2018 U.S election

[onejayhawk]

One week to go. Time for predictions.

RCP averages have the Democrats winning the House with a whopping three vote majority. Republicans gain two seats in the Senate.
538 has Democrats winning the House with a 31 vote majority. Republicans gain one seat in the Senate.
These are the mean numbers of the probability spread.

Literally, everyone expects Democrats to win seats in the House. The question is whether they are +23, which is the number they need for a majority.

The Iowa University political market is about 8-1 Democrats gain control and the same for Republicans maintaining or increasing their Senate advantage. There is also a market for overall control. It is showing about 2/3 preference for a split Congress, but almost 20% for "Other", primarily a Democratic House and a 50/50 Senate.

I have just seen over/under lines of 31 and 1. That translates to a 17 vote Democrat majority in the House and a four vote Republican majority in the Senate.

https://www.realclearpolitics.com/epolls/2018/house/2018_elections_house_map.html
https://www.realclearpolitics.com/epolls/2018/senate/2018_elections_senate_map.html
https://projects.fivethirtyeight.com/2018-midterm-election-forecast/senate/?ex_cid=rrpromo
https://projects.fivethirtyeight.com/2018-midterm-election-forecast/house/?ex_cid=midterms-header
https://iemweb.biz.uiowa.edu/

J

 

[Timsup2nothin]

One week to go. Time for predictions.

Why? I'm quite satisfied with probabilities at this point, and have far more interest in what I can do in this final week to oust my horrific incumbent congressman. All that matters is that however much of a slap in D'ump's face happens on November 6th I will have done my part.

 

[Zkribbler]

One week to go. Time for predictions.

I predict all predictions will be wrong.

 

[Takhisis]

Zkrib, you disagree with this here post.

 

[rah]

We have about a dozens models running and needless to say, the output is based on the input.
We did have a couple of models that predicted a Trump win back in 2016 (out of about 20)

 

[AdrienIer]

I like the 538 articles, they give a lot of context around their model. So it's more than just a "85% chances for democrats to win the house". If there's generally a small polling error in favor of the democrats the republican will keep the house and will have a large majority in the senate. If the fundraising (translated into name recognition and awareness) ends up having a larger effect than expected the dems could potentially get 50 or 60 seats in the house and win the senate. Also they're pretty angry that some districts are polled all the time while others have literally 0 polls. So there may be a few blind spots in the models, which would rate a district "likely R" because of its history when it's actually very close.

 

[onejayhawk]

I predict all predictions will be wrong.

Impossible, because all possible outcomes will be predicted.

We have about a dozens models running and needless to say, the output is based on the input.
We did have a couple of models that predicted a Trump win back in 2016 (out of about 20)

No, we didn't. Not one.

Specifically not 538, Sabato, Cooks, or any of the poll collectors, such as RCP, HuffPo, KOS, electoral-vote, or NYT. LA Times had a recurring poll that was closer than any of the others, but it still had Hillary winning. The closest we had to someone getting it right were correct calls of a few specific states, like Michigan and Wisconsin.

It should be noted that the ~3% popular vote margin covers more than half the national poll lead Clinton enjoyed. That's very relevant to this election because the House mimics the electoral college.

J

 

[metalhead]

You still don't know how a statistical model works. Saying there is a 70% chance Hillary will win is not the same thing as predicting Hillary will win.

 

[Timsup2nothin]

You still don't know how a statistical model works. Saying there is a 70% chance Hillary will win is not the same thing as predicting Hillary will win.

C'mon man, you didn't seriously expect any change in J's apparent knowledge level.

 

[onejayhawk]

You still don't know how a statistical model works. Saying there is a 70% chance Hillary will win is not the same thing as predicting Hillary will win.

It actually is saying that Hillary will win, with 70% confidence. When confidence exceeds 90%, which it did, then that is as close to an outright prediction as poll-heads are willing to get.

Regardless, the point is that the 70% confidence was proven to be misplaced. 20/20 hindsight is like that. There were several long articles why they blew it and why it was not as bad a fumble as it seems. However, they admit that they did miss the mark.

J

 

[JollyRoger]

Impossible, because all possible outcomes will be predicted.


No, we didn't. Not one.

Specifically not 538, Sabato, Cooks, or any of the poll collectors, such as RCP, HuffPo, KOS, electoral-vote, or NYT. LA Times had a recurring poll that was closer than any of the others, but it still had Hillary winning. The closest we had to someone getting it right were correct calls of a few specific states, like Michigan and Wisconsin.

It should be noted that the ~3% popular vote margin covers more than half the national poll lead Clinton enjoyed. That's very relevant to this election because the House mimics the electoral college.

J

The House does not mimic the Electoral College. First, plenty of one seat states get 3 electoral votes. Plus the electoral college is winner take all I most states while House seats are decided on a district by district basis.

 

[rah]

No, we didn't. Not one.

Specifically not 538, Sabato, Cooks, or any of the poll collectors, such as RCP, HuffPo, KOS, electoral-vote, or NYT. LA Times had a recurring poll that was closer than any of the others, but it still had Hillary winning. The closest we had to someone getting it right were correct calls of a few specific states, like Michigan and Wisconsin.

What do you mean "we." The company I worked for did. As I said we had a couple out of all the ones we did. I notice you didn't list the company I work for.

 

[onejayhawk]

What do you mean "we." The company I worked for did. As I said we had a couple out of all the ones we did. I notice you didn't list the company I work for.

Which company is that? I'm going off the cuff, so I might have missed one. I certainly do not remember any.

We refers the people in this forum that pay attention to such things.

J

 

[Timsup2nothin]

We refers the people in this forum that pay attention to such things.

J

At a guess, there is hardly anyone in this forum who would appreciate you referring to them in a "we" with yourself.

 

[JollyRoger]

Considering the number of Senate seats the Dems are defending in Trump states, the GOP will far underperform the map if they don't pick up more than 5 to 10 seats. Looks like a Red Wave folks.

 

[metalhead]

It actually is saying that Hillary will win, with 70% confidence.

It most certainly is not.

 

[onejayhawk]

It most certainly is not.

Fine. If you must be picky, it means that the probability of the null hypothesis--Hillary Clinton does not win--is at 30%. As I said, semantics. This level of precise language is actively unhelpful.

J

 

[Timsup2nothin]

This level of precise language is actively unhelpful.

J

Why would you expect anyone to actively help you with what you are doing? It seems pretty apparent that your posting makes most people want to vomit, not help. There are things I would help you with, but there's no sense listing them.

Moderator Action: Warned for trolling. Please minimize the personal attacks. --LM
Please read the forum rules: http://forums.civfanatics.com/showthread.php?t=422889

 

[metalhead]

Fine. If you must be picky, it means that the probability of the null hypothesis--Hillary Clinton does not win--is at 30%. As I said, semantics. This level of precise language is actively unhelpful.

J

It's not semantics, you just don't understand what you're talking about. As evidenced by your gross misuse of the term "null hypothesis."

 

[rah]

Which company is that? I'm going off the cuff, so I might have missed one. I certainly do not remember any.

We refers the people in this forum that pay attention to such things.

J

Ipsos
We do lots of different scenarios but usually only publish a few of the higher percentage models. You'll see a lot of our stuff used by Reuters
I would be shocked if all the other companies didn't do the same.