538 Model Megathread

[Erich Maria Remarque]

For the first time in forever, Trump has passed 40% in 538's polls-plus model.

Projected as the favorite in both Ohio and Florida. That doesn't make any sense.

I think this model is too sensitive to trendlines for specific pollsters, namely Quinnipiac.

I also don't get why they don't seem to have taken into account the convention bounce they projected. If Trump wants to be on track to get a 50-50 finish, he should probably be up a point or two in the swing states right now at least.

But there were almost no polls after convention!

[Erich Maria Remarque]

For the first time in forever, Trump has passed 40% in 538's polls-plus model.

Projected as the favorite in both Ohio and Florida. That doesn't make any sense.

The model doesn't make any sense this far out.

Trump now projected as the favorite in Nevada!!! They're adjusting all polling in the wrong direction!

Nate is fascist!

[Erich Maria Remarque]

I am surprised the Nov Polls-only forecast has Trump's chances at 47%.   I would imagine the model takes into account the temporary GOP post-convention bounce.   

It did

It doesn't. Only polls-plus does that.

It does, but much less aggressively than the poll-plus model. The poll-only model gives latest polls less weight even though they... yeah, recent.
http://projects.fivethirtyeight.com/2016-election-forecast/national-polls/

[Erich Maria Remarque]

For the first time Trump leads in polls-only forecast 50.1% vs 49.9%. Of course, it's due RNC-bounce, since we don't have any post-DNC polls yet. But still...

[Erich Maria Remarque]

Polls-Plus showed Trump not getting any convention bounce for Election Day 2016. His odds still stayed at 40% of winning the election. Maybe I should go back to polls-plus instead of now-cast.

The polls-plus model assumes, that there is 4% convention bounce for each candidate, and that is about what Trump got on average.

[Erich Maria Remarque]

I'm still trying to wrap my head around why they have Minnesota so ridiculously close. I mean, I get it, Mason-Dixon isn't the best pollster, but they're only counting two polls from Mason-Dixon, and then massively unskewing them. Do they really think Minnesota will be a two-point race, even if Trump narrowly wins?

But they write why, don't they? It mostly depends on trend line and house effects.


1.   Polling average   46.7%-   35.8%-   — -
Adjust for likely voters   +0.6--   +0.7--   —--
Adjust for omitted third parties   -3.6--   -3.6--   —--
Adjust for trend line   -3.3--   +3.2--   —--
Adjust for house effects   -1.4--   +0.8--   —--
2.   Adjusted polling average   39.1%-   36.9%-   — -
Allocate undecided and third-party voters   +7.2--   +7.2--   —--
3.   Polls-based vote share   46.3%-   44.1%-   — -
Calculate demographic regression   45.7%-   44.7%-   8.4%-
4.   Projected vote share for Nov. 8
Weighted average 52% polls-based, 48% demographics
46.0%-   44.3%-   8.4%-

[Erich Maria Remarque]

The current Now-Cast map:

Clinton 342
Trump 196

RIP TN Volunteer

[Erich Maria Remarque]

They're putting some stock in Reuters. That somehow flipped NC back on the polls-plus map.

Reuters/Ipsos used to be a live phone pollster, and they were generally good, but then they switched to being an internet only firm, and now they're all over the place. If Nate and Co. are smart they'll change their weight in the model.

Are you sure? I thougt they've always been using internet polling only.

[Erich Maria Remarque]

Reuters/Ipsos used to be a live phone pollster, and they were generally good, but then they switched to being an internet only firm, and now they're all over the place. If Nate and Co. are smart they'll change their weight in the model.

Are you sure? I thougt they've always been using internet polling only.

This poll from 2012 says "live telephone interviewing". They had Obama +3 which was pretty accurate.

If 538's rating is based on their "live telephone interviewing", it is really odd

Someone can probably ask them.

[Erich Maria Remarque]

Yeah, because just three polls from one B pollster that didn't poll since April is not enough as evidence for polls-only/plus models.

They might indicate something. Or not.

But everybody's waiting for Labour Day's polls anyway

[Erich Maria Remarque]

After Arizona Clinton +1 poll, it's now showing 20.5%
Seems like their system updates pretty slow.

[Erich Maria Remarque]

All right! I figured it out. Their latest polls:

AUG. 1   OH Predictive Insights / MBQF   Clinton +3
JUN. 20   OH Predictive Insights / MBQF   Clinton +4

Since 538's model is based both on level and trend, this poll showed a positive trend for Trump and drove his chances little bit up.

[Erich Maria Remarque]

New A+ Monmouth poll is showing Clinton +7 from (+13) for 4 weeks ago.

