Rassmussen vs Nate Silver

[mjrpes3]

His blog is here:

http://fivethirtyeight.blogs.nytimes.com/

Last election he got 49 states right.This time looks like he's going to get 50 right. Well done Nate.

 

[Phokus]

A seventh grader could make a predictive system. It isn't magic, I'll give him credit for believing the polls because they ended up being right.

 

[shira]

Nate Silver has actually discussed Rasmussen several times on his blog, and it's not all negative. He tends to hold them up as a better-than-average pollster, and the accusations of bias against them are often overblown. I do find it hard to believe that Wisconsin is a tie, against what every other polling firm is predicting. We won't know until tonight or tomorrow.

There have been a few cases where they were way off in favor of the Republican. Notably, a senate race in Hawaii where Rasmussen called a 13% lead for the Democratic incumbent and in reality wound up with a 53% lead.

EDIT: Also I agree, state polls are more important. Popular vote doesn't matter. I'd be interested to see the final Rasmussen state-by-state predictions for 2008, but I can't find them, only their prediction of the national popular vote (which was very close).

Rasmussen is actually careful in their methodology and consistent in their results, and Silver incorporates their polls into his aggregate after first applying his pollster-specific adjustment to account for Rasmussen's "house effect." All polling firms have house effects, and as long as they're accurately modeled, the aggregate will have a much smaller uncertainty than the individual polls (assuming, of course, there isn't the same systemic error in all polls).

To put this another way: As long as you recognize that Rasmussen's numbers are going to be 3 points right of where they should be, you can get a lot of info from them.

 

[jonks]

One thing no one on the left has addressed is the outsized Democrat representation in all polls except Rasmussen and Gallup. In an election where everyone agrees Obama has less support than 2008, how does one justify assuming a heavier Democrat voting base?

I can't tell if you're asking rhetorically, but the answer in any event, is that this is their job. Being accurate, not being biased. They care about the final number and getting it right. Whatever metric they used to determine likely voters and turnout was apparently correct, even if unskewedpolls, Dick Morris, Rove, George Will, and many others refused to believe it because it made their guy look bad. You get your own opinions, not your own facts.

 

[buckshot24]

Take a fucking cue from Romney and be gracious in defeat. Being a snippy little bitch about it is precisely the wrong way to act when you are shown incorrect. In particular because when you say shit like a seventh grader could have done what Silver did, the question becomes how did you, presumably an adult, fuck it up so badly?

Accept that Nate Silver is skilled at parsing polls and getting to the heart of what they say and move on.

I never said Silver was wrong in his methodology. If his model was incorrect it would be because of the polling data not because of the model itself. I thought the polls were over estimating Democratic turnout which turned out to be 100% wrong.

If we want to be picky. If Silver "predicts" 50 out of 50 states this is actually a failure of his model, at least of his percentage calculations, since there should have been some misses.

Colorado and Virginia both with 80% chance for Obama, 84% chance for Iowa and 50.3% chance for Florida to go Obama. If these percentages actually mean anything the odds are that one of them would have missed. There should have been a 27% chance of getting all of those states right.

If you take all of his probabilities he should have been right 10% of the time on every one of his picks.

His projections on probabilities will be tweaked, I'm sure of it. He's mentioned this in his blog.

Anyway, I was wrong about the polls not Nate Silver.

 

[Abraxas]

I never said Silver was wrong in his methodology. If his model was incorrect it would be because of the polling data not because of the model itself. I thought the polls were over estimating Democratic turnout which turned out to be 100% wrong.

If we want to be picky. If Silver "predicts" 50 out of 50 states this is actually a failure of his model, at least of his percentage calculations, since there should have been some misses.

Colorado and Virginia both with 80% chance for Obama, 84% chance for Iowa and 50.3% chance for Florida to go Obama. If these percentages actually mean anything the odds are that one of them would have missed. There should have been a 27% chance of getting all of those states right.

