Media stories on the US Senate race between Democratic Congressman Ed Markey and Republican Gabriel Gomez are breathlessly describing Gomez as a 2013 version of 2010 Scott Brown: independent, moderate, handsome, wears a cool jacket, and a grave danger to the Democrats (which also describes 2012 Scott Brown). Why, Public Policy Polling shows Markey with only a four point lead! Danger for the Democrats?
Probably not, suggests a model developed by political scientists Kedron Bardwell and Michael Lewis-Beck, “State Level Forecasts of U. S. Senate Elections,” PS: Political Science and Politics 37: 821-826 (2004).
I say suggests because, well, I just can’t get to devising a proper model for Massachusetts right now. But in 2002 Maine Democrats asked Bardwell and Lewis-Beck to build a prediction model for the US Senate race between incumbent Senator Susan Collins and Democratic challenger Chellie Pingree. The scholars built three models, the best of which predicted Collins vote percentage within three points – in July, over four months before the November election. Their summer forecast was more accurate than any poll taken in the race except the final one, conducted a week before Election Day.
So here’s how they did it.
First they modified a national forecasting model to predict that the incumbent’s vote would be a function of national economic and political conditions, plus incumbent party previous vote. So the more popular a president is nationally, the higher vote the candidate of his party should receive. Also the higher the rate of national economic growth, the higher the vote of the president’s party. In Maine, just those variables were not a good predictor.
But for another model the scholars added a simple dummy variable for a high quality or low quality candidate. That improved the predictive capacity of the forecasting model a great deal. But Bardwell and Lewis-Beck were still not satisfied. So they constructed an index they called the Incumbent Spending Advantage, as the ratio of incumbent to challenger spending in a two-year cycle. When they added that variable they had a very good model, explaining 97% of the variance in the incumbent party’s vote share.
How good were the nine public polls taken through the year? Well they all had Collins winning. But they vastly overestimated her margin of victory, which likely suppressed Pingree’s ability to raise money and build a volunteer base. Except for the last survey all the polls were woefully off in predicting vote percentages, and the last one was of little use for party strategic purposes because it was taken a week before the election. So the scholars’ forecasting model is much more helpful for a party in allocating its resources than are polls.
I wonder if the forecasting model might be refined further, for example, using the president’s state popularity rather than national popularity, or the state’s economic growth rate.
Thinking of this broadly for the Gomez-Markey race, incumbent party previous vote obviously favors Markey. Outside of the Brown win in 2010, the last Republican to win a senate election in Massachusetts was Ed Brooke’s re-election in 1972. Most of those races were blowouts (many with low quality candidates). Presidential popularity and economics both would favor Markey. The economy may not be great but it was good enough for Obama and Elizabeth Warren in November 2012 and it probably still is (think of the economy and President Obama’s popularity in January 2010 and you have a better insight into Brown’s win than you get from fawning over his pickup truck). Sorry Gomez fans, but Bardwell and Lewis-Beck would judge him a low quality candidate. Gomez should be able to spend money and there will be outside spending on his behalf (not a factor in the Maine model) but really, are the Republicans going to outspend the Democrats in Massachusetts?
I leave you to your own predictions.