538 Model Megathread

[Wells]

To me the thing that makes the least amount of sense is there have been 3 Florida polls released since yesterday with Clinton up (+1,+5,+5), yet Trump's odds in Florida keep spiking?

If you look at his poll adjustments, one of them is Adjust for trend line, which helps Trump by 2% in Florida. There could be an error which causes a feedback loop, or it could be that the new trend hasn't fully manifested yet, or the model is getting overrun by junk polls.

[Likely Voter]

IIRC the NYC model doesn't factor in national polls and they aren't using these 50 state online polls.  538 model basically uses pretty much every poll there is (albeit with weighting) and those national polls are used to create momentum at the state level. So there is a level of garbage in/garbage out with 538, especially in periods where there are fewer live voter polls coming out, yet lots of data from these tracking polls and '50 state' polls.

[Ebsy]

Nate Silver is this cycle's premiere unskewer of polls so it is not surprise that his idiocy is evident to all.

[bilaps]

IIRC the NYC model doesn't factor in national polls and they aren't using these 50 state online polls.  538 model basically uses pretty much every poll there is (albeit with weighting) and those national polls are used to create momentum at the state level. So there is a level of garbage in/garbage out with 538, especially in periods where there are fewer live voter polls coming out, yet lots of data from these tracking polls and '50 state' polls.

Yes, his premise in a way is that states follow national numbers. So, nothing too surprising in fact that model gives far more meaning to ipsos national than fl monmouth 400 sample size.

[Likely Voter]

I think the issue is that Silver's model most wants lots of live caller state polls. If every day it had Fox, Monmouth, Marquette, Marist, CNN, etc state (and national) polls feeding it, it would work best. It is essentially an engine tuned for those. But this cycle the vast majority of data being put into the model are the online polls and notably the trackers and more recently, the '50 state polls'.  If they knew what data was coming this cycle, I bet they would have built the model differently, but they have what they have. 

I suspect that after the debates there will be more live-call polls and they will increase in frequency both at the state and national level. Then perhaps you will see less of a disparity between it and the Upshot model. 

[KingSweden]

I think the issue is that Silver's model most wants lots of live caller state polls. If every day it had Fox, Monmouth, Marquette, Marist, CNN, etc state (and national) polls feeding it, it would work best. It is essentially an engine tuned for those. But this cycle the vast majority of data being put into the model are the online polls and notably the trackers and more recently, the '50 state polls'.  If they knew what data was coming this cycle, I bet they would have built the model differently, but they have what they have. 

I suspect that after the debates there will be more live-call polls and they will increase in frequency both at the state and national level. Then perhaps you will see less of a disparity between it and the Upshot model. 

I think this is it

[Lief 🗽]

If you look at every other probability-based model (NYT Upshot, Sam Wang, Daily Kos), none of them have swung as wildly as Silver's. They have mostly stayed within a 70-90% Clinton range, which makes sense because the fundamentals of the race haven't really changed that much. But a good model that doesn't swing wildly from day to day doesn't bring in the page views. Nate just wants clicks. He's not interested in accurately forecasting anything.

[Maxwell]

If you look at every other probability-based model (NYT Upshot, Sam Wang, Daily Kos), none of them have swung as wildly as Silver's. They have mostly stayed within a 70-90% Clinton range, which makes sense because the fundamentals of the race haven't really changed that much. But a good model that doesn't swing wildly from day to day doesn't bring in the page views. Nate just wants clicks. He's not interested in accurately forecasting anything.

Silver's old model looked fairly stable even as Obama was only leading by like 2-4 points.

[Mallow]

If you look at every other probability-based model (NYT Upshot, Sam Wang, Daily Kos), none of them have swung as wildly as Silver's. They have mostly stayed within a 70-90% Clinton range, which makes sense because the fundamentals of the race haven't really changed that much. But a good model that doesn't swing wildly from day to day doesn't bring in the page views. Nate just wants clicks. He's not interested in accurately forecasting anything.

People who think 538 is intentionally being misleading are annoying. Their methodology was largely determined long ago, they haven't changed anything substantial as far as I'm aware (though based on some recent commentary on here, maybe I missed something?). Their model is showing what their model is showing, for better or worse. Complain about their methodology all you want, but to claim intentional bias for page clicks is dumb.

As for a possible explanation for the recent Trump swing, I think this is a possible culprit: the 538 model relies pretty substantially on the "trend" line. The Ipsos poll was added in three sections, the first being the 9/13-9/17 results, and the last being the 9/15-9/19 results. The three polls are Clinton +2, Trump +1, and Trump +2, so it looks to the model like a strong short-term trend from a highly-rated pollster.

[elcorazon]

Agreed. Plus I disagree with notion his model has been wildly moving. The Polls Plus in particular, but Polls Only as well, were the steadiest for awhile as it was other models that had Hillary way ahead early on. It's really been that Trump is continually chipping away at Clinton's lead for so long the numbers have ultimately changed dramatically. If you're saying the race really hasn't changed despite the fact that polling has changed, you're talking about something other than a model based primarily on polling.

[bilaps]

Also, he just gave Nate Cohn an explanation for adjusting Siena poll to Trump +1. He basicaly says that Siena was D leaning this cycle because their NY polls were strongest for Clinton out of 10 different pollsters.

Also says that trend goes into that 2 point adjustment, because his model shows Trump still gaining.

[heatcharger]

The problem is Florida. Why is Trump still favored, and why did his odds go up today? I know, the trend, but Monmouth last polled it a month ago, so at that point, the statute of limitations for a trend should go out the window. It could've been a lot worse if Monmouth conducted the poll last week.

Also, he just gave Nate Cohn an explanation for adjusting Siena poll to Trump +1. He basicaly says that Siena was D leaning this cycle because their NY polls were strongest for Clinton out of 10 different pollsters.

Silly reasoning. Safe state polling is almost always less accurate than swing states. Are Nate & Co. really basing their 2 point D-house effect on some NY Siena polls? Isn't the methodology of the FL Upshot/Siena poll also completely different?

[amdcpus]

The problem is Florida. Why is Trump still favored, and why did his odds go up today? I know, the trend, but Monmouth last polled it a month ago, so at that point, the statute of limitations for a trend should go out the window. It could've been a lot worse if Monmouth conducted the poll last week.

Also, he just gave Nate Cohn an explanation for adjusting Siena poll to Trump +1. He basicaly says that Siena was D leaning this cycle because their NY polls were strongest for Clinton out of 10 different pollsters.

Silly reasoning. Safe state polling is almost always less accurate than swing states. Are Nate & Co. really basing their 2 point D-house effect on some NY Siena polls? Isn't the methodology of the FL Upshot/Siena poll also completely different?

St Leo University Poll today of Florida had Clinton at +5 when a month ago they had her at +14.

[Wells]

The problem is Florida. Why is Trump still favored, and why did his odds go up today? I know, the trend, but Monmouth last polled it a month ago, so at that point, the statute of limitations for a trend should go out the window. It could've been a lot worse if Monmouth conducted the poll last week.

Also, he just gave Nate Cohn an explanation for adjusting Siena poll to Trump +1. He basicaly says that Siena was D leaning this cycle because their NY polls were strongest for Clinton out of 10 different pollsters.

Silly reasoning. Safe state polling is almost always less accurate than swing states. Are Nate & Co. really basing their 2 point D-house effect on some NY Siena polls? Isn't the methodology of the FL Upshot/Siena poll also completely different?

St Leo University Poll today of Florida had Clinton at +5 when a month ago they had her at +14.

Read the first paragraph of that post and try again.

[Classic Conservative]

The problem is Florida. Why is Trump still favored, and why did his odds go up today? I know, the trend, but Monmouth last polled it a month ago, so at that point, the statute of limitations for a trend should go out the window. It could've been a lot worse if Monmouth conducted the poll last week.

Also, he just gave Nate Cohn an explanation for adjusting Siena poll to Trump +1. He basicaly says that Siena was D leaning this cycle because their NY polls were strongest for Clinton out of 10 different pollsters.

Silly reasoning. Safe state polling is almost always less accurate than swing states. Are Nate & Co. really basing their 2 point D-house effect on some NY Siena polls? Isn't the methodology of the FL Upshot/Siena poll also completely different?

St Leo University Poll today of Florida had Clinton at +5 when a month ago they had her at +14.

Read the first paragraph of that post and try again.

That's why, it's the poll has a large Democratic bias of like 4 points so 538 put the poll at +2 I think. Which fits in with the NYT Poll

[PaperKooper]

I just think that switching the "default" prediction to poll-only was a mistake.  He originally did that since his polls-plus model significantly underestimated Trump in the primaries.  His polls-only model seems to understate the probability of the polls shifting significantly.  Part of that may be that they are shifting more this election than the '12 and '08 elections due to the widely-disliked candidates. 

[Fmr President & Senator Polnut]

Silver reacted way too much to the beating he got for underestimating Trump. This "every data point matters" attitude is why his model is so much more volatile. His reputation is based on getting things right, so he's covering his arse by including EVERYTHING and hoping the model will do the work.

[Gass3268]

In the polls only, Democrats have a better chance of winning the Senate than Clinton winning the Presidency, lol.

[ProudModerate2]

Why didn't 538 update the system to include the two new Florida polls that were released earlier today (Tue Sept 20th) ?
I can possibly understand that they are unsure to include the St Leo University poll, but why would they not include the Monmouth poll ?
(PS: Both polls are +5 Clinton.)

[Atlas Has Shrugged]

Why didn't 538 update the system to include the two new Florida polls that were released earlier today (Tue Sept 20th) ?
I can possibly understand that they are unsure to include the St Leo University poll, but why would they not include the Monmouth poll ?
(PS: Both polls are +5 Clinton.)

See, you are trying to rig the polls now! Sad!

[ProudModerate2]

Why didn't 538 update the system to include the two new Florida polls that were released earlier today (Tue Sept 20th) ?
I can possibly understand that they are unsure to include the St Leo University poll, but why would they not include the Monmouth poll ?
(PS: Both polls are +5 Clinton.)

See, you are trying to rig the polls now! Sad!

You lost me.

[Classic Conservative]

Why didn't 538 update the system to include the two new Florida polls that were released earlier today (Tue Sept 20th) ?
I can possibly understand that they are unsure to include the St Leo University poll, but why would they not include the Monmouth poll ?
(PS: Both polls are +5 Clinton.)

Monmouth is on 538 as C+4

[JRP1994]

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.

[Fmr President & Senator Polnut]

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?

[Mallow]

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.