Calculating the Odds of Who Will Be the Next President?

One of the sportsbook site I frequent let’s you “bet” on the outcome of the US Presidential Election.

I’ve been thinking about buying some McCain shares just because it seems like a good value.

I’m rooting for Obama to win, but the odds for buying his “Sportsbook Stocks” are not worth it. (No Value on that bet)

McCain on the other hand is a different story.

You purchase one McCain share for $22, and if he wins, it’s worth $100.
So according to them, McCain is roughly a 4-1 shot of winning.

Does this sound right to you guys? What would be a good site for trying to calculate the odds of the outcome?

I’m thinking of buying 10 shares.

If Obama wins, I lose $220, but at least the guy I’m rooting for is elected.
If McCain wins, my $220 is now turned into $1,000. This will help ease my pain of not having my candidate win. (a little)

The site owner is a well known (especially among baseball geeks) statistician and he takes a conservative apprroach to forecasting the election. He runs every scenario 1000 times to come up with his odds. He currently says that Obama is a 9-1 favorite to win the election. That means that the (likely) value of your Obama bet is probably quite good, and the McCain odds are terrible. does extensive statistical analysis on all of the available polls, and calculates the chance of each candidate winning (among other things) from that. Their methods seem pretty solid, and they’re currently giving McCain only a 9.5% chance of winning. Assuming that’s correct, then betting at 4.5 to 1 on McCain is a losing proposition.

If you want insurance, go for it (insurance is of course always a losing proposition, on average), but it’s pretty expensive insurance.

The percentages on the site aren’t actually odds; they are the results of his simulation. Not the same thing, and how the numbers relate to the real world is very vague, since any simulation is 1) based on the simulator’s assumptions and 2) can’t really take into account every possible occurance. If tomorrow there’s a terrorist attack, everything changes.

Ultimately, the simulation here only says, “The candidate who is leading in the polls has the best chance of winning.” That’s pretty trivial. To actually relate it to odds (which, in the sports book case, are determined by the relative amount of money bet on each candidate) is a mug’s game.

I ran similar simulations using data just for kicks a few weeks back. A big flaw in these methods is that they treat the states independently, which is a horrible assumption. The battleground state outcomes are going to be strongly correlated (e.g., if Obama loses PA, he’s much less likely to win OH.)

The questions that actually drive the victory probability are more like: How effective will the final three weeks of attack ads be? What will the state of the economy be at the end of October? … In other words, the polls are telling you “If the election happened now, who would win?” but that’s the wrong question for your wager. The correct question is, “If the election happened three weeks from now, who would win?”

This is true for most red/blue state analyses, but Nate’s analysis (538) has correlated polls to the election results, so his odds are answering the question of “historically, how likely is it that a candidate who is N points ahead in the polls by today’s date carries the election?” He implemented that change in late August and discusses his methods in this post. By correlating polling numbers and demographics to the electoral outcome, he can ignore things like attack ads and assume that their impact appears in the polls. There are three things that make this year’s election unusual: 1) First appearance of an African-American on the ballot, 2) Hi Opal!, and 3) The pending (current?) financial meltdown.

Nate’s likelihood numbers account for the first by using the African-American population in each state as one of twenty variables that correlate scores between states. His model appears to ignore the meltdown except where its impact can be seen in polls.

Thanks for the info. I did not realize that that site did such analysis. I’ll have to take a look at his methods…

The results of the simulation do give the probability, which gives more information than just saying which candidate is more likely to win-- It also tells you how much more likely that candidate is to win. And what the OP wants to know is the probability: He already knows what odds the bookies are offering, and whether it’s a good bet or not depends on how the bookies’ odds compare to the true probability.

The sportsbook’s odds are not based on the actual odds of the candidate winning. Remember, they’ll take either end of the bet. The odds are based on the odds necessary to have as close to a 50/50 split of bets. All the 4:1 odds tells you is that that’s how much of a margin it takes to get half the people to bet on McCain. I figure 4:1 is about right, I might take a flyer at those odds myself.

I thought they wanted the money bet to be 50/50, not the people betting to be 50/50. In other words, using your numbers, four people would be betting Obama for every one person that bets McCain.

We get the likelyhood of Obama winning via 538, and the odds of him winning via InTrade. Right?

You’re right of course, they want the money bet to be 50/50. It’s usually more common that the long shots have many small bets and the favorites have fewer large ones.

It’s not clear that InTrade’s numbers actually mean anything. See this recent thread for discussion and links.

They had the guy from five thirty eight on Colbert the other night. One of the things he says he does is try to “level” the polls. For example, a poll from Fox News will skew conservative and a poll from New York Times will skew liberal. This is not (just) because of the way the questions or the survey is framed (I’m sure they try to eliminate question bias) but because someone who is conservative, for example, might just hang up on someone polling from the NYT, and vice versa.

From the short bit I caught, it sounds like he’s really trying not to include bias, to the degree that is possible. Of course there will be factors beyond his control, but it seems the most neutral site available.

He calls these sources of error “house effects” and explains how he gets rid of them in the linked post.

I’ve been following him since early June, and have now gotten into the habit of checking his site daily. I think he has perhaps the best understanding of what’s going on this season of anyone in America who is not on a campaign payroll. If you’ve got more questions about his methodology, this link will take you to all of his posts tagged “methodology”. Enjoy!

Last odds I saw from Vegas were
Obama -450
McCain +250