Welcome to DU! The truly grassroots left-of-center political community where regular people, not algorithms, drive the discussions and set the standards. Join the community: Create a free account Support DU (and get rid of ads!): Become a Star Member All Forums Issue Forums Culture Forums Alliance Forums Region Forums Support Forums Help & Search

cthulu2016

(10,960 posts)
Tue Oct 9, 2012, 04:47 PM Oct 2012

Nate Silver's Nowcast Projections (Baseball, Bacteria and Ballots)

The daily Nowcast projections are not predictions. (The November 6 Forecasts are.) They are computer modeled analysis of what the election outcome would be today given all the polls to date.

The method is classic SABERmetrics, the form of computer assisted baseball analysis that Nate used to do. (SABER is, I think, Society for American Baseball Research)

You take every statistical thing you know about a situation. For most things there will be too many variables to calculate a single right answer.

Say Ohio polls, with their margins of statistical certainty, suggest a current 17% chance of Romney winning. And Nevada shows a 15% chance of Romney winning. What are the odds of Romney winning both? That is easy to calculate. But then apply that to every possible combination of 50 states... well, there are 52 cards in a deck and IIRC, there are more possible deals in bridge than photons in the universe. (Maybe only in our galaxy... I don't remember exactly but it's a crazy huge number.)

Computers allow us to program all the data we have, all the probabilities, and have the computer run trial elections with everything randomized within it's probabilities. So we can run 10,000 trial elections on the computer and get a close approximation of the (impossible) calculation. This is sometimes called a brute force calculation.

And it works. You get something usefully close to the calculated answer. (If you flip a coin 100 times you'll get the idea it's roughly a 50-50 proposition, but may well not get exactly 50 heads and 50 tails. A coinflip is easy to calculate so we know it is 50-50. But the general idea applies to things too complex to calculate. If you could, for some reaon, not calculate the odds of a coin-flip you would do many trials to approximate the answer.)

(Bill James used to do this with baseball games. He would do things like run 10,000 seasons of Joe Dimaggio in his prime to see how often he would beat his own consecutive game hitting streak. Seldom, it turns out. That streak was a truly rare statistical event!)

(Another cool form of brute force calculation is using vats of bacteria to model "traveling salesman" problems with too many variables to calculate. I digress... but it's cool. You can probably google it.)

So if the election were held thousands of times with precisely the polling data we have today Obama would win 90% of the time, or whatever the Nowcast projection is from day to day.

But it is a projection of the interaction of current polling data, not a projection of the future. It is not a prediction. To be a prediction it would have to predict what the polls will be in the future.

Nate does do some November 6 predicting also, but I think he would agree that his predictions of what will happen in Novemeber are a very different creature from his projections of current polling. Those involve a lot of historical trend analysis and extimation and guesstiamtion. For insatnce the current Nowcast is 82.5% Obama but the forecast is only 75% Obama. Nate is predicting some lower polls in the future. But that is not super scientific... it is smart educated guesstimation. For instance, he has no way to really figure the odds of Isreal bombing Iran, or even predicting for sure how the polls would react to that.

I think the Forecasts are presented as a bit more objective than they plausibly can be, but the Nowcasts are very reliable analysis of what current polling means in the context of a real multi-varriable election.

2 replies = new reply since forum marked as read
Highlight: NoneDon't highlight anything 5 newestHighlight 5 most recent replies
Nate Silver's Nowcast Projections (Baseball, Bacteria and Ballots) (Original Post) cthulu2016 Oct 2012 OP
No, there's a Nov. 6 forecast, and a "Now" forecast frazzled Oct 2012 #1
Right you are. Thanks. OP corrected to focus on Nowcasts. cthulu2016 Oct 2012 #2

frazzled

(18,402 posts)
1. No, there's a Nov. 6 forecast, and a "Now" forecast
Tue Oct 9, 2012, 04:51 PM
Oct 2012

The former (Nov. 6 forecast) has Obama currently at a 71.4% chance of winning, with 297.2 electoral votes, and a slim 50.5 to 48.5 vote margin.

The forecast if the election were held today has Obama at a 76.4% chance of winning, with 297.0 electoral votes and a 50.4 to 48.4 margin.

(These forecasts are roughly the same at the moment.)

Latest Discussions»General Discussion»Nate Silver's Nowcast Pro...