With recounts, absentee ballots, and voting by mail, we don’t yet know the precise number of votes that the two parties won in the 2010 House races. But I want to develop a point John suggested just after the election: to the extent that the GOP outperformed the expectations of political science forecasts based on the generic ballot, it seems to have done so by translating its votes into seats more effectively than expected.
Consider the July 2010 forecast by political scientists Joseph Bafumi, Robert Erikson, and Christopher Wlezien. At that time, their estimate of the GOP share of the two-party vote nationally was 52.9%, which is not far from the 53.4% that GOP candidates are estimated to receive. Certainly, the 95% confidence interval for this estimate—with the Democrats receiving between 43.1% and 51.1% of the two-party vote—included the true value of 46.6% near its midpoint. Yet interestingly, their model translated that national vote into an average GOP majority of 229 seats, with a 95% confidence interval from 199 to 259. Since the right number is likely to be around 245 seats, the implication is that the model was perhaps a bit better at predicting the national two-party vote (where it was off by only 0.5 percentage points) than in translating the national vote into seats (where it understated GOP gains by around 16 seats, which is a bit more than the 0.5 percentage point change implies). [Author’s note 11/09/10: I corrected the 95% confidence interval for seats from the original post. Thanks to Chris Wlezien for pointing out the error.]
Pundits and political scientists alike have long recognized that a combination of geographic concentration, majority-minority districting (gated), and perhaps partisan redistricting (gated, gated) mean that GOP voters are distributed more efficiently for our first-past-the-post elections. So the GOP will typically take slightly more districts with the same national vote share. The recent past bears this out. Last week, the Republican Party won approximately 245 seats with around 53.4% of the two-party vote, while in 2006, the Democrats won an almost identical share of the two-party vote (53.6%), and with it took 233 seats. Still, any structural imbalances should be built into models like that of Bafumi, Erikson, and Wlezien, as these models take into account the district’s past voting patterns. So what’s going on?
One prime suspect—anticipated by a comment on the Monkey Cage last week—is the size of the incumbency advantage. The Bafumi, Erikson, and Wlezien model gives freshman representatives a boost, expecting to see a “sophomore surge” as the incumbency advantage kicks in. On account of that, their translation of votes to seats actually expected a slight structural advantage for the Democrats in 2010, since the Democrats had many more incumbents. Expecting a robust incumbency advantage is in keeping with a long tradition of political science research. But it is plausible that under specific conditions (say, a down economy, several tough votes in Congress, and well funded challengers), the value of incumbency might be muted for the incumbent party. In the average midterm year between 1974 and 2006 (excluding 1982 and 2002 due to redistricting), a Democratic incumbent was expected to win an additional 12.8 percentage points of the two-party vote even accounting for the outcome of the last election. That is not a causal estimate, but it provides us with a baseline. In 1994, that figure dropped by 2.6 percentage points. Did Democratic incumbents not have the expected advantage in the 2010 midterms either? And what might that tell us about the capacity of candidates to cultivate a personal vote in their first few terms in office? Or in a period of nationalized elections? These seem to be central questions in understanding the outcome, and in understanding the translation of votes to seats in today’s America.