More Election Reflections

Nov 5 '12

* WEATHERING THE ELECTION: With The Washington Post and USA Today writing about the impact of tomorrow’s weather on turnout in different swing states, it seems useful to point to a recent political science source on this.  A 2010 Quarterly Journal of Political Science article by Bernard Fraga and Eitan Hersh shows that rain depresses turnout only in uncompetitive elections–and it’s hard to argue that tomorrow’s Presidential election fits that description.

* ON JUDGING BINARY PREDICTIONS: Early in the fourth quarter of yesterday’s game between the Steelers and the New York Giants, Pittsburgh coach Mike Tomlin took a risk by faking a field goal that would have tied the game at 20 points a piece.  Many judged Tomlin’s decision to be an error based on the outcome: the Giants made the tackle and took over possession of the football with their lead intact.  But there is a random element to football, and the right decision or the right process doesn’t always produce the right outcome.  Why bring that up now?  The last few weeks have seen heated clashes about the value of forecasting election outcomes with statistics–and we should fully expect that come Wednesday, we’ll see horse-race style articles assessing the winners and losers among forecasters.  But at least some of those articles are likely commit the same error as Tomlin’s critics, by assuming that a single binary outcome validates or invalidates a stochastic model.  Yes, The Times‘ Nate Silver thinks an Obama victory is more likely than a Romney victory–but as of this writing, his model still has Romney winning the Presidency just under 14% of the time.  To put that in different terms, if we lived in a Groundhog Day-style world, and held the same election every day this month, Romney would be elected President on four different days. The best way to evaluate predictions like that is not based on a single binary outcome, however much our inclination to narrative might encourage us in that direction.  Instead, it’s to make lots of forecasts and to repeatedly evaluate their performance.  Voting in the 2012 U.S. Presidential election comes to a close tomorrow.  But opportunities to forecast election outcomes and to hone our models surely do not.