538 Democratic primary model is up

[Slander and/or Libel]

The most irritating thing about them having this primary model is that now their articles and tweets all filter other results through it. They're like, "Sanders had a good poll in Iowa, and you know it's good because his probability of receiving a majority of pledged delegates in our model ticked up by 2%." It's like saying, "You know this is a good result because I fed it into my guess machine, and my guess machine gave me a thumbs up."

[Vaccinated Russian Bear]

A model is though the opposite to guess machine and/or cherry-picking.

[bilaps]

LOL, they had an error in a code which overplayed home state, region bounces instead of downplaying them and now Biden has surged from 41 to 45 when they corrected it.

[Fmr. Gov. NickG]

So I know the 538 model include a probability that each candidate will drop out after each set of contests, presumably based mostly on how well they performed. 

I wonder how they are modeling Bloomberg’s probability of dropping out prior to Super Tuesday.  Presumably he will perform so poorly in the first four contest that any model based on past candidate behavior would predict he is almost certain to drop out before Super Tuesday. 

But we all know he doesn’t care about these contests and is virtually certain to stay in through Super Tuesday.  How does the model incorporate this?  Is there just some ad hoc fix?

[Mr. Morden]

So I know the 538 model include a probability that each candidate will drop out after each set of contests, presumably based mostly on how well they performed. 

I wonder how they are modeling Bloomberg’s probability of dropping out prior to Super Tuesday.  Presumably he will perform so poorly in the first four contest that any model based on past candidate behavior would predict he is almost certain to drop out before Super Tuesday. 

But we all know he doesn’t care about these contests and is virtually certain to stay in through Super Tuesday.  How does the model incorporate this?  Is there just some ad hoc fix?

I'm sure the model doesn't do anything to account for that.  Just like I'm sure it wouldn't have accounted for the fact that, for example, McCain skipped Iowa in 2008, and wasn't going to drop out of the race after Iowa no matter how badly he did there.  But yes, that's a good argument for why it's presumably underestimating Bloomberg at least somewhat.  If you removed contingencies for candidates dropping out, then I'm sure Bloomberg wouldn't be as far behind Buttigieg in the model as he is now.

[Vaccinated Russian Bear]

The most irritating thing about them having this primary model is that now their articles and tweets all filter other results through it. They're like, "Sanders had a good poll in Iowa, and you know it's good because his probability of receiving a majority of pledged delegates in our model ticked up by 2%." It's like saying, "You know this is a good result because I fed it into my guess machine, and my guess machine gave me a thumbs up."

This has always been Silver's trick. There's a limit to how much someone can say about it, but it is helpful to keep commentary based in reality. Unfortunately the writers employed by 538 lack the insight to add much value.

This "trick" was how 538 was the only forecast that gave reasonable chance to Trump in 2016.

[Vaccinated Russian Bear]

So I know the 538 model include a probability that each candidate will drop out after each set of contests, presumably based mostly on how well they performed. 

I wonder how they are modeling Bloomberg’s probability of dropping out prior to Super Tuesday.  Presumably he will perform so poorly in the first four contest that any model based on past candidate behavior would predict he is almost certain to drop out before Super Tuesday. 

But we all know he doesn’t care about these contests and is virtually certain to stay in through Super Tuesday.  How does the model incorporate this?  Is there just some ad hoc fix?

I'm sure the model doesn't do anything to account for that.  Just like I'm sure it wouldn't have accounted for the fact that, for example, McCain skipped Iowa in 2008, and wasn't going to drop out of the race after Iowa no matter how badly he did there.  But yes, that's a good argument for why it's presumably underestimating Bloomberg at least somewhat.  If you removed contingencies for candidates dropping out, then I'm sure Bloomberg wouldn't be as far behind Buttigieg in the model as he is now.

Does it look so strange though? Pete's ONLY chance is to "win" IA and/or NH and surge. Otherwise his chance would be near 0 as well.

[Sorenroy]

https://www.youtube.com/watch?v=EeunWvAkjqM

538 mentioned on their most recent podcast that the debate is factored into the model as the equivalent of 6 days of campaigning. So, it is saying there are 10 days in campaign time between yesterday and the New Hampshire primary. Now that we're past the debate, the polls done will be much more aggressively factored into the model.

[Mr. Morden]

The latest iteration now has Bloomberg up to 3rd place (behind Sanders and Biden) in terms of average number of delegates won in all the runs of the model.  He still runs a little behind Warren and Buttigieg in probability of getting a delegate *majority*, presumably because the model gives an outside shot of some Buttigieg and Warren wins (or at least overperformances) in the remaining pre-Super Tuesday states giving them momentum going into Super Tuesday.

Of course, as mentioned earlier, the model is presumably underestimating Bloomberg at least a little, since it includes scenarios where Bloomberg drops out of the race before Super Tuesday.  This is obviously not going to happen, since he's not even bothering to compete in the pre-Super Tuesday states, so it's not like a poor showing in those states is going to lead to him quitting.

[Beefalow and the Consumer]

We now have a 27% chance of no candidate having a majority of pledged delegates.

If you want Trump defeated in November, pray this doesn't happen.

Pray HARD.

[Fmr. Gov. NickG]

Nate just updated the model, and somehow it has Biden's chances going UP in response to the New Hampshire result.

[Slander and/or Libel]

Mostly a big gain for "no majority," though, a scenario that the model doesn't really appear to have been built for...

Yeah, this model was already built on scant data even if this primary season looked like a lot of others, which it doesn't. I don't understand why 538 is so all in on using it as their interpretive filter for everything.

More, it's fluctuating so much, what's even the point? A forecast should respond to events, but shouldn't it also be somewhat stable? If it's not, then what's the point of it over simple polling averages?

[Penn_Quaker_Girl]

We now have a 27% chance of no candidate having a majority of pledged delegates.

If you want Trump defeated in November, pray this doesn't happen.

Pray HARD.

Just jumped up to 34% (1 in 3) about ten minutes ago. 

[GeorgiaModerate]

Mostly a big gain for "no majority," though, a scenario that the model doesn't really appear to have been built for...

Yeah, this model was already built on scant data even if this primary season looked like a lot of others, which it doesn't. I don't understand why 538 is so all in on using it as their interpretive filter for everything.

More, it's fluctuating so much, what's even the point? A forecast should respond to events, but shouldn't it also be somewhat stable? If it's not, then what's the point of it over simple polling averages?

At this stage of the election, it's not surprising that a model should fluctuate a lot in response to events.  We've had one caucus and one primary, both of which had some unexpected results.  As more primaries occur, the model should become more stable.

[Slander and/or Libel]

Mostly a big gain for "no majority," though, a scenario that the model doesn't really appear to have been built for...

Yeah, this model was already built on scant data even if this primary season looked like a lot of others, which it doesn't. I don't understand why 538 is so all in on using it as their interpretive filter for everything.

More, it's fluctuating so much, what's even the point? A forecast should respond to events, but shouldn't it also be somewhat stable? If it's not, then what's the point of it over simple polling averages?

At this stage of the election, it's not surprising that a model should fluctuate a lot in response to events.  We've had one caucus and one primary, both of which had some unexpected results.  As more primaries occur, the model should become more stable.

But again, what's the value added of this over simple poll averages and delegate math at that point?

[GeorgiaModerate]

The newest update shows a big drop for Bernie, perhaps in response to the YouGov poll?

Sanders 37
No one 36
Biden 17
Buttigieg 5
Bloomberg 4
Warren 3
All others 0.1

[Gass3268]

I have to imagine the model expected a stronger result for Sanders in NH.

[Vaccinated Russian Bear]

I have to imagine the model expected a stronger result for Sanders in NH.

This.

Sanders' vs field performance in NH and national polls was ~good to get plurality, but not so for majority. So Bernie is a front-runner, but not the one who easily wins against the field. Not for now that is. Virtually no polls from NV/SC and few national.

And the model shows that.


Majority:
Sanders 37% (-9 since yesterday I believe?)
No one 36% (+6 I believe?)

Plurality:
Sanders 51% (-3 I believe?).


It sounds ~reasonable IMO. Besides, the model is nog a GOD LMAO, but a tool to simplify a comprehension of where the race is at.

[Mr. Morden]

Does the model currently have Sanders and Biden (usually in that order) as the top two most likely to win in every state that isn't the home state of another candidate?  That seems to be the case in all the states I checked, but I didn't check them all.

[GeorgiaModerate]

Does the model currently have Sanders and Biden (usually in that order) as the top two most likely to win in every state that isn't the home state of another candidate?  That seems to be the case in all the states I checked, but I didn't check them all.

Yes, it's currently Sanders/Biden in every state except for the following:

AL: Biden/Sanders
DE: Biden/Sanders
IN: Sanders/Buttigieg
MA: Sanders/Warren
MN: Sanders/Klobuchar

[atheist4thecause]

https://projects.fivethirtyeight.com/2020-primary-forecast/

Current chances of winning >50% of pledged delegates:

Biden 42%
Sanders 22%
No one (contested convention) 13%
Warren 12%
Buttigieg 9%
All others 2%

It also includes individual state forecasts.

A reminder of how drastically 538 changes it's "forecasts". These are not to be trusted. They are now at:
Sanders 37% (+15%)
No one (contested convention) 35% (+22%)
Biden 17% (-25%)
Buttigieg 4% (-5%)
Bloomberg 4% (N/A - All others was at 2%)
Warren 3% (-9%)
All others .2% (-1.8%)

[Mr. Morden]

Bloomberg's now up to 3rd place (after Sanders and Biden) in most likely to win a delegate majority.  And he's up to 2nd place (now ahead of Biden) in average number of delegates won across all runs of the model.

[GeorgiaModerate]

Sanders and "No majority" are now tied at 37%.

[GeorgiaModerate]

No majority has now taken the lead 38-36, undoubtedly due to the Georgia poll that was pretty good for Biden.

[100% pro-life no matter what]

No Majority: 38%
Sanders: 35%

https://projects.fivethirtyeight.com/2020-primary-forecast/?ex_cid=rrpromo

This might be the year we see a contested convention.