There’s been a lot of digital ink spilled recently over the rise of election forecasting models. (Did you know there’s a 77 percent chance Republicans will take the Senate next month? Or maybe it’s 66 percent. Or 59 percent. Maybe 62 percent.) I won’t get into the debate over whose model is better, or whether forecasts are bad or good for political science. Rather, I wanted to address an interesting question raised by Brian Rosenwald, a graduate student at the University of Virginia: Could these forecasts influence the election?
The idea is that with so much information out there about who will likely win, some non-trivial number of potential voters may be influenced. You may plan to vote knowing nothing about the outcome and just assuming that your vote has some chance of mattering. But if Nate Silver tells you that one candidate has a 90 percent chance of winning, why bother to vote?
Voter turnout, while certainly not something to take lightly, is probably not going to be affected by the proliferation of forecast models.
This isn’t an idle question. In a famous example, NBC News called the 1980 presidential election for Ronald Reagan before polls had closed on the West Coast. Californians on their way to the polls may have changed their mind at that point and just skipped voting. This wouldn’t matter much if both sides were equally discouraged. That is, a suddenly depressed Jimmy Carter supporter may have been equally likely to stay home as a suddenly ebullient Reagan supporter. But as Michael Delli Carpini and others have found, the turnout effect was not equal; it depressed Democratic turnout more. That wasn’t enough to change the presidential election in 1980, but it may have affected more than a dozen congressional races, giving Republicans greater representation in Congress than they otherwise would have had. There may have been a similar effect in Florida’s panhandle region in 2000 after some channels prematurely called the state for Al Gore; the western part of the state lies in the Central time zone and had a later poll closing time.
Could we be seeing such an effect today? Probably not, for a few reasons.
For one, most of the famous examples of early calls affecting turnout come from, unsurprisingly, presidential elections. That’s partially because presidential elections get all the attention anyway and span multiple time zones, but also because presidential elections involve a more fickle group of voters. There are people who show up for a presidential election that just won’t vote in any other election, and those are the people most likely to change their mind about voting if they’re told it’s going to be a blowout. This year is, of course, a congressional mid-term; those fickle presidential voters aren’t showing up anyway. It’s a more hardcore electorate that’s less likely to be deterred.
For another reason, the forecasts that are getting all the attention are the nationwide ones about which party will control the Senate. Even if voters are paying attention to those forecasts, they probably won’t apply them to their own statewide race. Even if, say, a Colorado Democrat has accepted the notion that Republicans will control the next Senate, she will still likely vote for Democratic Senator Mark Udall simply because she doesn’t want to be represented by a Republican senator.
Finally, these concerns about turnout tend to massively overstate the amount of information that individual voters have at their avail. If you’re in line to vote or driving to a polling station and suddenly hear a radio report or see a TV headline that a presidential race has been called, that’s a very loud signal that few are likely to miss. But if you’re going about your day and a blog posts a forecast that Republicans have a 77 percent chance of taking the Senate? Even if hundreds of thousands of people read that, you’re still largely in geek territory there. The typical voter—especially one who may decline to vote based on the news—just isn’t likely to get that message in the first place.
So my take is that voter turnout, while certainly not something to take lightly, is probably not going to be affected by the proliferation of forecast models. Forecasts may be good or bad, but they’re probably not determinative.