If you take all of his probabilities he should have been right 10% of the time on every one of his picks.

His projections on probabilities will be tweaked, I'm sure of it. He's mentioned this in his blog.

Anyway, I was wrong about the polls not Nate Silver.

That is certainly true to an extent, but at the same time, what his model also produced is the scenario of greatest likelihood. While it is correct to say his model should not be exactly right every time, his predictions do come to the conclusion about which, if a selection was taken once at random, electoral map was most likely to come up, and that is the one we see tonight.

That is to say it was not most likely this would be the outcome but of all possible outcomes this was the most likely based on his model.

EDIT: And yes, I fully expect him to learn from this and refine his model further. That he got the states right does not mean his model is perfect and I'm sure he will work to improve it for 2014/2016 and beyond.

 

[Phokus]

I never said Silver was wrong in his methodology. If his model was incorrect it would be because of the polling data not because of the model itself. I thought the polls were over estimating Democratic turnout which turned out to be 100% wrong.

If we want to be picky. If Silver "predicts" 50 out of 50 states this is actually a failure of his model, at least of his percentage calculations, since there should have been some misses.

Colorado and Virginia both with 80% chance for Obama, 84% chance for Iowa and 50.3% chance for Florida to go Obama. If these percentages actually mean anything the odds are that one of them would have missed. There should have been a 27% chance of getting all of those states right.

If you take all of his probabilities he should have been right 10% of the time on every one of his picks.

His projections on probabilities will be tweaked, I'm sure of it. He's mentioned this in his blog.

Anyway, I was wrong about the polls not Nate Silver.

The whole point of aggregating polls is to eliminate bias you bitch idiot republican.

 

[First]

If Silver "predicts" 50 out of 50 states this is actually a failure of his model, at least of his percentage calculations, since there should have been some misses.

Colorado and Virginia both with 80% chance for Obama, 84% chance for Iowa and 50.3% chance for Florida to go Obama. If these percentages actually mean anything the odds are that one of them would have missed. There should have been a 27% chance of getting all of those states right.

Lord. The minority likelihood deltas of only 10% or 20% aren't normal odds like dice or cards, the remaining doubt for most battlegrounds was predicated on unlikely scenarios that the polls were systematically biased in one direction or the other. That was it. This is something Silver has mentioned in the last 24 hours.

As a layman, though, you wouldn't understand.

 

[buckshot24]

The whole point of aggregating polls is to eliminate bias you bitch idiot republican.

You have no idea on what you are talking about.

 

[Charles Kozierok]

Colorado and Virginia both with 80% chance for Obama, 84% chance for Iowa and 50.3% chance for Florida to go Obama. If these percentages actually mean anything the odds are that one of them would have missed. There should have been a 27% chance of getting all of those states right.

Step 1: Put down the shovel.
Step 2: Learn about how probabilities work.

 

[OneOfTheseDays]

You have no idea on what you are talking about.

He actually does. You are most certainly a bitch idiot. Also a loser. Go fuck yourself.

 

[Abraxas]

Step 1: Put down the shovel.
Step 2: Learn about how probabilities work.

To be fair to him, he isn't wrong, at least not completely. When you have multiple events of less than certain probability in a sample the odds of them all happening concurrently diminish. I think he is neglecting Silver's point that the state polls are interconnected and that a strong turnout in, say Virginia is likely to mean a strong turnout in Ohio etc. too, but it was always possible that Florida or Virginia could have had above average republican turnout or below average Democratic returns and the more states you had, the more possibility for at least one of them to deviate from the model.

That said, the model does produce a single most probable scenario and we're living in it, that we happened to land on it even though probability dictates in most cases we won't is in no way a knock against the model.

 

[ohnoes]

i doubled down on eleven 5 times in a row and lost all 5. the house must be cheating.

 

[Phokus]

http://www.freerepublic.com/focus/f-bloggers/2955832/posts

Crybaby republicans who hate nate silver should just admit they don't understand/hate math.

