Some notes on future improvements to the calculation. David Kline points out that to calculate the single-state probabilities, for small (<30) numbers of polls, a t-distribution is more appropriate than a normal distribution. This distribution has longer tails and will give less certainty in the estimates. Going in the converse direction, in retrospect some states were more certain than the one-week snapshot indicated, and outlier polls tended to introduce occasional inaccuracies. A more sophisticated averaging procedure is needed, one that uses more polls but gives more weight to recent ones.