On February 28, 2016 I released my fundamental model for the state-by-state presidential election. The model, using incumbency, presidential approval, economic indicators, and past voting had the Republican getting 292 Electoral Votes and the Democratic candidate 246. The final vote tally (assuming no stray electors) will be 306 to 232 … I (along with co-author Patrick Hummel) am awesome! Of course, I did not believe that forecast, because fundamental models provide a baseline prediction of a generic Republican and Democratic candidate. We did not have generic candidates, we never have generic candidates, and we had polling and prediction markets already showing a Democratic bump above the generic outcome.
Above is the same chart posted on February 28, but I have changed the colors to reflect the final outcomes and ordered the states to reflect the fundamental, not PredictWise market-based, predictions. The PredictWise market-based predictions missed four states (MI, WI, IA, PA), but they all went from the Democratic to Republican camp, flipping the overall result. The fundamental model missed five states, but was much closer to the final answer. It showed PA (the critical swing state!) going Republican but it also thought VA, and NV would as well. Also, the fundamental model missed MI, WI, and IA, as well.
I have no regret about allowing our model to progress as new information became available. Clearly some of the new information was right and some of it was wrong, but it is absurd to give up on filling in the dispersed and idiosyncratic data of the campaign, in favor of pitting generic candidates against each other due to the result of single outcome.