FiveThirtyEight releases more detailed state-by-state model

[JRP1994]

http://fivethirtyeight.com/interactives/senate-forecast/

[dmmidmi]

It's interesting to see how close many of their Projected Vote Share estimates are to simple polling averages: the difference in Louisiana is 0.1%, in Michigan it's 0.3%, in North Carolina it is also 0.1%.

If this is the case, what value does State Fundamentals really provide?

[Miles]

He's projecting Haugh/third parties to get 8.5% (!) in NC. 'Seems too high, IMO.

I'd say at least 3%, but I'd be surprised if it was more than 5%.

[King]

No True Atlasian gives to inks about 538. It's statistical analysis for casual observers who can't do it themselves. To be a True Atlasian, you should be able to open the toplines of any poll and interpret it for yourself.

[SteveRogers]

It's interesting to see how close many of their Projected Vote Share estimates are to simple polling averages: the difference in Louisiana is 0.1%, in Michigan it's 0.3%, in North Carolina it is also 0.1%.

If this is the case, what value does State Fundamentals really provide?

It carries more weight in states like Alaska where reliable polling is sparse.

[Nichlemn]

No True Atlasian gives to inks about 538. It's statistical analysis for casual observers who can't do it themselves. To be a True Atlasian, you should be able to open the toplines of any poll and interpret it for yourself.

Pretty much all poll analysis in this forum consists of coming up with reasons to disregard polls with results you don't like while being accepting almost at face value any poll with a result you do like. Also, not really caring about the polling average.

[Nichlemn]

It's interesting to see how close many of their Projected Vote Share estimates are to simple polling averages: the difference in Louisiana is 0.1%, in Michigan it's 0.3%, in North Carolina it is also 0.1%.

If this is the case, what value does State Fundamentals really provide?

Their primary function is in races where there hasn't been much polling, and for when election day is far away.

[dmmidmi]

No True Atlasian gives to inks about 538. It's statistical analysis for casual observers who can't do it themselves. To be a True Atlasian, you should be able to open the toplines of any poll and interpret it for yourself.

Pretty much all poll analysis in this forum consists of coming up with reasons to disregard polls with results you don't like while being accepting almost at face value any poll with a result you do like. Also, not really caring about the polling average.

I'd love to know how many people that do this have taken so much as an entry-level research methods/statistics class.

[Vosem]

It's interesting to see how close many of their Projected Vote Share estimates are to simple polling averages: the difference in Louisiana is 0.1%, in Michigan it's 0.3%, in North Carolina it is also 0.1%.

If this is the case, what value does State Fundamentals really provide?

The more polling there is, and the closer to Election Day it gets, the less value the state fundamentals have. The less polling there is, and the further from Election Day we are, the more value the state fundamentals have.

[Princess Nyan Cat]

When are they ever going to post a governor's model?

[The Other Castro]

When are they ever going to post a governor's model?

The best we can hope for at this point is just a final projection post.

[Attorney General, LGC Speaker, and Former PPT Dwarven Dragon]

When are they ever going to post a governor's model?

Never. Didn't have one in 2008 and 2012, and I'd be surprised if they start up a 2014 one this late. They did actually try it in 2010, but even though it did well (in terms of predicting winners (missing only IL), margins are another story), they must have lost interest in doing it again:

http://elections.nytimes.com/2010/forecasts/governor

[Slander and/or Libel]

It's interesting to see how close many of their Projected Vote Share estimates are to simple polling averages: the difference in Louisiana is 0.1%, in Michigan it's 0.3%, in North Carolina it is also 0.1%.

If this is the case, what value does State Fundamentals really provide?

Their primary function is in races where there hasn't been much polling, and for when election day is far away.

Wouldn't the value of those farther out projections then be judged by how accurate they tend to be? For some reason we only ever get analyses of how well the final projections modeled the races' outcomes, but shouldn't we be able to, say, back that all up by a couple of months and see if Nate Silver's early September projections are any better than Sam Wang's?

[IceSpear]

After falling for quite a while, Democrats had a mini rebound today. Up to 36% on 538 and 37% on the Upshot.

[muon2]

After falling for quite a while, Democrats had a mini rebound today. Up to 36% on 538 and 37% on the Upshot.

Given the number and variability of the inputs, any changes of less than 10% are probably within the margin of error.