Presidential Prognostications

As the presidential race enters the home stretch, my most recent Wall Street Journal column assesses the likely outcomes. My conclusion probably comes as no surprise: “Barack Obama is the hot favorite to win.”

I begin by noting that prediction markets currently rate Obama an 85 percent chance to win. But I think that these may slightly understate the probability of an Obama victory:

My reasoning is based on a phenomenon that behavioral economists call the “favorite-longshot bias.” In essence, people tend to overbet unlikely outcomes, and so the odds often overstate the chances of a longshot (like McCain) winning, and conversely understate the chances of the favorite.

Most of the previous evidence of the favorite-longshot bias came from analyzing horse-race betting markets. But recently, Andrew Leigh, Eric Zitzewitz, and I have turned to assessing whether a similar bias affects political prediction markets. Our findings — while preliminary — show that it is exceptionally rare that political underdogs win. When political races are close, this bias isn’t so important. But in lop-sided contests, prediction markets really tend to overstate the support for underdogs.

In fact, our best estimates suggest that when election-eve prediction markets suggest a 15 percent chance of victory — as they presently do for Sen. McCain — that the true probability may be as low as 4 percent. Thus, Sen. Obama is not quite a sure thing — but he’s close.

How does this assessment compare with the polls?

To my eye, they tell a similar story, and my adjusted assessment of the prediction markets — suggesting that Sen. Obama is a 96 percent chance to win the election — matches that of FiveThirtyEight.com’s Nate Silver, the current darling of the poll crunchers.

Our approaches differ. My assessment is based on a reading of the prediction markets, taking account of historical biases against the favorite. Mr Silver’s assessment is based on crunching data from countless state and national polls. But viewing the data through either lens, it is hard to see a path to a McCain victory.

My full column is available here. McCain’s pollster disagrees here. And Stan Greenberg responds here.

Related: Foreign Policy has a useful summary of prediction markets; BetFair has launched an interesting site, BetFair Predicts.


Karen

I've noticed this phenomenon in the office sweeps for the Melbourne Cup (horse race held on the first Tuesday in November that everyone stops work to watch). Every firm I've worked in with more than 20 employees has some form of auction system to allocate horses to office punters (if your horse wins, you get 50% of the pool, second place =30% of the pool, etc). Since most people I've worked on wouldn't know which way round to sit on a horse, the only information we have is the odds published in the papers. Now you'd think that a horse that has been given 3:1 odds to win would have a final price of ten times that of a horse given 30:1 odds, because there is a ten times greater chance that you'll win the office sweep with that horse.

Not so. In my experience, the favourite horse might attract a final price three to five times that of some of the seriously long-shot horses. In other words it supports your findings exactly - the long shots were way overvalued relative to the favourite. And these are economic research/consulting organisations; ie at least 2/3 of the office sweep participants were economists!

Read more...

a VERY confused reader

Is there any way, Justin, that you can summarize (in close-to-plain-English, your paper "Explaining the Favorite-Longshot Bias: Is it Risk-Love or Misperceptions?"? Thank you

Sean Samis

As of this morning; fivethirtyeight.com has Obama at a 97.2% favorite.

We shall soon see ...

The US Senate races are more interesting, and the California Proposition to protect gays from having in-laws ...

PaulK

I think there are other factors as well. In horse racing, few bet on a horse because they like the horse or owner or some such. In political betting, there are a fair few who will place a bet because they want that person to win, no matter the odds. We saw this with intrade where someone was placing large bets to skew the numbers - they were not basing it on odds, but using bias to affect it.

You are talking about a bias where you pick an underdog simply because the payouts are great if it happens to win. I know some people in any betting will do that: place a large bet on a favorite, but a smaller bet on a longshot just because the pay out would be so good.

NJR

Have you seen a limit to how high the price in a betting market will be driven up? Paying $96+fee with a chance to win $100 is neither fun nor financially rewarding.

DJH

For Kurt (#2), the cell-phone problem you mention, is not as big as has been claimed.

First, some polls are conducted by dialing phone numbers randomly. These polls will naturally include both landline and cell phones, so there will be no such bias in them.

Second, even in polls conducted from phone listings -- which all but exclude cell phone numbers -- there is still not that much difference. The Pew Center, for example, has shown that the results are not that far off (http://people-press.org/report/276/).

Third, some pollsters are developing lists of cell phone users using various sources, and can use these to calibrate the results if needed.

James

Given that the market must be right and that as we have analyzed the effect of one vote before, I guess that means I can stay home. Thank you for saving me the gas.

DK1

Maybe some folks are putting money on McCain as a hedge. If your preferred outcome doesn't materialize, at least you'll make some money.

frankenduf

not sure this is true in football- i would say that the tendency is for people to overbet favorites- this is what happened last year when the bookies were taking a bath on the patriots- they finally threw caution out the window and raised the patriot spreads to over 20 (!) points, but by then, they started to play much closer games- and certainly people overbet the favorites for the super bowl- it is a law of betting that if the favorite wins the SB, the bookies lose, but if the underdog covers, the bookies celebrate christmas while the bettors eat snow

Kurt

I've also heard that there's a cell-only-voter bias that skews the polls slightly towards the right. People who don't have a home phone and only use their cells are far less likely to be reached by pollsters. These people tend to be young and tech-savvy, a demographic that is generally Democratic. In this case, it would also contribute to an overstatement of McCain's favor in the polls.

Joe

Lots of evidence for the overestimation of unlikely events. Insurance, loterries, etc all hinge on overvaluing going from zero to slim probablities of good things and slim to zero probability of bad things.

XPolygamistWife

I hope it's true because the last man I want in the White House is John McCain who SPREADS AMERICA'S WEALTH to FLDS polygamists in Colorado City, Arizona who practice tyranny over women and receive 25-30 million in taxpayer handouts.

No thanks, McShame.

BANKING ON HEAVEN . COM

Dan M.

I think something else that differentiates sport betting from election betting is the size of the sample being betted on. In sports, you are betting that a relatively small group of people (teams or individuals) will or will not do something. This smaller group makes it inherently difficult to predict an outcome. This is not so in an election, especially a national one. I think this is why political upsets are so rare.

When the group becomes larger (especially an electorate), it becomes much easier to predict an outcome. The size of the group itself precludes any shocks that may upset a poll prediction. In this election there are several million people voting. It's hard to imagine them swaying too much from the sample polls, unless of course the poll itself is flawed. One never knows...