XKCD's Randall Munroe runs Monte Carlo on Intrade's per-state markets

Re:XKCD's Randall Munroe runs Monte Carlo on Intrade's per-s... by Eliezer on Saturday October 04, 2008 @01:23PM (#323564)

Conditional independence FAIL.

[ Parent | on HN ]

Re:XKCD's Randall Munroe runs Monte Carlo on Intrade's per-s... by randomwalker on Saturday October 04, 2008 @03:45PM (#323637)

Thank you, that's what I came here to say. All it takes is one gaffe from Obama to flip all the swing states in one shot.

I'm not sure the 90% number means anything at all. Perhaps it is an estimate of Obama's chances if the election were to be held NOW?

[ Parent | on HN ]

Re:XKCD's Randall Munroe runs Monte Carlo on Intrade's per-s... by jrockway on Saturday October 04, 2008 @05:35PM (#323670)

I'm not sure the 90% number means anything at all. Perhaps it is an estimate of Obama's chances if the election were to be held NOW?

This is exactly what the text at the bottom of the page says.

[ Parent | on HN ]

Re:XKCD's Randall Munroe runs Monte Carlo on Intrade's per-s... by robg on Saturday October 04, 2008 @05:42PM (#323675)

But it's not quite that dependent either. There are stages of dependencies. Some states hang together more tightly (Northeast, West Coast, and Deep South) than others (Midwest and Southeast). And gradual shifts can be detected through polling and pinned to real world events (conventions, crises, etc). The campaigns use game theory to move resources accordingly (see McCain now in Minnesota versus pulling out of Michigan). Obama has a very strong firewall.

That's where, I guess, 538.com has taught me so much. The polls are a constant flow of changing opinions. But if your models are sensitive and well-designed you can see the slow wave shifts developing. It's not predictive, but history helps too. For instance, no candidate has ever been this far up, with this little time left, and lost.

[ Parent | on HN ]

Re:XKCD's Randall Munroe runs Monte Carlo on Intrade's per-s... by mattmaroon on Saturday October 04, 2008 @05:35PM (#323671)

No. It means he's ahead in so many states, that he'd pretty much have to do something big (and it's going to take more than a little faux pas) to fuck it up, and there's about a 10% chance of that.

[ Parent | on HN ]

Re:XKCD's Randall Munroe runs Monte Carlo on Intrade's per-s... by neilc on Saturday October 04, 2008 @06:24PM (#323692)

"This snapshot approach doesn't really predict upcoming regional or national shifts"

So I don't think your interpretation is correct.

[ Parent | on HN ]

Re:XKCD's Randall Munroe runs Monte Carlo on Intrade's per-s... by mattmaroon on Saturday October 04, 2008 @11:43PM (#323825)

Well, I'm certainly not arguing that these odds are correct overall. I figure there's a 10% chance we'll find out that either candidate once murdered a hooker at a party of Ted Kennedy's.

I see now what is meant though by Eliezer's comment, I think I missed it at first. I'll have to think about the effect on the outcome though. Luckily I have lots of friends with math PhDs.

[ Parent | on HN ]

Re:XKCD's Randall Munroe runs Monte Carlo on Intrade's per-s... by mattmaroon on Saturday October 04, 2008 @05:41PM (#323674)

I'm curious how he modeled it. Do actual results follow a bell curve with prior polling as the median and some well-known standard deviation? If so, it would explain why a 10% lead in the polls is considered nearly insurmountable in a given state. I know that if someone is polling at 60-40 in terms of voters, their odds of winning the state are much higher. I've heard 90% thrown out, but don't have any hard data.

Perhaps a better Monte Carlo would use polling data and that.

[ Parent | on HN ]