November 3, 2004 12:25 PM PST

Who won? Online bookies

The results of the 2004 presidential election finally are in, and the winners are: the betting Web sites that correctly predicted what would happen on Election Day.

Unlike the amateur prognosticators who relied on their own computer models and statistical analyses of polls and frequently got it wrong, betting exchanges were eerily accurate in heralding George W. Bush as the winner.

Betfair correctly predicted that George W. Bush would stay in office and had offered odds of 2-to-1 last week for Sen. John Kerry to win.

About $4.2 million had been wagered on Bush, with Kerry drawing only $1.2 million in bets. Betfair also claims to have accurately picked the victories in the elections of California Gov. Arnold Schwarzenegger and Australian Prime Minister John Howard.

The odds at Dublin-based Tradesports.com were similarly accurate, giving Bush a 58 percent chance to win and Kerry a 42 percent chance as of late October.

One explanation for this stellar performance is that it's easy enough to offer predictions when being wrong means only potential embarrassment for a blogger or online pundit. But when there's a financial penalty to errors, results tend to line up far more closely with reality.

The Iowa Electronic Markets also correctly predicted a Bush win, with its winner-take-all market showing Bush with a slight but consistent edge over the last few months.

Economists believe that bettors have strong incentives to collect all the information available to them to make accurate predictions, while people responding to pollsters may lie. That may be what plagued well-known pollster John Zogby, who called the election for Kerry and incorrectly said on Monday that Kerry was ahead in Florida.

While Internet-based betting may be novel, political oddsmaking enjoys a venerable tradition. Paul Rhode and Koleman Strumpf, who teach economics at the University of North Carolina at Chapel Hill, said betting markets predicted the popular-vote winner in every presidential race but one between 1884 and 1940. (In that 1916 race, gamblers were offering even odds between Woodrow Wilson and Charles Hughes.)

"Wagering on presidential elections has a long tradition in the United States, with large and often well-organized markets operating for over three-quarters of a century before the Second World War," Rhode and Strumpf wrote in a recent paper. "The resulting betting odds proved remarkably prescient and almost always correctly predicted election outcomes well in advance, despite the absence of scientific polls."

Bloggers were all over the map
Sam Wang, a professor of molecular biology at Princeton University, was working in his office late Monday on a computer model that predicted Kerry winning a handsome victory on Election Day.

The Democratic senator would easily sweep Ohio and Florida, Wang predicted on his Web site, based on a compilation of statistical data and estimates of turnout and who undecided voters would choose.

Wang and many other amateur forecasters weren't exactly correct, of course: Bush delivered a victory speech on Wednesday afternoon after Kerry phoned him earlier in the day to concede the race.

"The polling data were fine, but my follow-up assumptions about net undecideds and turnout were wrong," Wang acknowledged Wednesday. But he defended his statistical approach, which essentially averaged state polls, as "still the best way to look at the data."

Even after Kerry's concession call, Electoral-Vote.com continued to list Iowa and Ohio as "exactly tied."

Behind that site is Andrew Tanenbaum, a computer scientist and Kerry supporter probably best known for writing MINIX, a Unix-derived operating system that was the precursor to Linux. Tanenbaum's final predictions showed Kerry claiming 306 electoral votes and Bush with 218 thanks to projected wins in Ohio, Florida, Pennsylvania and Michigan.

On Wednesday, Tanenbaum acknowledged Ohio seemed to be Bush territory--but speculated about possible fraud. "One thing that is very strange is how much the exit polls differed from the final results, especially in Ohio," he wrote. "Remember that Ohio uses Diebold voting machines in many areas. These machines have no paper trail."

A rival site at ElectionProjection.com correctly predicted that Florida and Ohio would tilt toward Bush, giving the president an estimated total of 286 electoral votes to Kerry's 252. But Scott Elliott, a self-described conservative Republican who runs the site, incorrectly predicted that Wisconsin would side with the president.

By Wednesday morning, an ebullient Elliott was beyond quibbling over numbers. "George W. Bush has won the presidential election of 2004," he wrote. "I feel an enormous sense of relief that America has confirmed in no uncertain terms that the right man will be in the White House for four more years."

Also missing in his prediction was Larry Sabato, who runs the University of Virginia's Center for Politics. His online map predicted an exact tie between Bush and Kerry.

Not one of the pundits surveyed by the Washington Post for its "Crystal Ball" prediction roundup got it right. Conservative writer William Kristol was probably the most inaccurate, predicting that Bush would win 348 electoral votes. (As of midday Wednesday, the near-final count stood at 274 electoral votes for Bush and 242 for Kerry.)

1 comment

Join the conversation!
Add your comment
Andrew Tanenbaum is an idiot
He fraudulently tried to manipulate the election by predicting Kerry with known bad data... then he cops this "holier than thou" attitude about election irregularities, WHEN HES NOT EVEN IN THE COUNTRY TO WITNESS IT. Just parroting whatever comes out of the DNC echo-chamber I guess.

Its pathetic for someone who worked as such a highly regarded programmer.
Posted by (1 comment )
Reply Link Flag
 

Join the conversation

Add your comment

The posting of advertisements, profanity, or personal attacks is prohibited. Click here to review our Terms of Use.

What's Hot

Discussions

Shared

RSS Feeds

Add headlines from CNET News to your homepage or feedreader.