The Fundamental Problem with Nate Silver's Model

[grantcart]

The fundamental problem with Nate Silver’s model is that the state polls cannot be reconciled with the national trend. You cannot have states like NY, CA, WA up 20 points, another dozen states where Clinton is up 15 points and the most populous red states like TX, AZ and Florida where she is even or down just a little and then have the national polls where Clinton is only up 3%, it just doesn’t add up.

EITHER HUNDREDS OF STATE POLLS ARE WAY OFF OR A FEW NATIONAL POLLS ARE A FEW POINTS OFF. SILVER HAS DECIDED THAT THE NATIONAL POLLS ARE GOLD AND ADJUSTS DOZENS OF STATE POLLS IN ORDER TO MAKE A SQUARE PEG FIT A ROUND HOLE.

The Princeton consortium never uses the national polls as the state polls are much more accurate and they don’t have to adjust to make them fit

Silver’s model is based on his sports models where there is a constant stream of games being played on schedule with a set data stream. One of the problems with the last 2 weeks of the election season is that Republicans launch a barrage of polls from RW pollsters (who ever heard of Remington?) that lean heavy Republican that are used to help with their GOTV.

Silver’s top down approach that gives preeminence to national polls is not as stable as the bottom up approach that other sites use that work on a larger and more accurate base of state polls. If you accept that Clinton is currently 6% ahead nationally then you can reconcile all of the state polls without major adjustment.

[manicraven]

1. Spot on!

The fundamental problem with Nate Silver’s model is that the state polls cannot be reconciled with the national trend.

That makes sense!

[grantcart]

4. Thats a great picture.

[Avalux]

2. Thank you for this concise explanation!

I agree - he's got it backwards.

[Hortensis]

3. Love the conclusion. I've seen a number of people say

national polls are mostly bogus at this point and that state polls are far more reliable. Also that the best are the extremely well funded internal polls of both candidates--algorhithms fed with vast amounts of information they've been gathering about voters, right down to what kind of underwear they prefer if believed relevant.

Come to think of it, that's a correlation that would be fun to check out.

[DLCWIdem]

6. HOW about early vote. How does that for in

Does Nate' s model look at early vote at all?

[grantcart]

7. He doesn't factor that in. And it doesn't make sense either

NV has a huge D lead and he still has it Red.

[GopherGal]

16. Doesn't explicitly include early vote

But his model prefers poll results for "likely voters". I don't know how that works within the polls themselves, but I would think that if there's an "already voted" category that boosts respondents' scores in "likely voter" weighting, the polls themselves (rather than the aggregation model) might be indirectly accounting for the early vote.

[yardwork]

8. So glad to see your posts again!

[Awsi Dooger]

9. Sorry, the state polls make no sense in relation to a 6 point Hillary lead

Not even close. They are much more in line with a 3 point national advantage.

For example, if Hillary leads nationally by 6 points but her Pennsylvania edge is 4-5 points, then you're trying to tell me Pennsylvania has a slight red tilt. The states you picked are not in play. No advertising blitz. The margins can shift considerably in situations like that from cycle to cycle. Republicans used to contest Washington state more than they do now, so naturally the blue edge can rise.

I've used the partisan index for 20 years and used to post the statewide numbers here more than a decade ago, before it was widespread to do so.

Keep in mind that this cycle is challenging from a partisan index perspective because Trump has unusual strength in states with high percentage of whites but not as many Hispanics or blacks than other swing states. The partisan index in states like Ohio and especially Iowa figures to shift into territory we haven't seen before, at least not recently. Those states may also move against us in general, similar to states like West Virginia and Missouri over the past 15 years or so.

Many states are in partisan index flux currently. Quite interesting. Most in our favor but some of them are still Hail Mary caliber, like Arizona and Georgia and Texas.

[grantcart]

10. Absolute nonsense

You are only quoting the polls in the swing states

California +22
New York + 20
Mass + 22
Washington, Maine, Conn, Deleware, IL Hawaii, MD, VT all around +15

On the Republican side you have a lot of low populations states that have wide margins but their large states are very narrow margins

Texas -9
AZ - 5

and a whole bunch of Atlantic states that have been red but are now tied or blue
Virginia
North Carolina
Florida
Georgia

Its not really a point of contention. Silver explicitly states that he is moving the state polls by 3 points so that they mesh with the national numbers. He has a term for it, he calls it "trend line adjustment".

[BzaDem]

11. His model is mostly based on state polls.

He does use the national polls to adjust state polls, but he says that only 10-20% of the uncertainty comes from that. He says the other 80-90% comes from other factors (larger than normal undecided/third-party factor, higher variability in the polls, correlations between states, etc.)

He also said that if the national polls all uniformly ticked up a point, his model would show Trump as still having a 25-30% chance of winning.

[pnwmom]

15. How Nate Silver "unskews" the polls -- and why he shouldn't.

http://www.huffingtonpost.com/entry/nate-silver-election-forecast_us_581e1c33e4b0d9ce6fbc6f7f

[Cha]

12. Mahalo, grant! This is kind of what I was gathering from reading

what was being said about Nate's models and Sam Wang's methods.

Just saw this tweet from Sam..

Sam Wang ‏@SamWangPhD · 6h6 hours ago
Sam Wang Retweeted Clara Jeffery
Good thing there isn't anything serious to do, like figure out what's happening in Senate, or interpret early voting, or go get out the vote

https://twitter.com/SamWangPhD?ref_src=twsrc%5Egoogle%7Ctwcamp%5Eserp%7Ctwgr%5Eauthor

[grantcart]

13. Hey Cha

Here is something to think about

With the enhanced Hispanic turnout I wonder if in the Southern States with large Hispanic population that we may do better in the house than people are thinking about. Maybe 20+ seats.

[Cha]

14. Good point! And, considering Team Hillary's ground game, too

tephanie ‎@FormasForever
Listen, there is no ground game better than @HillaryforPA. By 6 PM, our vols knocked on 621,720 doors. That's 69,080 an hour. #proudtobePA
2:44 PM - 5 Nov 2016
72 72 Retweets 100 100 likes

https://twitter.com/FormasForever?ref_src=twsrc%5Etfw

Good to see you back