Why Kamala Harris’s chances of victory just jumped

Our statistical model of America’s presidential election will be updated six more times before votes are counted. There are few opportunities for candidates to move the dial in an election which has been stubbornly close since Kamala Harris became the Democratic nominee. Today’s update will cheer her supporters: the vice-president’s probability of victory rose by six percentage points, making the race a dead heat.

There are three reasons. One is the volume of new polls—65 were added to our forecast today—giving the model more confidence about small changes. Another is that there is so little time left before the election. Up until now our model has been a forecast, with weeks or months left for candidates to make gains. Many pollsters are now publishing their final surveys of the cycle, so the forecast will soon become a “now-cast”.

The third is that the race is remarkably close, which means that even tiny changes in expected vote shares can yield large shifts in win probabilities. The most influential polls yesterday were concentrated in four states: Michigan, North Carolina, Pennsylvania and Wisconsin. In those states, Ms Harris’s forecasted vote share rose by an average of 0.4 percentage points (see chart)—a small move that was nonetheless sufficient to increase her chance of victory by an average of six percentage points across the four.

Read more of our coverage of America’s presidential election

On the surface, the new polls did not look unusually good for Ms Harris. Most showed results that were close to a tie. However, the firms that released surveys yesterday—particularly AtlasIntel, Quantus and Trafalgar—have tended to give Donald Trump better numbers this year than have other pollsters who surveyed the same races at similar times. Our model shifts all poll results to counteract such biases. And on average, these adjustments nudged vote margins in yesterday’s swing-state polls around half a percentage point in Ms Harris’s direction. 

Moreover, in recent days the model has been moving towards Mr Trump, and Ms Harris’s average projected vote share (excluding third parties) had fallen below 50% in every swing state besides Michigan. As a result, new polls showing a tied race (like those in Pennsylvania did on average after our adjustments) or even a slim lead for Mr Trump (as did those in North Carolina) still represented an improvement for Ms Harris, compared with the model’s relatively gloomy expectations for her yesterday.

New polls also came out in Arizona and Georgia yesterday with a wide spread of results, ranging from an eight-point lead for Mr Trump to a one-point edge for Ms Harris. However, after our adjustments, the average of these new surveys landed very close to the model’s previous expectation of a two-point lead for Mr Trump in both states. As a result, the forecasts for Arizona and Georgia were unchanged.

Ms Harris’s small gains have brought her back to parity in Nevada, Pennsylvania and Wisconsin and made her a narrow favourite in Michigan, whereas Mr Trump retains a small but clear edge in Arizona, Georgia and North Carolina. The two candidates each won exactly half of our model’s simulations in its latest run. On average, they both wind up with 269 electoral votes—which would leave the House of Representatives to break the tie, presumably in Mr Trump’s favour. However, the model assigns just a 1% chance to an actual electoral-college tie, which would probably require Ms Harris to win Michigan, Pennsylvania and Wisconsin while losing Nebraska’s second Congressional district.

The direction or size of polling errors cannot be predicted. But if history is any guide, surveys are likely to underestimate one candidate by a margin that dwarfs the small day-to-day shifts in our model’s average estimates. Any such error would probably deliver a decisive victory to whichever candidate it benefits. Despite the tight polls, our forecast gives a two-in-five chance of the winning candidate receiving more electoral votes than Joe Biden did in 2020 or Mr Trump did in 2016.

The other main source of uncertainty in our model, aside from polling errors, is the time remaining until the election. The forecast works by estimating the candidates’ current positions with the available data, and then simulating movement that could occur each day until November 5th. With just six remaining, there is little movement left to make.

The effect on our forecasted probabilities is counterintuitive. There are few opportunities for big changes in public opinion, meaning polls published now have greater weight. As a result, the forecasted probabilities may change more substantially from day to day than they would earlier in the cycle. The slight movement in Ms Harris’s favour today is harder to reverse in the next six days than it would have been a month ago.

The polls in today’s forecast update were mostly based on interviews conducted a few days ago, so it is hard to judge what, if anything, caused a small uptick in Ms Harris’s standing. Some polls now being published were conducted after Mr Trump’s rally at Madison Square Garden on October 27th—which is now roundly considered to have been a misstep for his campaign—but it is unlikely to be until after the election that we have a clear idea of whether that event moved many voters. It appears as though the final six days of the campaign will go in a similar fashion to the past three months: plenty to talk about, but no decisive leader.