Cook Report moves GA Senate race to "Toss Up"

[7,052,770]

Mississippi is not safe Republican...yet.

It's probably fair to list it as Safe Republican now.  We can change it in the unlikely event that McDaniel is the nominee, but don't count on that.

[7,052,770]

Silver currently sees GOP with 70% chance of winning GASen. That's better than KY and tied with AR.

To say that Pryor is as likely to win as Nunn is ludicrous.

Not sure which way you're going with it, but they're both within the margin of error in every single poll released thus far. Silver's work is obviously heavy on polling and there's not a lot of that yet - and we know his 2012 Senate results changed considerably between spring and fall - but the guy has a 96% success rate on Senate seats thus far.

By that logic, he should have both of them with a greater chance of victory. Instead, he uses his questionable 'state fundamentals' variable which has helped him blow some close races (see 2012 MT/ND). A 96% success rate isn't impressive. Calling close races correctly is impressive.

Again, Silver doesn't call races. Period. That's just not what he does. The idea that he has a "96% success rate so far" is absolutely an incorrect way to look at his record.

He gives a probability of each winning.  What makes him good is that 75% of the candidates he says have a 75% chance of winning end up winning.  If everyone he gives a 75% to wins, he's not good.

There was nothing "wrong" with him "missing" the ND race in 2012. He needs to be "wrong" 1 out of every 12-13 times he gives a 92% chance or else his probabilities aren't very accurate.

[7,052,770]

Silver currently sees GOP with 70% chance of winning GASen. That's better than KY and tied with AR.

To say that Pryor is as likely to win as Nunn is ludicrous.

Not sure which way you're going with it, but they're both within the margin of error in every single poll released thus far. Silver's work is obviously heavy on polling and there's not a lot of that yet - and we know his 2012 Senate results changed considerably between spring and fall - but the guy has a 96% success rate on Senate seats thus far.

By that logic, he should have both of them with a greater chance of victory. Instead, he uses his questionable 'state fundamentals' variable which has helped him blow some close races (see 2012 MT/ND). A 96% success rate isn't impressive. Calling close races correctly is impressive.

Again, Silver doesn't call races. Period. That's just not what he does. The idea that he has a "96% success rate so far" is absolutely an incorrect way to look at his record.

He gives a probability of each winning.  What makes him good is that 75% of the candidates he says have a 75% chance of winning end up winning.  If everyone he gives a 75% to wins, he's not good.

There was nothing "wrong" with him "missing" the ND race in 2012. He needs to be "wrong" 1 out of every 12-13 times he gives a 92% chance or else his probabilities aren't very accurate.

I hope your kidding. If you aren't, you just proved how useless Nate Silver is. What is the point of his model if not to "call" races. Why the hell have a model otherwise?

I don't really know what else to tell you. No one is going to build a computer model that can pick winners of 50-50 tossup races with any level of accuracy. Silver's model just gives probabilities, and while there hasn't been enough of sample (that will take years) to say it's definitely a good model, there's been no reason to doubt it yet.

Again, if Silver wasn't "wrong" once every 12-13 times he said a candidate had a 92% chance of winning, the model would be bad.

[7,052,770]

The point is to say what the probability is.  Whether "anyone can call 90% of these races" or not is completely irrelevant, because Nate Silver's model is not trying to call races -- it is assigning a win probability to each race.

[7,052,770]

Again, anyone who knows an ounce about politics can call up to 90% percent of races. The whole point of models is to predict tossup contests, otherwise there would be no point.

I'm not sure you understand what he's saying or what Silver does. His job isn't to call who is going to win or lose; his job is to determine the statistical likelihood of a candidate winning or losing an election. If he was just going to call the races, then he'd do it the same way that Cook, Sabato and Rothenburg do it by just guessing. Instead, he uses a series of data-sets and weights to project a confidence factor.

It makes more sense if you think about running each race in a simulation of sorts (which is what Silver and the Obama campaign both were doing to determine the likelihood of victory for one party or another in each state). So in May 2012, the GOP had a 70% chance of winning, according to Silver. That means if you ran the simulation 1000 times, the GOP would win 700 times and the Dems would win 300 times. Now obviously, you can't simulate such events in real life and there is only one election, but that doesn't matter in the context of how this data is supposed to be interpreted.

Yes, most races were off the table for one party or another, which is why those races had a 99%+ confidence factor - 18 Senate races, to be exact. North Dakota, in contrast, had a 70% likelihood of going to the GOP in May and a 80% chance in September. The final projection was 92.5% for the GOP in ND, and only 66% in MT - far cries from 99%+ confidence.

The fact is that his probabilities failed spectacularly in two very similar neighboring states. This points to a flaw in his model that emphasizes state fundamentals over polling.

That's possible, but we have a very small sample.  It's also possible that MT as a 1-in-3 event that Silver acknowledged was possible, and ND was also a 1-in-12 he accounted for.

[7,052,770]

Seriously, Nate Silver predictions for senate aren't extremely inacurrate. Nunn isn't going to have more chance to be electec than Hagan.

He wasn't that accurate in 2012 either.

Are we really going to do this again? Silver doesn't make predictions. He can't ever be "right" or "wrong." There is not nearly enough data out there to analyze his probabilities.

If candidates he gave a 5% chance to win never won, his model would be bad. Around once every 20 times, the 5% chancer should win. We need a larger sample size to know for sure, but since Heitkamp's the only senator to ever pull off the 5% upset by his numbers, and there have probably been around 20 races that he gave around a 95% on, it would be expected that one of them would have won.