Been following his site for the past week...well here's his final prediction.
Overview: The basic calculation derives from polls only. Using statistical methods of meta-analysis, I use polls to calculate a starting point, referred to as "decided voters only." This result is an uncorrected snapshot of where the polls stand. In addition to this, I estimate the effects of last-minute undecided/uncommitted voter decisions and differential turnout. My estimates are supported by evidence, but are by no means certain. Results based on uncertain assumptions are clearly labeled. To let you try your own assumptions, a table of medians with different bias values is given here.
The basic decided voter result. The median of Kerry 252 EV, Bush 288 EV among decided voters was calculated from 152 polls taken in 23 battleground states, and stepping probabilistically through all possible outcomes. Most of these polls were completed between October 24 and 31. The EV estimate carries a large amount of uncertainty: the 95% confidence interval is ±40 EV. Thus, if only decided voters counted, the Kerry win probability would be 18%, or 4-1 in favor of Bush.
Decided voters only (% Kerry win probability): AR 2, AZ 1, CO 2, FL 24, HI 67, IA 38, ME 100, MI 90, MN 72, MO 1, NC 0, NJ 100, NV 8, NH 96, NM 4, OH 38, OR 99, PA 85, TN 1, VA 1, WA 100, WV 0, WI 25.
Rank order of states: States currently in play in the 20-80% probability range, indicating a near-tie, are in bold. Turnout and how the undecideds break will shift which states are at a near-tie, but the order, from most Democratic to most Republican, should stay about the same.
Decided voters only: Democratic <- ME/WA/NJ/OR/NH(95-100%) / MI / PA / MN / HI / IA / OH / WI / FL / NV / NM/AR/CO/TN/MO/AZ/WV/VA/NC(0-5%) -> Republican
With undecided voters assigned. Undecided voters typically end up voting against the incumbent. In previous presidential races this has given a 2.5 ± 2.0% advantage to the challenger. I currently estimate that 3.0% of voters are undecided. A 3-1 Kerry-Bush split gives a +1.5% net advantage to Kerry. This leads to a median EV estimate: counting undecided voters, Kerry 280 EV, Bush 258 EV, and a Kerry win probability of 71%, or 2.4-1 in Kerry's favor.
Turnout estimates and other corrections. The principal factor not measured by polls is turnout. Pollsters ask respondents questions to determine if they are likely to vote. However, this cannot capture efforts by voter turnout organizations. In addition, newly registered voters have no track record. Finally, telephone polls may not accurately sample the voting population. I estimate that these factors sum to an advantage in battleground states of 2 to 3% for Kerry. As I am sure you are all aware, this number cannot be known with certainty. With that caveat in mind, I use +2.5% as my turnout figure. Combined with the +1.5% undecided allocation this makes a +4.0% bias as plugged into the MATLAB script. This leads to my final prediction. Predicted electoral outcome (11/1/2004 noon EST): Kerry 323 EV, Bush 215 EV, Kerry win probability 99.93%. Based on the probabilites below, of the 23 states modeled, Kerry's expectation value of states is approximately 16 of them, for a total of 25 states plus the District of Columbia.
Prediction, undecideds assigned, plus turnout (% Kerry win probability): AR 48, AZ 10, CO 47, FL 90, HI 99, IA 96, ME 100, MI 100, MN 100, MO 34, NC 1, NJ 100, NV 72, NH 100, NM 60, OH 95, OR 100, PA 100, TN 8, VA 12, WA 100, WV 13, WI 91.
The popular vote. To estimate the popular vote I use two approaches: (a) one based on presidential preference polls and (b) one based on Bush's job approval numbers. In 16 national polls the medians (± SEM) are Bush 48.0 ± 0.4%, Kerry 47.0 ± 0.4%. Assuming 2.0% for Nader/other, the fraction of undeclared voters ("undecideds") is 3.0%. Assuming Cook's incumbent rule that undecideds split 3:1 for the challenger (2.25% and 0.75%), this gives a net of 1.5 ± 1.2% to Kerry. This predicts a national popular vote (not corrected for turnout) of Kerry 49.3 ± 0.9%, Bush 48.7 ± 0.9%, Nader/other 2%. The second measure uses job approval ratings. In ten polls taken since mid-October the median ± SEM is 49.0 ± 0.9%. Based on historical trends, this places an upper bound on Bush's share of the popular vote. Thus, both approaches indicate that Bush's popular vote share will be 49% or less.
I use the turnout factor to make a final estimate. National turnout should be less enhanced than battleground state turnout, so I assume that the margin will be increased by about 2.0%. Predicted popular outcome: Kerry 50.3%, Bush 47.7%, Nader/other 2%. National polls come from davidwissing.com, RealClearPolitics, and yougov.com. Job approval numbers come from pollingreport.com.
http://synapse.princeton.edu/~sam/pollcalc.html