In 2012, two sets of journalists eagerly awaited the jobs report each month: economics and election reporters. In 2014, the economics reporters are still following the news with great interest. Yet Friday’s jobs report was the last before the midterm elections, and we barely heard a peep about it from politicos. In fact, I think I’ve written the word “economy” just once (or twice) to explain this year’s election trends.
The reason is simple: Accounting for the state of the economy doesn’t hold much, if any, predictive value for congressional elections. It’s not that the economy doesn’t affect House or Senate races — Democrats might be doing even worse if the economic recovery wasn’t somewhat decent. It’s just that other variables already sufficiently account for the economy’s effect.
We can see this when looking at FiveThirtyEight’s presidential and Senate models. Both rely primarily on the polls — and more and more so as we get closer to the election. Both also take into account “fundamental” variables that help us better understand the races. Both have been back-tested to ensure that including fundamentals makes for more accurate projections.
In the presidential model, FiveThirtyEight includes an economic index variable. This index combines seven different measures of economic strength in the lead-up to the presidential election. In the Senate model, FiveThirtyEight looks instead at the generic congressional ballot and congressional approval ratings. Right now, the generic ballot is not looking good for Democrats, while congressional approval is a little bit more hazy given the split control of Congress.
FiveThirtyEight is not alone in its different treatment of congressional and presidential elections. This year’s American Political Science Association conference showcased three well-known forecasting models for the midterms — none of them uses an economic variable, but the same authors include an economic growth variable in their presidential models.
Emory University Professor Alan Abramowitz’s Senate model looks at the generic ballot in early September, seat exposure (i.e. the number of seats each party has among the seats up for re-election) and whether or not the election is a midterm. His presidential model includes a variable for GDP growth in the second quarter of the election year.
SUNY Buffalo Professor’s James Campbell’s House and Senate models look at the number of seats in trouble (as measured by the Cook Political Report), the president’s approval rating in late August and the number of seats held by each party going into the election. His presidential model also includes second quarter GDP growth.
For midterm elections, Dartmouth College Professor Joseph Bafumi, Columbia University Professor Bob Erikson and University of Texas Professor Chris Wlezien’s House model looks at the generic ballot at different points of the midterm cycle and which party controls the White House. Erikson and Wlezien’s presidential model includes a variable for the cumulative growth of an index of leading economic indicators.
The distinction between how the authors treat congressional and presidential elections is clear. Importantly, their explanation matches our reasoning behind not including an economic variable in our model. As Bafumi, Erikson and Wlezien note, “economic performance has no clear effect on the midterm vote. We tried two economic measures. … Neither variable (multiplied by presidential party) is significant when added to our models.”
Put it all together and you can see why we haven’t spoken that much about the economy in regards to the Senate elections this year. Other measures tell us far more about where the election is going to end up.