538 Model Debut: 64% Chance of Republican Majority; R+7 Most Likely

[JRP1994]

Guys, if you add up the individual probabilities, the model is projecting R+9.

[JRP1994]

Guys, if you add up the individual probabilities, the model is projecting R+9.

I think you missed the fact that GA, KY, and KS are currently Republican held seats.

No, because WV and MT aren't even on the list. What he/she meant was that the model shows Republicans having a greater than 50% chance in 9 states: WV, SD, MT, AR, AK, LA, NC, CO, and IA. Gauss is correct in pointing out that it doesn't actually mean the model is predicting 9 seats though.

Yes, exactly what I meant.

Guys, if you add up the individual probabilities, the model is projecting R+9.

'
But events aren't independent here.  The races tend to go together.......

Thus, you can't just add up expected values like that.

Good point. But isn't that how Silver "correctly projected the outcome" in all 50 states, by adding the expected value? The model showed Obama winning something like 318.5 electoral votes, but 332 if you added up ever state in which he had a >50% chance.

[JRP1994]

Meh. Wang is better anyway.

Wang uses polling only and then applies way too much certainty in his models. If he's right, he looks smart and can bleat on about how Nate Silver adds too many bells and whistles, but when he's wrong, it can be spectacularly bad. For instance, in 2010, his model was off by 6.5x his stated standard error, which would occur by chance something like one in a billion times. Nate is much more realistic about the amount of uncertainty in prediction models.

His current prediction for KS-Sen, Orman as 80% to win, is clearly ludicrous. A single poll of a then-purely hypothetical race is enough to make such a confident prediction on? This is exactly the sort of case where non-poll factors like "state fundamentals" really are clearly needed.

PREACH IT.