 

[woolfe9999]

Yep, 50 of 50 this time. Even I still thought Florida would go narrowly to Romney. I said before that his prediction of Obama up by 2.5% on the PV was a stretch, and it looks like his actual margin will be a bit over 1%, but we'll see.

 

[First]

Yep, 50 of 50 this time. Even I still thought Florida would go narrowly to Romney. I said before that his prediction of Obama up by 2.5% on the PV was a stretch, and it looks like his actual margin will be a bit over 1%, but we'll see.

Actually don't think so, looks like Silver is going to get it exactly right. Did the math for remaining CA, NY, etc. precincts and if their margins stay the same (which may be a horrible assumption, didn't bother looking at the counties) Obama will win with somewhere in the mid-2.5M range.

Silver nailed the shit out of this election.

 

[stlc8tr]

Yep, 50 of 50 this time. Even I still thought Florida would go narrowly to Romney. I said before that his prediction of Obama up by 2.5% on the PV was a stretch, and it looks like his actual margin will be a bit over 1%, but we'll see.

How did he come up with a 313/225 EC split? Was that an average of the various scenarios?

Was that EC split even possible given the polls?

 

[Abraxas]

How did he come up with a 313/225 EC split? Was that an average of the various scenarios?

Was that EC split even possible given the polls?

The 313 was the average of then thousands of simulations he ran, not a specific prediction of how the states would fall out.

 

[Screech]

313 was basically the average of obama winning (332) and losing (303) florida biased by the possibility that he would lose ohio, virginia, or colorado (or a bunch of other less likely options).

Nate wins this round, and all the pundits who published columns along the lines of "how dare this asshole try to predict elections when we should be the sole arbiters of truth" look like the morons they are.

 

[stlc8tr]

The 313 was the average of then thousands of simulations he ran, not a specific prediction of how the states would fall out.

Interesting. How did the other aggregators do? I heard that Sam Wang at PEC was even more bullish on Obama than Silver.

 

[ohnoes]

This shouldn't be that hard to believe. Companies have been using statistics and mining data for a very long time. This is just applying the same principles to the election, and as much as we like to think of momentum and gut instinct as an unknown, it's not actually that big of a deal.

 

[First]

This shouldn't be that hard to believe. Companies have been using statistics and mining data for a very long time. This is just applying the same principles to the election, and as much as we like to think of momentum and gut instinct as an unknown, it's not actually that big of a deal.

Yup, basically this. A great example is sports; for the all bluster about "intangibles", in basketball there are great uni-aggregator statistics like PER that quite accurately encapsulates floor impact.

Bottom line; follow the numbers carefully and you'll get it right mostly.

 

[randomrogue]

And that settles that. Quite an impressive model on Nate's part.

Permanent ignore for Buckshot for being so incredibly ignorant.

 

[stlc8tr]

This shouldn't be that hard to believe. Companies have been using statistics and mining data for a very long time. This is just applying the same principles to the election, and as much as we like to think of momentum and gut instinct as an unknown, it's not actually that big of a deal.

Depending on how FL's count goes, Sam Wang's predictions seem pretty friggin spot on!

http://election.princeton.edu/2012/11/06/presidential-prediction-2012-final/

"ELECTORAL PREDICTION (mode): Barack Obama 303 EV, Mitt Romney 235 EV. The mode is the single most frequent value on the EV histogram. It corresponds to the map below, and has a 22% chance of being exactly correct. The next-most-likely outcome is Obama 332, Romney 206 EV."

"As I wrote late last night, Florida is a hard case. Several new polls came out this morning, making the median basically zero. As a tie-breaker I resorted to mean-based statistics. I will be unsurprised for it to go either way. Nate Silver and Drew Linzer went the other way. We are all tossing coins. I am prepared to lose the coin toss."

 

[Bird222]

That "gay" math is a motherfucker! LOL