07 November 2016

Almost Showtime

Election forecasts for the Presidency and the U.S. Senate have been remarkably volatile, although Hillary Clinton has still been the favorite or at closest tied with Donald Trump for most of the race. The Presidential race has been gradually tightening over the last couple of weeks, and the U.S. Senate race has dramatically tightened over the last week, according to polling.

It isn't at all clear how much of this is bad modeling and how much of this is reality. Much of the gain for Trump appears to come from reluctant, eleventh hour decisions to cease to be an undecided voter or to back the Libertarian candidate Gary Johnson. 

But, the tightening of the polls doesn't seem to comport with unprecedented early voting turnout by Democrats and may be due in part from a failure of prediction models to adequately reflect the impact of early voting and to weight mere "likely voters" who have not yet voted and tend to vote Republican properly relative to the votes of people who have already cast voters which seems to be weighted towards Clinton. For example, perhaps people who have already voted are less likely to respond to polling or are weighted equally when already voted has a 100% chance of impacting the final result, while a likely voter may only have a 90% chance of voting.

My intuition is that Clinton and Democratic candidates for U.S. Senate will over perform late polling across the board, but that could be simply cognitive dissonance at work. I find it hard to believe that Comey's disclosures at the FBI (which in the end amounted to nothing) rocked the vote very much.

In a day and a half, we will learn which predictions were on target and which were far afield.

No comments: