[Satadru]

How about covariance between races. Are you assuming that each race is independent? For example, there should be big correlation between SD and MN.

[crystalk]

No, no use getting too technical, since we do not really care whether the prob of (R 50 or more) is 87% or 89%; we are trying to get a handle on whether it is 20%, or 80%!

Personally, I think S Dakota is in the bag for R, remember I do not care about the MARGIN, only the chance of winning at all, by however thin a margin.

Obviously MN is much closer and less certain, but I think Coleman is well ahead of Wellstone in chance of winning. (Say 70-30)

[Doctor Stochastic]

I gave that some thought myself. Except for 1994, it seems that in most races, politics is local. National trends are weak. Maybe I'll write a program of my own (including covariances.)

I'll start with the assumption that the poll results are binomial with the error at a 5% level and then fit a multivariate normal through the poll numbers. (Easier to simulate.) Then fudge some covariance results; probably by taking several polls over time and seeing if they move up or down together. The I can draw samples and see what happens.

This may take some time as I haven't looked for poll summaries. I could also model the drift in polls as Brownian motion and use this to add uncertainty to my results.

[Jim Noble]

>>How about covariance between races<<

Not just regionally. The probability of a Johnson win (SD) will rise as the probability of the GOP taking over rises.