HN Explorer

Nate Silver vs. The Betting Markets

22 points by davidmathers ↗ HN
For the record, on the night before the election:

(Obama is first number, McCain second)

fivethirtyeight.com:

  OH: 86% - 14%
  FL: 67% - 33%
  NC: 59% - 41%
  MO: 47% - 53%
  IN: 36% - 64%
  ND: 34% - 66%
  MT: 23% - 77%
intrade.com:

  OH: 75% - 24%
  FL: 77% - 25%
  NC: 65% - 43%
  MO: 52% - 50%
  IN: 39% - 67%
  ND: 27% - 76%
  MT: 33% - 78%
betfair.com:

  OH: 80% - 18%
  FL: 78% - 24%
  NC: 61% - 40%
  MO: 53% - 51%
  IN: 41% - 65%
  ND: 31% - 73%
  MT: 30% - 72%
Final Scores:

Nate Silver (McCain gets MO): Obama 353, McCain 185

Betting Markets (Obama gets MO): Obama 364, McCain 174

Karl Rove (McCain gets MO and NC): Obama 338, McCain 200

pollster.com (McCain MO, NC, FL): Obama 311, McCain 227

14 comments

[ 4.7 ms ] story [ 41.7 ms ] thread
jyothi
I think the bradley effect would happen again and McCain might get a surprise win.

It is indeed astonishing that no over is McCain a favorite to bet on.

robg
Conversely, there's the cellphone effect:

http://www.fivethirtyeight.com/2008/11/cellphone-effect-cont...

dejb
True. Both effects might true. However the 2 don't cancel each other out probabilitisticly. Given that McCain is so far behind the mere possibility of a Bradley effect favours him. Let's say both effects are 50% likely to exist and would sway votes by enough to tip the election. Then the results would be

25% Bradley + No Cellphone = McCain wins 50% Bradley + Cellphone = Obama wins 25% No Bradley + Callphone = Obama wins

So in this example McCain now has a 25% chance. Of course this is an exagerated example but hopefully you get the point.

gamble
The Bradley effect is a myth. Like Truman v Dewey and the exit poll controversies in 2000 and 2004, the Bradley effect presupposes that the polling was accurate. In each case, it's apparent from the historical evidence that unreliable polling was the culprit. Just look at how much variation there is in the polls this election - the most heavily polled yet. A gubernatorial campaign twenty-five years ago can't even compare.
robg
The likelihood numbers cannot be tested empirically. The experiment only gets run once. In other words, there is no error term for the likelihood numbers.

I'll be looking at Electoral Votes and the popular margin. There the difference between pollster and 538 is the most interesting to me. The former is a simple average of all polls while the latter uses weighted averages. Individual pollsters might get closer but within their margins of error.

jedc
But the markets (and fivethirtyeight projections) exist for every state. Together they're a decent dataset, though still a bit light.

I, too, am very interested in comparing betting (prediction) markets versus poll-based probabilistic forecasts. I wrote more about this here:

http://blog.mercury-rac.com/2008/11/03/election-tuesday-what...

patrickg-zill
Nate Silver is conducting a propaganda campaign on behalf of Obama. He is not even close to being an honest pollster.

See http://seanmalstrom.wordpress.com/ (long post, search for "silver" in the text)

jedc
A- He's not polling, he aggregates publicized polls B- He doesn't claim that he's not biased in his commentary. C- His numbers come from polls alone. While he could potentially skew through manipulating an algorithm, his results to date have shown that he's done much better than the pundits.
jonknee
He has been open with his support for Obama. But he's a numbers guy, I don't see where his math is biased. Likewise, he's a Cubs fan but his baseball research isn't slanted towards the Cubbies.

Your link just says because Nate Silver talks about Obama in his blog the site is a propaganda tool. People don't visit for the blog, they visit for the numbers and research. Personally I think it would be the other way around, if he were also a popular Obama blogger but did FiveThirtyEight anonymously that could be considered sinister (it still wouldn't be unless the model was proven wrong, but there would be a perception of deception).

I've seen him a couple times on HDNET and he's extremely clever. I'll be watching him tonight instead of the hacks they put up on the major networks.

gamble
I find it informative to compare Silver's projections with the Real Clear Politics averages. RCP is a group project, but it's pretty clear that their editorial views skew heavily toward the conservative end of the spectrum. Assuming either is biased, I figure that Obama can't be doing any worse than RCP is projecting or much better than FiveThirtyEight.
jfarmer
He's not a pollster, he's a guy who analyzes other people's polls.

And how is he being dishonest? It's not like he's hiding his support for Obama.

The numbers are right or they aren't. We'll see how close the election really is and how his predictions match up with the pollsters.

jpavlik
Silver also said the Tampa Bay Rays would win 88 games, and people thought he was nuts. They won 97.
ca98am79
So far the actual results are:

OH: Obama

FL: Obama

NC: ?

MO: ?

IN: Obama

ND: McCain

MT: McCain

All were wrong with IN.