At the moment, I gotta say, I'm dubious about trusting Nate. Here's why: his calculations right now are way, way off those of his election prediction peers.
Nate Silver, fivethirtyeight.com, presidential prediction, polls-only forecast:
The Upshot Presidential Prediction, NY Times:
Princeton Election Consortium (Sam Wang):
Electoral-Vote.com (computer scientist Andrew Tanenbaum and his colleague-in-statistics Christopher Bates):
These analysts/sites all do election analysis based on polling. The methodology is obviously different to get such wildly different results. I will have to read the fine print, but Nate Silver is very much the outlier here.
Oh, wait: here's the fivethirtyeight.com polls-plus forecast.
Okay, that's the model that includes historical data and the state of the economy.
Be sure to read Silver's methodology and check out his other "polls plus" model, which includes economic indicators.
ReplyDeleteAlso, noting that Silver's predictions are different from others is more an indication of how different his model is and how similar others' models are.
Lastly, if somebody predicts a 90% chance of something and it doesn't happen, the prediction isn't inaccurate. Conclusion: Always worry. :-)
I've got the polls plus model right at the bottom, but I ddi add it maybe a half-hour after the rest of the post went up. Yeah, I know his methodology is different!
ReplyDeleteHis explanation: http://fivethirtyeight.com/features/election-update-why-our-model-is-bullish-on-trump-for-now/
ReplyDeleteThanks, Dave!
ReplyDeleteWhile we're at it, here is a clickable version of the link. No, I don't know why Blogger is too stupid to publish clickable links.
ReplyDeleteI have an Oakland mathematics professor friend who is running a presidential polling prediction blog that might interest you: http://abovenota.blogspot.com/
ReplyDeleteAnd that was quite an opera we saw yesterday at the train station.
Oh, yes, it was.
ReplyDeleteHow well did Nate Silver perform in the primaries? I found Benchmark Politics to be pretty accurate in their predictions (based largely on demographics). They are currently predicting Clinton to win 347 EVs.
ReplyDeletehttp://www.270towin.com/maps/benchmark-politics-electoral-map
Not sure - I wasn't reading him during the primaries.
ReplyDelete538's predictions this week are much more positive for Clinton.