FairVote’s Monopoly Politics projection model is designed to predict the results of as many U.S. House elections as possible using the most simple of inputs. Using only data about the incumbent’s past margins of victory and the district’s vote in the most recent presidential election, the 2014 model predicted 368 House races in 2014 with 99.4% accuracy. New data from Tuesday’s election, however, allows us to evaluate the model more completely. When we incorporate two basic pieces of new information into the model’s default assumptions, it correctly projects the makeup of the entire House in 2014.
The Monopoly Politics model’s outputs are influenced by two kinds of national sentiment, both of which vary in magnitude considerably across elections: the electorate’s preference for one party over another and the electorate’s preference for incumbents over non-incumbents. The model accounts for these trends with two numerical inputs. First, the national two-party preference, which describes the electorate’s overall partisan lean and is expressed as the percentage of major-party voters at the polls who, all else being equal, preferred Democrats over Republicans; a national partisan swing of 46%, for example, would indicate a strong Republican wave year. Second, the national incumbency bump, which measures the degree to which incumbents benefited at the polls relative to non-incumbents, all else being equal; it’s expressed as a percentage value indicating how much better incumbents performed compared to a hypothetical generic candidate of the same party in an open-seat race in the same district. A national incumbency bump of 3% would indicate that incumbents tended to receive an extra boost of about 3 percentage points solely as a result of their incumbency.
However, these numbers aren’t set in stone. They vary unpredictably between elections, and their values for a given year can only be determined after the fact by examining actual election results. Thus a key feature of the Monopoly Politics model is that these values can be tweaked by users to simulate different national trends, such as a strong wave year for one party or a strong anti-incumbent sentiment. The 2014 model is archived here; the values in the top-left hand corner can be edited by viewers to simulate these effects.
The model’s default projections for 2014 assumed that voters as a whole would prefer Democrats and Republicans to the same degree, i.e. a national two-party preference of 50%. As well, they assumed that the incumbency bump would remain at its 2012 value of 4.5%. The model predicted 368 of 2014’s races with 99.4% accuracy using these assumptions. But now that the 2014 election is over, we can plug the actual 2014 national two-party preference and incumbency bump values back into the 2014 model and see how it performs. Let’s take a look at the model’s projections using the actual 2014 national two-party preference of 48.11% (see methodology notes below) and the actual national incumbency bump of 2.36%:
Strikingly, the model predicts the final tallies with complete accuracy – even including those races which it categorized as too close to make an official projection.
These results speak to the strength of the model, but also to the impact of 2014’s strong climate for Republicans and relatively weak climate for incumbents. In a nationally even partisan year in which incumbents were favored at the same rate as in 2012, the model predicts Republicans would have won 13 fewer seats, adding just two to their existing majority. Instead, with 51.89% of major-party voters preferring Republicans to Democrats, 2014 was a substantial if not overwhelming Republican wave year in the House. Meanwhile, 2014’s 2.36% incumbency bump was unusually low by recent standards: the national incumbency bump from 1998 to 2010 averaged 6.34%. However, peering below the aggregate number, it was Democratic incumbents who bore the brunt of low pro-incumbent support: while the median Republican incumbent enjoyed a reasonably healthy boost of 4.26%, the median Democratic incumbent’s advantage was just 0.46%.
A more detailed explanation of our methodology for calculating national two-party preference and national incumbency bump follows. A Google Sheets document containing the below calculations and relevant data is available here.
National Two-Party Preference. To calculate the national two-party preference, we first calculate each district’s partisanship value (expressed as a percentage of voters favoring Democrats), which measures the underlying partisan preference of the district’s voters independent of national trends. Partisanship is calculated by taking Barack Obama’s two-party vote share in that district in 2012 and subtracting half of his national margin of victory. Next, we take the two-party vote percentage for each House winner and subtract his or her district’s Democratic partisanship percentage to calculate the degree to which the vote in that district deviated from the district’s baseline partisanship. We then take the median result of these comparisons for each of three categories: open seats, seats held by an incumbent Republican, and seats held by an incumbent Democrat. We take the mean of these three medians to arrive at the national partisan swing, a percentage value indicating the deviance from an even 50-50 nationwide partisan split; negative values indicate a Republican lean and positive values indicate a Democratic lean. In 2014, the national partisan swing was -1.89%. We add this number to 50% to calculate the 2014 national two-party preference of 48.11%, which indicates that 48.11% of major-party voters preferred Democrats over Republicans.
Incumbency Bump. We use a similar process to calculate the national incumbency bump. We compare each incumbent’s performance to the partisanship of their district to calculate the degree to which incumbents over- or under-performed their district’s baseline partisan preference. We take separate medians of these values for Democratic incumbents and Republican incumbents. We then take the mean of these two medians to arrive at the 2014 national incumbency bump of 2.36%, which indicates incumbents could expect to perform 2.36 percentage points better than a generic candidate of the same party would have performed in an open-seat race.