Thursday, November 8, 2012

Nate Silver was 100% right

Picking a public fight with Nate Silver about his model demonstrated poor judgment by the right. Nate Silver’s model predicted 50 out of the 50 states. Realclearpolitics polling average got one state (Florida) wrong. In 2008 Nate Silver got 49/50 states right while the polling average got 48/50 states right.

Nate Silver over-predicted Obama’s national vote only marginally. Nate Silver predicted Obama would get 50.8%; currently Obama is getting 50.5%. Who knows, maybe when all ballots are counted Silver will be proven exactly right about this too. In the Senate he was only wrong by 0.5 seats (with a pro-Republican, not pro-Democrat bias).

When Nate Silver’s model was criticized, Paul Krugman become hysterical and wrote “This is really scary. It means that if these people triumph, science — or any kind of scholarship — will become impossible.” This is silly, obviously part of social science is discussing behavioral models and their assumptions. The model is then tested against reality.

Nate Silvers model did a fantastic job against reality in the 2008 election and the harder to predict 2012 election. The only time Silver’s model did not do a great job was the 2010 election. Silver predicted Democrats would keep the house in the summer, and only gave republicans a 30% chance of taking the house with a 60-seat margin, which they ultimately did.

By scientific standards Nate Silver has grown in stature in my book while his Republican critics have shrunk.   

The only significant pollster on the right I know of who predicted 49 out of the 50 states was American Enterprise Institute Henry Olsen. I recommend you read his article in full. The most prophetic passage is:

“Win or lose, we are in the twilight of the Age of Reagan. Romney’s efforts have almost recreated the Reagan coalition, but in today’s America that is no longer enough.”

5 comments:

  1. today I went online and found your blog really amazing, I think you have an amazing creativity I am happy to learn from you even can't be met in the real world.wallpapers

    ReplyDelete
  2. “Win or lose, we are in the twilight of the Age of Reagan. Romney’s efforts have almost recreated the Reagan coalition, but in today’s America that is no longer enough.”

    Actually, no. In 1984, Reagan got 64% of the white vote, 9% of the black vote, and 37% of the Hispanic vote.

    Romney did worse with all three group. However, it was the fall in white votes (to 59%) that cost Romney the election. Apparently, white turnout was depressed as well. Conversely, black and Hispanic turnout matched 2008.

    Note that in 1980, Reagan margins were lower with whites and Hispanics (but better with blacks). Simply doing as well as Reagan did in 1980 would not suffice today.

    The real irony is that Reagan pushed for the 1986 illegal alien Amnesty that did so much to create the Democratic friendly electorate that doomed Romney. At the time, sensible voices said that Amnesty would eventually hurt the Republicans very badly. They were right. Actually, Republicans started losing Hispanic votes as soon as Amnesty passed (in 1988).

    Why Republicans would be even contemplating repeating the disaster of 1986 is hard to grasp. A death wish perhaps?

    ReplyDelete
  3. I predicted Obama would win by applying the Franken/Coleman model where Obama wins every state without Voter ID laws and with Early Voting laws.

    ReplyDelete
  4. The triumph of Silver's model, and the failure of his critics illustrates the famous observation of statistician George Box.

    All models are wrong, but some models are useful.

    I'm a mathematician by training and to a degree, by trade. The right point to attack a model is in the assumptions going into it, before the crank is turned. Differing with its conclusion despite accepting its construction and inputs amounts to hysterical hand-waving. As Karl Rove illustrated.

    One thing - praising Silver's electoral model because it got 49 of 50 states (2008) or 50 of 50 states (2012) right over-states its performance. We all know that the real metric here is how it does in cases where the vote is close. A model that predicts Texas will go red, or California blue, doesn't really tell us anything we don't already know.

    Still, it was an impressive showing for Silver.

    ReplyDelete

Google Analytics Alternative