Wednesday, July 30, 2003

The Pentagon’s terrorism death pool is gone, but there are still people all over the Internet -- most but not all of them conservatives -- who believe that it was a fine idea and that its opponents are hypersensitive and closed-minded. At site after site-- here, here, here, here, here -- we’re told that betting pools predict outcomes with amazing accuracy.

But is it possible that the astonishing predictive ability of widely scattered non-experts who place online bets in their Jockey shorts is one of those facts that’s just too good to check?

One of the great engines of market-based prediction, say the defenders of the Pentagon plan, is the Iowa Electronic Markets. IEM’s bettors bet on election outcomes, among other things, and often predict those outcomes better than experts, we’re told.

Well, not quite.

IEM keeps data online for some bygone betting pools. Here’s the main page for the 2000 presidential popular-vote pool. We all know how that vote turned out: a Gore-Bush dead heat. IEM allowed bets on Democrats, Republicans, and the Reform Party (Perot’s old party, with Buchanan as the nominee), and liquidated contracts after the election at 49.9 cents for the Democrats, 49.7 cents for the Republicans, and .4 cents for Reform (ignoring Nader and all other candidates, this was the three parties’ share of their joint total; it adds up to $1.00, i.e., 100%). Now, most polls leading up to the election were predicting a Bush victory. If the IEM bettors had predicted a dead heat, or a slight Gore victory, that would have shown real predictive power. But here’s what contracts were going for on election eve, November 6, 2000:

Republican (Bush): 52 cents

Democrat (Gore): 47.5 cents

Reform (Buchanan): 1.7 cents


Did IEM's bettors do a better job than professional pollsters? Hardly -- see Polling Report's summary of election eve polls.

IEM also ran a presidential winner-take-all pool -- you pick the winner, you get a buck per successful bet. On election eve, that one looked like this:

Bush: 71.1 cents

Gore: 29.1 cents

Buchanan: 0 cents


This was the right answer, but by an eyelash. A few tweaks here and there -- no butterfly ballots, for instance -- and Gore could have declared victory.

The 2002 congressional control pool was just dreadfully inaccurate. Election day was November 5, 2002; at the close of the day on November 4, here were the numbers:

Republican House, Republican Senate: 28.3 cents

Republican House, non-Republican (Democrat + Independent) Senate: 57 cents

Non-Republican House, non-Republican Senate: 10.3 cents

Non-Republican House, Republican Senate: 1.7 cents


So why do these betting pools have such a sterling reputation for accuracy?

No comments: