Forecasting

I occasionally give a lecture on political forecasting, about the different methods of predicting election outcomes. A key element of this talk is the uncertainty of forecasts: anybody who nails a congressional election or presidential vote percentage 6 months in advance is just lucky, and the farther into the future the event is, the bigger the range of uncertainty. In December 2003, the top three contenders -- in both polls and among pundits -- for the Democratic presidential nomination were Dean, Gephardt, and Clarke. Within a month, they had all crashed and burned.

I always make a point of mentioning this whenever I'm asked to forecast a result -- "here's what it looks like today, but things can easily change." At times, I'm accused of waffling, but it reflects the reality that there is no way to predict things like this: A GOP Congressman from Florida, Mark Foley, has suddenly resigned after sexually suggestive emails and instant messages to a Page were revealed.

Foley was expected to win in a landslide, and this race has now become an instant tossup. To make things more complicated, the ballots have already been printed, so even though the Republican Party can replace him, voters will still see his name when they go to the polls on November 7th.

When you're on a razor's edge -- I had thought that the Democrats would win 212-215 seats, and that range has just gone from 213-216 -- something like this could easily have national repurcussions. And there ain't a forecasting model in the world that can predict this before it happens.