When people talk about well pollsters or economic models or information markets predict the election, what is the best statistical way to do this? For instance, the Weekly Reader kids poll has been right 12 out of 13 times (http://www.weeklyreader.com/election_vote.asp). Would you simple say that 6 out of 12, to make it easier, would be expected? Or would you toss out the “obvious” like Reagan/Mondale and Clinton/Dole wins? Same with markets: how do you know when to declare them accurate? The day off, day before, week before? Only who close they are if the race is truly up for grabs, or in all cases?
Just thinking outloud, but I hear so much about this or that method working out, that it seems silly after awhile, especially if kids can hit it on the nose 12:13 times.
Count the number of correct predictions they’ve made and divide it by the total number of predictions they’ve made. That’s the percentage of the time that it’s right.
With respect to percentage of the vote, there are a number of techniques one could employ: correlation coefficients, mean squared error, regression analysis. The best choice would depend on exactly what question you were trying to answer.
As far as forecasting the outcome is concerned, there isn’t much more to it than what ultrafilter said: The poll has correctly forecast 12 of 13 races. Of course, keep this in context: Just by predicting a Republican victory every time, you could have gotten 8 out of 13. Any minimally effective polling technique which called the blowouts correctly and guessed at the remainder would have gotten 10 or 11 right. And at least three elections (1960, 1968, and 2000) were so close that they were beyond the forecast accuracy of any poll, so getting them right was purely a matter of luck.