In reaction, Clarity Campaign Labs‘ Tom Bonier, who had partnered with Issenberg, wrote this rebuttal. Part of Bonier’s response is that his data, which come from actual voter files, are superior to the Current Population Survey. The voter files allow Bonier a closer look not just at the demographic composition of the electorate, but which members of different demographic groups actually voted.
So here’s where I step in. My notion was this: What if we had a survey asking people how they planned to vote in their House race, and we knew from the voter file whether they had voted in 2010 and 2012? This would exploit the voter file data that Bonier prefers and allow us to conduct a hypothetical like Enten’s — that is, what if we re-ran an election with the electorate from a different election?
Thanks to Bonier and David Margolis of Greenberg Quinlan Rosner Research, I have the precise data. The data come from a September 2012 GQRR poll conducted in 54 competitive congressional districts. The poll asked whether respondents planned to vote for the Democrat or Republican running in their House race. But unlike in most pre-election polls, the vast majority of poll respondents (89 percent) were ultimately matched to voter file data that indicated whether they voted in 2010 or, several weeks after the poll was taken, in 2012.
Among respondents who voted in 2012, 48 percent supported the Democratic House candidate and 46 percent supported the Republican House candidate. That’s the Democratic advantage that we might expect in a presidential election. What about respondents who voted in 2010? This group does not include voters who turned out only in the presidential but not the midterm election. Among 2010 voters, the generic ballot results were reversed: 46 percent supported the Democratic House candidate and 48 percent supported the Republican candidate. In other words, switching from the 2012 electorate to the 2010 electorate shifted the generic ballot from a 2-point Democratic advantage to a 2-point Republican advantage.
What would this have meant in terms of House seats? In 2012, the Democrats had a 1.2-point edge in the national House vote and ended up controlling 201 seats. If the electorate had resembled 2010 and Republicans had had a 2-point advantage in the national House vote, there would have been 3-point swing to the GOP overall. A simple votes-seats curve from 2012 suggests that a 3-point swing in Republicans’ favor would have left the Democrats with 181 seats, or 20 fewer than they controlled after the 2012 election.
Twenty seats is not nothing, of course. But it suggests that simply shifting from midterm to presidential electorates, or vice-versa, can’t explain all, or even most, of the differences in outcomes between these two types of elections. Turnout is not going to explain a 63-seat gain for Republicans in 2010.
So here is where I come down in this debate. No one disagrees that “turnout matters,” and of course Democrats should work hard at turning out Democratic voters in 2014. This is what made Issenberg’s piece and Bonier’s analysis so interesting.
The question is how much turnout matters. My sense is that commentators still put too much emphasis on it. That is, there is not enough grappling with what changes in the electorate do not explain — such as, perhaps, the majority of Republican seat gains in 2010. There is not enough grappling with how Democrats did so well in 2006 despite a midterm electorate, as political scientist Michael McDonald has noted. For more, see Mark Mellman’s four excellent columns on this, and especially political scientist Seth Hill’s research.
[Correction: The post originally said that the GQRR poll had asked people the “generic ballot" question to measure their vote intention in House races. In fact, the poll provided respondents with the actual names of the House candidates running in their district.]