaaaaa?

[Beefalow and the Consumer]

Clinton's closest states (polls only):

Arizona
North Carolina
Colorado
Ohio
Iowa
Florida <-- Tipping point
Virginia
New Hampshire
Nevada
Pennsylvania

Clinton's closest states (polls-plus):

Ohio
Colorado
Florida <-- Tipping Point
Iowa
New Hampshire
Virginia
Pennsylvania
Nevada

Ohio I expect to see past the tipping point.  Colorado I don't.  Iowa I especially don't.  And Pennsylvania being that far behind the tipping point is weird.  What's up with PA?

Oh, it's the Marist College and Evolving Strategies outlier polls.  WTF are they doing in there, completely jacking up his model?

[Beefalow and the Consumer]

Kansas and Mississippi are some of the most competitive states? I like Nate Silver's forecasts, but just like his primary prognostications, I think these will end up totally wrong (Gary Johnson probably won't get anywhere near the 9% this forecast shows).

To be fair this will adjust with new polling (both statewide and nationally). Obviously those two states should move towards Trump.

I believe his models don't just take polls in isolation.  They take polls in the context of polls in other states and form a holistic model.

For instance, say you have several polls in Mississippi, when averaged, saying Trump is only ahead by five points.  Is that statistical noise, or is it meaningful?  Well, take a look at surrounding states, or states with similar demographics.  Do their polls corroborate this?  If so, weigh that data heavily.  If not, weigh it less.  Similar states move in concert.

If his model says Mississippi is competitive, Mississippi is competitive.  I have good reasons of my own to believe it should be, so this does not surprise me.  Kansas genuinely does.  There's a massive revolt against the Brownback regime going on, but that shouldn't blow up-ticket.

Silver's data analysis is very, very good.  His rating of polls is very, very good.  His political punditry is crap.

[Beefalow and the Consumer]

It looks like they fixed the Trump-Obama state and Clinton-Romney state numbers. Now its:

Clinton wins at least one state Mitt Romney won in 2012 87.4%
Trump wins at least one state President Obama won in 2012 63.3%

Seems about right.

It's hard to model correlations between regions.  OH and CO are mostly independent variables, but CO and NV are closely linked.  Assuming OH and CO are completely independent (they aren't), Clinton has about a 50% chance of winning them both.  If she does win them both, there's a good chance she'll retain the rest of Obama's 2012 states, but a small chance Trump could pick off a FL or VA.

[Beefalow and the Consumer]

Most likely states to flip (polls only):

NC   59.00%
AZ   53.90%
MO   47.30%
GA   41.90%
SC   38.50%
KS   32.60%
CO   31.50%
MS   31.10%
SD   30.00%
IN   29.60%

Most likely states to flip (polls-plus):

NC   46.00%
OH   40.10%
CO   38.80%
AZ   37.70%
IA   37.10%
FL   36.70%
NH   32.90%
MO   31.80%
VA   27.70%
NV   27.00%

[Beefalow and the Consumer]

Clinton's closest states (polls only):

Arizona
North Carolina
Colorado
Ohio
Iowa
Florida <-- Tipping point
Virginia
New Hampshire
Nevada
Pennsylvania

Wouldn't that order make Virginia the tipping point?

Heh.  You're right, my bad.  How'd I bork that one?

Clinton 270.

[Beefalow and the Consumer]

If you sort the states by Trump's %chance to win, then NH is actually the tipping point at 32.8% vs 27.6% for VA. 

So Trump's narrow path ( sorted by % chance to win) is Romney + OH + CO + IA + FL + NH

His percentages seem to update in real-time.  I think there's a constant punch tape loader feeding into his comp-u-tron.

[Beefalow and the Consumer]

Are we really going to pretend that Johnson will get more than 2-3% of the vote at best for this entire cycle?

Given two major-party candidates with the highest unfavorables ever?  Given that an activist who never held public office running on a leftist fringe ticket won nearly 3% of the vote in 2000, and Johnson was a two-term governor running as a third-way (ish) candidate?

