What an election model tells us, and what it leaves out.
We’re far enough into the 2018 midterm election cycle that we have some sense of what the fundamentals of the political environment that will govern it look like. A pretty simple forecast model that relies on economic performance and presidential popularity predicts Democrats will pick up 45 to 50 House seats this fall, and take over 15 to 20 state legislative chambers. A loss of just 24 House seats would flip House control to the Democrats.
Okay, that’s the headline. How realistic is this model, and what is it based on?
The economic performance variable is per capita real disposable income (RDI), measured from the second quarter of the year before the election to the second quarter of the election year. Presidential popularity is the president’s Gallup approval rating on Labor Day of the election year. The data ranges from 1950 to 2014.
Most years, this model works fairly well. It predicted Democrats losing 46 House seats in 2010 (they lost 63), and it predicted Republicans losing 40 House seats in 2006 (they lost 31). The model has an R-squared of .568, meaning that just those two variables explain about 57 percent of the variation in the shifts in House seat shares. I can improve the model fit considerably by including a measure of the majority party’s exposure (how many seats in the House it currently controls), but that barely budges the forecast for 2018.
The graph below shows a range of predicted outcomes according to the model, based on three levels of presidential approval (35 percent, 50 percent, and 65 percent) and economic growth ranging from -2 percent to 8 percent.
As the graph reminds us, the president’s party tends to lose seats in congressional midterms. Since 1950, only in 1998 and 2002 did the president’s party manage to eke out a few seats, and both of those were during unusually popular presidencies and solid economic expansions. The rest of the time, it’s a loss for the party, affected substantially by the economy and presidential approval.
Of course, we don’t know precisely what President Trump’s figures will be heading into this election, but his approval rating has generally ranged between 35 and 40 percent. It’s possible that strong economic news could buoy his approval somewhat (his numbers have improved slightly in recent weeks), but his personal behavior has managed to step on what would otherwise be good news for presidents ever since he took office.
Economic figures, meanwhile, are showing only modest growth so far — per capita real disposable income only rose by around 1 percent between the last quarter of 2016 and the last quarter of 2017. It’s possible the recent tax legislation could put more money in people’s hands and boost these figures. Even if RDI growth jumped to 3 percent, though, the model would still predict Republicans to lose 37 House seats, more than enough to lose control of the chamber, and 14 state legislative chambers.
(Note that I’m not saying which state legislative chambers might flip, although this is a useful guide. Nor am I forecasting the US Senate, where it will be difficult for Democrats to take over the chamber.)
What doesn’t this model account for? A big factor is district safety. Due to the concentration of Democratic voters in Democratic House districts, which leads to a lot of “wasted” Democratic votes, Republicans can lose the national House vote substantially and still retain a majority of seats. Conversely, Democrats have been running very strongly in special elections over the past year, and they’ve recruited record numbers of congressional candidates, usually a positive sign for a party’s prospects. A wave of retirements by prominent Republicans in Congress also suggests good fortune for Democrats, making it harder for Republicans to hold some competitive seats.
More generally, as with any forecast model, past performance is no guarantee of future results. It’s possible things are just different now. Perhaps Trump’s Republican supporters are so loyal that they’ll turn out no matter how challenging the fundamentals are, limiting Democratic gains. Perhaps Democrats are so angry right now that their turnout will far exceed that of the out party in previous elections.
But it’s worth remembering that despite all the weirdness of the 2016 election, the results that year were astonishingly close to the predictions of conventional forecast models. Even as we continue to note how abnormal much of the current political system is, it seems a safe bet that voters will continue to behave as they have in the past.