I guess, Trump will get 1-3 pps now. What do you think?

[Erich Maria Remarque]

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.

[Erich Maria Remarque]

Iowa seems strange to me. 50/50 in nowcast, but in Polls Only Hillary is at 60 and in Polls Plus Hillary is at 53. Seems very strange to say if the election were held today it is a total tossup, yet, somehow JUST USING POLLS, Hillary has a 60% shot on election day??? thoughts?

Polls Only model is not aggresive. On the election you in generelly don't give any weight at all for polls older than say 5-6 days. Simplied.

[Erich Maria Remarque]

Iowa seems strange to me. 50/50 in nowcast, but in Polls Only Hillary is at 60 and in Polls Plus Hillary is at 53. Seems very strange to say if the election were held today it is a total tossup, yet, somehow JUST USING POLLS, Hillary has a 60% shot on election day??? thoughts?

Polls Only model is not aggressive. On the election you in generally don't give any weight at all for polls older than say 5-6 days. Simplied.

I'm not following. Polls Only model is based on ONLY POLLS. Are you saying Polls Only uses different polls than Nowcast? I would think if polls show if election were TODAY the results are 50/50, but if we wait 2 months, Hillary is more likely to win, something doesn't really make sense. I could deal with 53% or something like that, but 60% is pretty significantly different, no?

Just pay attention on how the weights change when you switch between different models.

[Erich Maria Remarque]

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).

I don't always understand, the reaction on particular poll.
After recent Yougov's poll that showed compared to week before

You must be logged in to read this quote.

the changes were:
Polls-plus 27.0% (+0.1)
Polls-only 22.6% (-0.1)
Now-cast 22.8% (-1.2)

Why did Trump's chances increase in Poll-Plus? Insignificantly by just 0.1, but still

[Erich Maria Remarque]

And after recent Trump +3 (+0.6) LA Times poll:
Polls-plus: 27.2% (+0.2)
Polls-only: 22.3% (-0.3)
Now-cast:  23.8% (+1.0)


Do they update their fundamentals for Polls-plus model as well, each time they add new polls? Or is it the results of noise causing by too few simulation?

[Erich Maria Remarque]

Some thoughts about their model.


To those who still thinks LA Times is of no use.

Look at 538's model. Last two polls were expected [more or less] by their model: Monmouth didn't change it at all (just slightly), while Marquette was little bit better than expected. What did the magic? They DID LISTEN to those "junky" nationall polls. That's it.

I'm not kidding, just look at it http://projects.fivethirtyeight.com/2016-election-forecast/wisconsin/#now

The expected margin (for yesterday) in WI was D+3.8

What did we get today? D+5 and D+3

[Erich Maria Remarque]

Trump isn't going to take the lead unless he, you know, actually starts leading in the polls again.

[Erich Maria Remarque]

Some thoughts about their model.


To those who still thinks LA Times is of no use.

Look at 538's model. Last two polls were expected [more or less] by their model: Monmouth didn't change it at all (just slightly), while Marquette was little bit better than expected. What did the magic? They DID LISTEN to those "junky" nationall polls. That's it.

I'm not kidding, just look at it http://projects.fivethirtyeight.com/2016-election-forecast/wisconsin/#now

The expected margin (for yesterday) in WI was D+3.8

What did we get today? D+5 and D+3

Nate Silver today

You must be logged in to read this quote.

[Erich Maria Remarque]

Clinton's now below 70 in polls-plus.  The surge is real!

For the first time in a month.

Some of that was the jobs report

Probably 0.5%

[Erich Maria Remarque]

Clinton's now below 70 in polls-plus.  The surge is real!

Nate Silver doesn't have a clue. There is no path, let alone a single one that gives a 30% chance.

Come on now...

You need to get more realistic. A large portion of this country despises Trump. Minds are made up. GOP not winning PA, WI, (or NH! )

Wake up! According to the latest polls they are pretty much equally ~hated. Muahaha

[Erich Maria Remarque]

Clinton's now below 70 in polls-plus.  The surge is real!

Nate Silver doesn't have a clue. There is no path, let alone a single one that gives a 30% chance.

Those chances are there for "convenience" sake. The difference between 538 and others is that 538 assume that Undeciders are "pure", i.e. very elastic, or we can say that 538 doesn't make any assumption at all = we know nothing about undeciders and their preferences.

Other models assume that Undeciders will split evenly more or less.

That's it. Sort of.

But if you just look at what the models think each candidate score at each state and/or nationally, they are pretty much the same

[Erich Maria Remarque]

Converging!

After today CNN and SurveyMonkey polls.

Trump:
Polls-plus   32.5%
Polls-only   30.8%
Now-cast   32.3%