Each candidate needs at least 270 electoral votes to win outright, and a tie at 269 that throws it into the House isn't impossible.
Obama | McCain
Votes Probability | Votes Probability
230+ 0.9882 | 230+ 0.4442
235+ 0.9841 | 235+ 0.3829
240+ 0.9789 | 240+ 0.3253
245+ 0.9723 | 245+ 0.2724
250+ 0.9639 | 250+ 0.2249
255+ 0.9532 | 255+ 0.1834
260+ 0.9398 | 260+ 0.1479
265+ 0.9230 | 265+ 0.1180
266+ 0.9192 | 266+ 0.1126
267+ 0.9152 | 267+ 0.1075
268+ 0.9111 | 268+ 0.1026
269+ 0.9066 | 269+ 0.0979
270+ 0.9021 | 270+ 0.0934
271+ 0.8974 | 271+ 0.0889
272+ 0.8925 | 272+ 0.0848
273+ 0.8874 | 273+ 0.0808
274+ 0.8820 | 274+ 0.0770
275+ 0.8764 | 275+ 0.0733
280+ 0.8454 | 280+ 0.0573
285+ 0.8087 | 285+ 0.0444
290+ 0.7659 | 290+ 0.0343
295+ 0.7174 | 295+ 0.0262
300+ 0.6635 | 300+ 0.0199
305+ 0.6050 | 305+ 0.0149
310+ 0.5432 | 310+ 0.0111
315+ 0.4797 | 315+ 0.0081
320+ 0.4157 | 320+ 0.0058
325+ 0.3533 | 325+ 0.0041
330+ 0.2940 | 330+ 0.0028
335+ 0.2394 | 335+ 0.0019
340+ 0.1902 | 340+ 0.0012
345+ 0.1475 | 345+ 0.0008
350+ 0.1115 | 350+ 0.0005
355+ 0.0822 | 355+ 0.0003
360+ 0.0592 | 360+ 0.0002
365+ 0.0414 | 365+ 0.0001
370+ 0.0285 | 370+ 0.0001
375+ 0.0192 | 375+ 0.0000
380+ 0.0126 | 380+ 0.0000
Posts
2 comments:
Good work for doing this - I remember our discussion after dinner a few weeks ago!
That non-independence thing is a real problem though... Could you (in principle) use polling timeseries from past campaigns to estimate very roughly the structure of the dependencies. Of course there's no saying it'd be the same this time around - the people and the issues have both changed - but it might be better than nothing...
Adrian
Sure, or any other data source you wanted. What you do with that data is another, harder question. I agree that the independence assumption is problematic, but relaxing it would involve putting together a model that would involve a lot more decision making.
The main problem is that there are correlations not just from state to state, but also from day to day. You could model the state distributions as changing according to a multivariate normal random walk, but a problem with that is that the way the distributions change is probably itself changing over time. You could try to model that based on past elections, but maybe this election's different. Etc.
Not that all that wouldn't be a lot of fun, but unfortunately I'm not sure I have time for it at the moment.
Post a Comment