HomeLatest ThreadsGreatest ThreadsForums & GroupsMy SubscriptionsMy Posts
DU Home » Latest Threads » Forums & Groups » Retired » Retired Forums » 2016 Postmortem (Forum) » The "How many states do y...

Sun Nov 6, 2016, 12:25 PM

The "How many states do you think 538 will get wrong?" poll

The famed 538 model evolved from baseball (and then football, etc) where a regular schedule uses a steady flow of information of equal quality. It suffers during primaries where quality and quantity are at odds. The quality of polls suffers as election day draws closer as Republicans flood results from Republican pollsters who have not been active previously and appear to be aimed at raising voter turnout among Republicans in key swing states.

The state polls simply are not in line with SOME of the national polls as discussed here

http://www.democraticunderground.com/12512589061

You canít have state polls showing an aggregate huge margin (i.e. NY/CA up 20 and Texas down 10, etc) and a national poll at 3 points or less. To make the state polls come in line with the national Silver makes manual adjustments which have been running 2-3 points for Trump.

FiveThirtyEight has doubled down on their logic with this article just out that argues that the early voting in NV could be wrong and their polls right because there is large numbers of Democrats voting for Trump (even though the state polls that they prefer in fact donít show that) I donít know how they justify Nevada going blue for the Senate but Red for the Presidential election.

http://fivethirtyeight.com/features/has-trump-already-lost-nevada/?ex_cid=2016-forecast

Being a registered Democrat doesnít necessarily mean youíre going to vote for Clinton. If Trump is winning more registered Democrats than Clinton is registered Republicans, the early vote data in Nevada may not mean what we think it means. Indeed, some Nevada polls (though not all) show Trump getting a higher percentage of self-identified Democrats than Clinton gets Republicans. Itís also possible that Republicans turnout in disproportionately strong numbers on Election Day, despite previous trends


With this logic, which is not based on supportable data, 538 shows Democrats winning the senate race in Nevada and Trump winning the Presidential election.

With 48 hours left 538 is showing the following states red: Florida, North Carolina, Nevada, Ohio, Iowa, Georgia and Arizona for the President

He is showing the following states going red for Senate:

Florida, North Carolina, Missouri, Indiana.

In these 11 states how many do you think he will get wrong?

15 votes, 0 passes | Time left: Unlimited
Wrong 9 out of 11 - "I guess some polling should be done in Spanish"
1 (7%)
Wrong 8 out of 11 - "You mean Republicans publish dishonest polls?"
0 (0%)
Wrong 7 out of 11 - "You mean people were actually listening to me?"
1 (7%)
Wrong 6 out of 11 - "But the National trend line has to be close"
2 (13%)
Wrong 5 out of 11 - "So Early Voting Numbers actually count"
7 (47%)
Wrong 4 out of 11 - "But Trump fooled me in the Primaries"
0 (0%)
Wrong 3 out of 11 - "So what if Wang got em all right, I have the model"
3 (20%)
Wrong 2 out of 11 - "The model is sad"
1 (7%)
Wrong 1 out of 11 - "Well I did get the Electoral College winner right".
0 (0%)
Silver is Gold, all correct
0 (0%)
Show usernames
Disclaimer: This is an Internet poll

12 replies, 1313 views

Reply to this thread

Back to top Alert abuse

Always highlight: 10 newest replies | Replies posted after I mark a forum
Replies to this discussion thread
Arrow 10 replies Author Time Post
Reply The "How many states do you think 538 will get wrong?" poll (Original post)
grantcart Nov 2016 OP
Foggyhill Nov 2016 #2
grantcart Nov 2016 #4
obamanut2012 Nov 2016 #8
grantcart Nov 2016 #11
Amishman Nov 2016 #5
Imperialism Inc. Nov 2016 #6
metroins Nov 2016 #7
grantcart Nov 2016 #10
BootinUp Nov 2016 #9
LineLineReply
Dec 1969 #

Response to grantcart (Original post)

Sun Nov 6, 2016, 12:30 PM

2. He;He;ll adjust his crap to follow concensus so he's not wrong at the last minute. He's done for me.

Reply to this post

Back to top Alert abuse Link here Permalink


Response to Foggyhill (Reply #2)

Sun Nov 6, 2016, 12:31 PM

4. And that is why I am doing the poll now, lol


Should have done it yesterday when he had NV Senate going red.

Reply to this post

Back to top Alert abuse Link here Permalink


Response to grantcart (Reply #4)

Sun Nov 6, 2016, 01:06 PM

8. His Neveada numbers is the reason I am 86ing him

At least until he updates his model in the future. Hopefully.

Reply to this post

Back to top Alert abuse Link here Permalink


Response to obamanut2012 (Reply #8)

Sun Nov 6, 2016, 01:33 PM

11. You might be interested in reply # 10.

There is good peer review analysis that simpler models out perform complex ones, that the more you try to manipulate the data the more variables you have to account for.