[Beefalow and the Consumer]

So the 2012 election boiled down to four states -- Colorado, Florida, Ohio, and Virginia, any of which would decide the election. Those states are different enough and separated enough that there is no way to make an appeal that could resonate in all four of those states without
shifting America on the whole.

You must be logged in to read this quote.

There has been movement in party affiliation due to demographic shifts.  Not seismic, but incremental.  Virginia looks a little bit more like Maryland.  North Carolina looks a little bit more like Virginia.  Georgia looks a little bit more cosmopolitan and less Southern every year, though this change is not enough to put it in play just yet.  The industrial Midwest is getting whiter, and more Republican, while the Sunbelt is getting more diverse, and more Democratic.

Combine this with the type of candidate Donald Trump is, and how he built his appeal, and the model may need tweaking.  Florida and Virginia may be too diverse to be within Trump's reach, and Ohio may be too white and industrial to be within Clinton's reach - in a close election.  I also believe Nate Silver's model for PA is built too heavily on outlier polls (+11 and +13 polls when there is a preponderance of tied or +1-4 polls). 

I think if there are four states that decide the election, they are now Colorado, Iowa, North Carolina, and Pennsylvania.  It still is, of course, Clinton's to lose, and Trump's to lose spectacularly, as you say, but those four states are now the critical path.

[Beefalow and the Consumer]

We're in the midst of a week-long commercial for Trump, with fake smiles and a fake conservative consensus (except Ted Cruz, who for once won my admiration).

The RNC will end, conservatives will go back to expressing their reservations about Trump, while the DNC's Hillary lovefest will make the RNC look like the awkward forced marriage that it is.

Hillary will be up 10 points in mid-August before things stabilize and we settle back to the 5-point equilibrium that will probably be the ultimate outcome.  And as time goes on and there's less chance for movement, Hillary's chances on 538 will continue to rise.

This is looking like a Clinton victory, 50-44-6.

[Beefalow and the Consumer]

Of course, I would think that the real fundamental of this race is that Donald Trump is the Republican nominee, and that the natural "mean" in such a race is a 10-point loss for Trump.  Of course, such thinking wasn't really helpful in the primary.

He's been legitimized as a candidate and a human being, as much as I vociferously disagree with both of those notions.  The sight of the rest of the GOP enabling this massive, crass ego trip is alarming and disturbing.  In years past this convention would be seen for the perversion that it is.  But at this point the new normal has been established.  A man like Donald Trump as the face of a major political party and possible commander-in-chief is now totally ok with at least 44% of voters.  My feelings, and the feelings of much of the pundit-sphere, are irrelevant.

[Beefalow and the Consumer]

Turnout game matters a lot less in the general than for caucuses, for obvious reasons, though it could matter on the margins.

It's not going to swing the vote nationally by any significant margin.  However, if there's a close fight for a pivotal state, ground game can make all the difference.  Particularly in a certain 20-EV eastern state where the Democratic vote is geographically concentrated and easily mobilized.

[Beefalow and the Consumer]

According to today's now-cast, if Trump picks up EVERY state he has a 25% or greater chance of winning, the map looks like this:

Trump 270
Clinton 168

If Clinton picks up every state she has a 25% or greater chance in, the map looks like this:

* Missouri is at 24.9% for Clinton

Clinton 384
Trump 154

[Beefalow and the Consumer]

1984 also.

Yeah, Roosevelt did it 3 times actually, 36, 40, and 44

Wow, I'm amazed he won New Hampshire three times.

[Beefalow and the Consumer]

Another interesting game to pay is to match red states with blue states by probability of flipping.

In polls-plus:

DC, MD, HI = OK, AL
CA = WV, ID
NY, MA = NE-3, WY
IL = NE-AL, KS, UT, AR, LA, TN
VT, RI = KY
WA = TX, IN, MS
NJ, DE, CT, ME-1, OR = MT, ND, SD, AK
NM = SC
MI = NE-1, MO
VA, MN, ME-AL = GA
PA, CO = AZ
NV = NE-2
IA = NC

Trump is closer to these than Clinton is to any Trump states:
FL, OH, NH, ME-2

[Beefalow and the Consumer]

I don't think we see South Carolina flip. That seems like a bit of a stretch.

