Dear Kossacks,
Here is an updated version of my meta-analysis of state polls. It's a calculation that takes advantage of all the state polls we read in small dribbles, reduced to a single confidence band. As of today it predicts a very close race, but perhaps surprisingly, it favors Kerry. The bottom line: Kerry 278 electoral votes (95% confidence band 247-317 EV), Bush 260 EV.
The calculation is done by taking the last three polls for each state, averaging them, and converting this to a single-state win probability. This is done for up to 21 states in which the outcome is in reasonable question. Then
every possible combination is tabulated, and the most likely range of outcomes worked out.
As of today, the surprise is that the median outcome is Kerry 278 EV, Bush 260 EV. This is surprising because as we all know, Kerry lags in current national polls. This discrepancy may be due to the fact that, as Ruy Texeira has pointed out, Kerry is currently leading in battleground states.
Analysis of state-level polls is prone to problems such as small sample size and survey flaws. However, the use of multiple polls reduces skew contributed by any single poll. For those of you who have opinions about polling biases, there is a calculation on the site that shows the effect of systematic bias in either direction.
Also note this history graph (updated less often, so it ends in late August). It shows very clearly the effect of major political events such as the Fahrenheit 9/11 release, the addition of Edwards to the ticket, and the Democratic convention.
I'll be updating this regularly.
Regards,
Sam Wang