Will America’s midterm elections vindicate the polls or once again confound them? In 2016, most of them wrongly predicted a victory for Hillary Clinton; two years ago, they reckoned that Joe Biden would win far more easily than he did. So, their figures are biased towards the Democrats, right?
Not so fast. Yes, they were out in the last two presidential contests, but not that much: Clinton did in fact win the popular vote fairly comfortably; and their national figures two years ago overstated Biden’s support only slightly. Moreover, at the last midterm in elections in 2018, the polls were commendably accurate.
There are three risks in translating the plethora of recent polls into a firm prediction as to whether the Democrats will hold the Senate and/or the House of Representatives this time.
First, vote swings vary far more from place to place in the US than in Britain. National figures can indicate overall shifts in the national mood—for example the popularity of Joe Biden and Donald Trump—but are not as good at predicting the likely winners of individual key races for the House or Senate. Polling is needed at state- and district-level to capture these variations.
Second, such state and district polls have a worse track record than the national polls. In 2016, while the national polls correctly showed Clinton winning the national vote, they failed to pick up the scale of the shift to Trump in the Rust Belt states that ensured his victory. Pollsters investigated what went wrong. One of their main conclusions was that their state-level samples contained too many graduates (who were pro-Clinton) and too few voters without any higher education (who favoured Trump). Two years ago, they put this right—but still understated Trump’s support.
Various theories have been advanced for this—for example, perhaps some Trump supporters regarded pollsters as part of an establishment they distrust, so wouldn’t take part in their surveys. This seemed to me perfectly credible, and could explain the polls’ successes in 2018, when Trump’s name was not on the ballot. However, pollsters, working together to tackle a problem they know they must solve, have been unable to either prove or disprove this hypothesis.
Third—and this is an additional problem, not an alternative to the two above—pollsters can never be quite sure which of their respondents will actually turn out to vote. They can question a perfect sample, but still get the result wrong because of the intrinsic difficulty of distinguishing between those semi-interested electors who decide on the day to make the journey to the polling station and those who choose at the last minute to stay at home.
This is a particular problem in midterm elections. Turnout is never high, and in recent decades has gently declined, from 52 per cent in 1982 to 42 per cent in 2014. Then four years ago, when the Democrats made big advances, the figure rose by more than a quarter, to 53 per cent.
What happened was that Democratic voters turned out in record numbers, defying the historic pattern of Republican supporters being keener than Democrats to vote in midterm elections. Analysis by the US Census Bureau reported massive increases in turnout among black and Hispanic voters and men and women under 30—precisely the groups who, if they vote, tend to vote Democrat (though Republicans are now making some inroads with Hispanic voters).
This year, Democrats hope that they can repeat the trick of winning the ground war and getting out their supporters more successfully than the Republicans. Hence their concentration on such issues as abortion and hatred towards Trump. If the Democrats retain the House and add to their ranks in the Senate, it will be the party’s success in the ground war that matters most.
All this does not mean we can say nothing about the mood of America. This is where the national polls help. Biden’s approval rating is poor, with 42 per cent approving of his record and 53 per cent disapproving—a net score of minus 11. In late summer he made progress: In July, his net rating was minus 19. In early October this had improved to minus nine. Since then, the gap has widened slightly. At the same time, polls in key Senate races have reported a trend towards the Republicans, for example in Pennsylvania, Georgia, Arizona, New Hampshire and Nevada. Plainly the recent rise in inflation has done the Democrats no favours.
The latest figures are close in all five contests. But the cumulative impact has been to dent Democrat hopes of retaining the Senate. This is clear from the FiveThirtyEight polling site, which has a sophisticated model for combing the myriad of polling numbers to forecast the overall result. In mid-July it gave the Democrats a 46 per cent chance of retaining the Senate. Two months later the figure was up to 71 per cent. It’s now back to 46 per cent. (FiveThirtyEight reckons that the Democrats have only an 18 per cent chance of retaining the House, down from 32 per cent six weeks ago.)
I wouldn’t bet against FiveThirtyEight’s forecast; but neither would I bet much on it coming true. Will turnout be high enough for the Democrats to retain at least the Senate, and limit their losses in the House? That big question remains; we shall know the answer soon enough.