The 538 now-cast gives Clinton a better chance of taking SC than Trump has of taking OH (?!).

0-1%
TRUMP: DC, HI, CA, MD, NY, MA, IL
CLINTON: WV, AL, OK, ID

1-5%
TRUMP: VT, RI, NJ, WA
CLINTON: WY, NE-3, TN, LA

5-10%
TRUMP: OR, CO, NM, DE, CT, WI, MI, ME-1
CLINTON: NE, KS, AR

10-25%
TRUMP: MN, PA, VA, ME
CLINTON: IN, TX, MT, ND, SD, NE-1, UT, KY, MS, MO, AK

25-40%
TRUMP: NV, IA, OH, FL, NC
CLINTON: SC

40-50%
TRUMP: ME-2, NH
CLINTON: NE-2, AZ, GA

[Beefalow and the Consumer]

A bit off topic, but MSNBC just plotted out Hillary's ceiling at this time: Here was the map:

Pretty much lines up with all of the states where Hillary has >25% chance of winning today in 538 Nowcast.

Today's >25% nowcast for Clinton:

*SD is at 24.9%

At >20% SD, ND, MT, TX, and MS flip.
At >15% IN flips.

>25% nowcast for Trump:

At >20% NV flips.
At >15% IA, OH, FL, and NH flip.  This will get Trump to exactly 270.

In other words, if Trump's ceiling is 270 EVs, Clinton's ceiling is 468:

[Beefalow and the Consumer]

Happy South Dakota Farmers strike again.

If the price of oil skyrockets above $150/bbl, Clinton may win North Dakota but lose the election.

[Beefalow and the Consumer]

Here's the "pecking order" of Romney states:

Polls-Plus

1. North Carolina
------------------- 50% line
2. Nebraska CD2
3. Georgia
4. Arizona
------------------- 25% line
5. Missouri
6. South Carolina
7. Nebraksa CD1
8. Montana
9. South Dakota
10. North Dakota
11. Alaska
12. Mississippi
13. Texas
14. Indiana
15. Utah

Trump's 25% line is between Ohio and Pennsylvania

Polls only

1. North Carolina
2. Georgia
3. Arizona
4. Nebraska CD2
--------------------- 50% line
5. South Carolina
6. Missouri
7. Texas
8. Mississippi
9. Nebraska CD1
10. South Dakota
11. Montana
12. Alaska
--------------------- 25% line
13. North Dakota
14. Indiana
15. Utah

Trump's 25% line is between North Carolina and Ohio

Now cast

1. North Carolina
2. Georgia
3. Nebraksa CD2
4. Arizona
5. South Carolina
--------------------- 50%
6. Missouri
7. Texas
8. Alaska
9. Nebraska CD1 (29.1)
10. South Dakota
11. Mississippi
12. Montana
13. North Dakota
---------------------- 25%
14. Indiana
15. Kansas
16. Utah

Trump's 25% line is between Georgia and North Carolina

In the now-cast, Clinton is more likely to win Wyoming than Trump is to win the Electoral College.  He has a 10% chance in OH and NH, which is worse than Clinton's 11% chance in AR and NE at large.

[Beefalow and the Consumer]

Since polls during and right after the convention are weighed less in polls-only and polls-plus, we're going to see those maps turn a lot bluer if things keep going this way.

[Beefalow and the Consumer]

Any reason for the sudden across-the-board downtick for Clinton?  Were the recent batch of polls that bad?

[Beefalow and the Consumer]

Intresting, that all the models started converging (don't really understand why).

Probability that Trump wins,
30/08, now:

Polls-plus 26.9%
Polls-only 22.7%
Now-cast 24%

Polls-plus model was pretty stable. Trump gained just 3-4 pps over the last 2 weeks.
In Polls-only and Now-cast Trump gained 10 and 14 pps, respectively, over the last 2 weeks.

We should expect all three models converging as Election Day approaches.  Polls-only factors in how much polls can drift between now and 11/8, and polls-plus figures in factors like convention bounces (which we're now past) and possible changes in economic conditions (the time for which is running out).