Reply to this post

Back to top Alert abuse Link here Permalink


Response to grantcart (Original post)

Sun Nov 6, 2016, 12:43 PM

5. I think seven, 5 going hillary and two going trump

for some reason I really think Trump is going to surprise and take NH and CO. NH on opposition to gun policies (that state loves their guns), and I think without the marijuana ballot initiatives CO is isn't quite a true blue state yet (but will be soon)

I think Hillary will take FL, NC, NV, AZ, and AK. The first four are due to the rise of the hispanic community, and AK is just a gut feeling.

The electoral college is not going to be close at all once the dust settles. Hillary has this locked up.

Reply to this post

Back to top Alert abuse Link here Permalink


Response to grantcart (Original post)

Sun Nov 6, 2016, 12:48 PM

6. I've been really embarrassed about fellow Democrats and their reaction to

Nate Silver's model this time around. It has taken away one of my great joys from 2012, being able to laugh at how Republicans were poll-truthers. The way they ignored models like Nate Silver's (that was basically the same then as now minus some minor tweaks) that showed a steady, high chance of Obama winning. And, who can forget that wonderful moment with Rove on Fox when Megyn Kelly asked him, "are those numbers real, or is that just math you do as a Republican to make yourself feel better?"

The story this time has been every time Silver's model tightens he is spat upon, but when it shows things going very well he is embraced. This time people have gone searching for the number that makes them feel the best. Just look at the love Sam Wang gets. I wish we could have a 100 elections so I could bet against the people who really think Clinton has a 99% chance of victory. Given the polls and the levels of uncertainty (such as possible polling error, and number of undecided and third party voters) anyone who thinks only one out of a hundred times would end up being a loss are delusional.

Is Silver's number too low? Probably. None of the models account for early voting, for example. Still, it is nowhere near a 99% win rate.

Reply to this post

Back to top Alert abuse Link here Permalink


Response to Imperialism Inc. (Reply #6)

Sun Nov 6, 2016, 01:03 PM

7. I agree with you

538 doesn't include anything other than polls, because that's what they do. Trying to get him to add in early voting is ludicrous, especially so close to the election.

I personally think Hillary will win FL but I'm not going to bash Nate Silver. His website does good analysis.

The republicans ignored polls in 2012 and I got an amazing laugh at Karl Rove that night.

Reply to this post

Back to top Alert abuse Link here Permalink


Response to Imperialism Inc. (Reply #6)

Sun Nov 6, 2016, 01:31 PM

10. Have you been reading Nate's commentary?

He has clearly stated that he is the "most bullish on Trump" of all of those that are projecting.

He was way off on the primaries and he is now making manual adjustments to his numbers to enhance Trumps numbers. This "unskewing" is unprecedented for 538, discussed here by Huffington Post

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

You missed the point about counting Early Vote. Count them or don't count them and your model will stand or fall by the conclusions it arrives at. Enten in his article today is making arguments against the EV totals being accurate predictors because he is saying polls indicate that there is a sizeable move of Democrats voting for Trump. There are NO polls that show that there is any move by Democrats in Nevada to support Trump at lower levels of Republican support for Clinton, none.

The reason that Sam Wang gets more love is because he uses the most reliable polls: state polls and doesn't try to manipulate state polls with national polls which Silver openly concedes that he does.



National polls versus state polls

To recap, the model mostly uses state polls. But national polls can influence the forecast in some subtle ways:
◾Theyíre helpful for calculating adjustments to the polls, especially the trend line adjustment and house effects adjustment.
◾Theyíre used, in conjunction with the state polls, in estimating the national popular vote.



Its Silver's bad luck that a few national polls are probably under rating Clinton by 2-4 points, but that is what the early voting is showing, probably because they are a) under estimating Republican women voting for Clinton and b) significantly under estimating Hispanic turnout.

You suggest that people are simply "cherry picking" to find the polls that make them feel the best. That might be partially true that people are looking for polls that they think most accurately reflect their "feeling" but actually DUers are much more sophisticated consumers of polling data than your slam would give them credit for.

538 is the most complex model. Peer review analysis has showed however that more complex models don't have a better outcome than simpler models.

The most simple model is Polyvote http://pollyvote.com/en/ and are showing Clinton 53 Romney 47, while 538 is showing 48/43. Let's revisit the numbers after the election and see who got closer.

The fact is that for all its pretense of being an objective data driven model 538 "readjusts" the data for historical, economic, national poll factors. There is substantial criticism against 538 that has nothing to do with "cherry picking" polls but with substantial criticisms with their methodology and the frustration that they have a higher media profile than other election forecasters that are, in fact, more empirically based that Silver.

Reply to this post

Back to top Alert abuse Link here Permalink


Response to grantcart (Original post)

Sun Nov 6, 2016, 01:21 PM

9. The problem with polls is that many

People are not committing to one the top 2 candidates. Therefore, there is more uncertainty. I think Clinton will take FL, Nev and either NC or OH.

Reply to this post

Back to top Alert abuse Link here Permalink

Reply to this thread