The best way to make election predictions is to use polls, and as many as possible. While this may sound simple, in practice it is quite complicated. We have chosen to combine polls, taking three factors into account, one objective, two somewhat subjective. Clearly, polls with a larger sample space should be given more weight and we apply a weighting proportional to the sample size. Then there is the issue of age. Polls do get stale, so an aging formula is applied to deemphasize older polls. Finally, there is the issue of past performance. I look at how well each polling organization did in each state and apply a past performance factor. All three of the above factors are assigned individually to each poll when it is added to the list.

The prediction calculations use accepted statistical techniques combined with highly optimized algorithms, which in turn invoke special functions from mathematical physics (think Gauss, Chebyshev, and the incomplete gamma function for all you applied mathematicians out there). All of this is a nice way of saying that the results are accurate and that things run fast, very fast, but don't let that fool you -- there's a lot going on beneath the covers.

From the following list of polls for each state, choose as many as you would like to combine. You can get results for multiple states on each run and, within a state, you can use only the polls you wish to include in the calculations. Also, it is important to remember that each state is run independently of the other states, so that if you want to make a prediction for any state you can run with just polls from just that state or along with polls for other states.

One final note: Sometimes the sample size for a survey is unknown. In such cases a default, small size is assumed for calculational purposes and this is noted.