The Democrats are highly likely to take the House

“WHEN THE facts change, I change my mind,” said—perhaps apocryphally—the economist John Maynard Keynes. Over five months ago, The Economist launched a statistical model to predict the results of this year’s mid-term elections for the United States House of Representatives. At the time, we called the Democrats a “clear but narrow favourite” to regain control of the lower chamber of Congress, with a 65% chance of securing a majority.

Our model’s forecast has been updated every day since then. However, the variables it used to make its prediction—primarily national polls, results of special elections for vacant legislative seats and economic conditions—have remained within a very narrow range. As a result, its prediction has barely budged, hovering around a 70% chance of a change of control for the past few months.

During that time period, however, new sources of data have become available that did not exist when our model was launched. We now know which candidates the major parties have nominated in every race, allowing us to account for their experience in politics, their ideological orientation, how much money they have raised and whether they are the subject of criminal or ethics investigations. We also know the results of the “top-two” primary elections held in California and Washington state, which tend to correspond closely to final vote totals. And we can complement national-level surveys with hundreds of polls of voters in specific districts, almost none of which had been conducted by May. Thanks to a record fundraising haul and favourable district polls, these factors provide a sizeable boost to the Democrats’ chances.

With just four days left until the election, there are no further types of information that will emerge before voters head to the polls. So we are pleased to publish a revised version of our model, which combines all of these new variables with the data sources already used by its predecessor. Our prediction for the national popular vote (after imputing results in uncontested districts) is unchanged, at 54% for Democratic candidates. However, our estimate of the Republicans’ structural edge in the House—the degree to which the district lines favour the GOP, thanks to a combination of gerrymandering and Democratic voters’ tendency to cluster inefficiently in cities—has declined, from around 3.5 percentage points of vote share to just under three. As a result, our best guess for the Democrats’ final seat total has risen from 222 seats to 229. Moreover, the addition of district-level surveys reduces the uncertainty about the result in races where pollsters have been active, increasing our estimate of the probability that a candidate with a modest lead will go on to win. The combination of a higher average prediction for Democratic gains and less volatility surrounding it raises the party’s chances of re-taking the House to 86%.

Our revised model still sees a Republican victory as squarely within the realm of plausibility. There are only 210 districts—eight fewer than the bare minimum needed for a majority—in which the Democratic candidate has at least a 60% chance to win, and just 220 in which the Democratic nominee is favoured at all. However, the playing field is impressively broad, and sits overwhelmingly in Republican-held territory. There are 254 districts in which the Democrats have at least a 20% shot, and a whopping 277 in which they have at least one-in-ten odds. Whichever party beats expectations is likely to win a comfortable majority of these races—but not necessarily all of them. Even if the Democrats do worse than expected, they only need to win a handful of these seats to capture a narrow majority.