Friday, September 12, 2008

How Obama Wins, Ctd.

Someone writes in:

Two points:

1) I can't figure out how to comment.

2) The individual probabilities aren't independent. If McCain wins Ohio and Florida, his probability of winning NC is more like 99% (or whatever). This is a serious methodological error. Fix it before someone figures out how to comment and lampoons you publicily.

Yeah, not sure I want to turn on comments or not. Not that I'm expecting anyone I don't directly know to actually read this thing, but even then, most comment sections of blogs kinda suck. Plus, I like to squash dissent...

As for the probabilistic assumptions... first, I totally understand what you're saying. But, I still think that states can be treated independently. First is the case of true toss-ups. There are several states that are, in my mind, legitimately 50/50 states, or so close that we can treat them as such: NH, NV, CO, & OH. I think you'd agree that, if a state is truly a toss-up, then what one does is a poor predictor of what another will do (i.e. if Obama wins Colorado, I can't think of an explanation for why he would or wouldn't win Ohio). In this case, it is fair to treat them as independent events.

Your concern comes with states that are more on the periphery, such as North Carolina. While my gut tells me that, yes, I can't really imagine Obama winning North Carolina but losing Florida, why should I trust my gut? In other words, I think the royal we have ranked various states in terms of their winnability for a given candidate, and made the assumption that once he wins state #27, he must also have won states #26 and up. I think that is a bad assumption. I think there are enough micro-conditions -- other races going on in that state, referendums on the ballot, cost of and commitment to advertising, variation in strength of get-out-the-vote operations -- that assumptions about macro-trends need to be taken with a huge grain of salt.

Put another way, I think the 'if state A is won, then state B will also be won' argument assumes that polls of states are all going to be wrong in a consistent way. In other words, I gave McCain a 90% chance of winning NC because he is leading in 9 out of every 10 polls, or tends to lead by 6 points, or whatever subconscious arithmetic I did to come up with 90%. I give Florida 60%, because McCain's lead there happens less often and is smaller when it does. This argument then assumes that, if Obama wins NC, he did it by making up 6 points (and/or pollsters were undersampling some key demographic and not adjusting accordingly, some October surprise occured, etc.), and thus those 6 points should be added to every state. Again, for the same micro-conditions listed above, I don't think this is a good way to go about it. Maybe I'm more of a federalist than I realized, but states really are autonomous units (how else to explain Utah?)

I readily admit that going with simple probabilities is not the most sophisticated way of doing it, but especially for the states that are true toss ups (and thus carry by far the most weight when doing the math), they are reasonable assumptions.