Our midterm predictions point to major gains for the Republicans

SIX WEEKS AGO, the Democrats looked like they were on track to pull off a remarkable triumph. The president’s party almost always loses power in midterm elections. Yet in mid-September The Economist’s statistical forecasting model suggested that they had an 80% chance of retaining control of the Senate. The contest is now neck-and-neck, with the odds slightly in Republicans’ favour (see chart). Where have the Democrats lost ground?

Almost everywhere, is the answer. Yet their slippage in states that once appeared nearly sure to vote Democratic is especially striking. Take Pennsylvania . On October 4th our average of polls estimated that John Fetterman, the Democratic candidate, was up on his Republican opponent, Mehmet Oz, by six percentage points. By November 4th his lead had slipped to one point—well within the uncertainty interval for our averages. In Arizona , Mark Kelly, the Democratic incumbent in the Senate, was up by eight points. He now leads by two. Democratic candidates have slipped by similar margins in vying for Democratic-held seats in Georgia and New Hampshire, with a negligent decrease in already-close Nevada.

They have also lost ground in once-competitive, Republican-held seats. Races in Ohio, North Carolina and Wisconsin now all lean decisively towards Republican candidates. That decreases the odds that the Democrats will mitigate their possible losses with a surprise upset.

The Economist’s weekly polls with YouGov, an online pollster, offer possible reasons for this change in fortune. First, undecided voters are less supportive of Joe Biden’s presidency than decided voters; among the first group, 30% say they approve of how the president is doing while 53% disapprove. That is in stark contrast with the 47% of decided voters who support Mr Biden (and 51% who do not). It is possible that as the election drew nearer, this group of undecided voters naturally drifted into the Republican camp.

Another possibility is that polls fielded earlier in the campaign missed groups of voters that were going to cast ballots for Republicans all along, thereby inflating Democrats’ leads. While this theory is hard to verify, past elections suggest polls struggle to reach these voters. A recent report by the New York Times about the paper’s own pre-election Senate polls, for instance, indicated that white registered Democrats were 28% more likely to respond to them than white registered Republicans.

Our crack team of election modellers is careful not to overstate the chances for either party. While the trend is not the Democrats’ friend, the race remains a coin toss. With less than a week to go, our analysis is that either side could win as few as 45 or as many as 55 Senate seats. If we told you that there was a 55% chance of rain in your neighbourhood you would probably grab an umbrella.■