four years ago The Economist magazine asked me to build a model for forecasting the results of the US presidential election. My colleagues and I did a pretty good job of capturing the uncertainty—we predicted Joe Biden would get between 259 and 415 votes, and he won 306.

This year, we’re back at it, for an even tighter race. We currently estimate that the two candidates, Kamala Harris and Donald Trumpthey have roughly equal chances of winning.

What is the fairest electoral system? The mega-election year sparks debates

Our model, like other election forecasts, uses national and state polls along with political and economic data from past elections. Combining these data sets generates correlated uncertainties about the election outcome in each state, which are added together to forecast the total candidates.

I believe that the main value of forecasts is not in the predictions themselves, but in how they describe the uncertainty and stability of the race over time.

Daily polls are attention grabbing but easy to overdo. Electoral forecasts — properly interpreted — can help us all keep our heads in an environment of information overload. After all, one of the most important roles of science is to temper enthusiasm for outlandish claims, whether of miracle cures or perpetual motion.

This year, the numbers coming out of our model won’t grab the headlines. Not much has changed in recent months. Harris’ winning probability hovering between 45% and 55% is hard to distinguish from the noise.

How algorithms influence and mobilize propaganda and distort reality

Based on forecast uncertainties, we estimated that typically a 10% change in a candidate’s probability of winning corresponds roughly to a 0.4 percentage point change in the national vote.

Four tenths of a percentage point is nothing – and in a close election it can be decisive. But it’s beyond the precision that forecasters can expect to get from any poll, or even any aggregate of polls, since the margin of error in most polls is about three percentage points. And polling biases could even double that margin.

It is impossible to know which forecaster is the most successful except in extreme cases. Evaluating forecasters based on their track record of predicting the winner reveals little. The differences between the results are too small and the elections are too infrequent for researchers to statistically identify which predictor is better.

For example, in the 2016 US election, the polling website FiveThirtyEight predicted that Trump had a 30% chance of winning, while the newspaper The New York Times gave him a 15% chance. Trump won, so the 30% prediction looks better than the 15% estimate – but it was just a roll of the dice. Indeed, if you estimate that an event has a 15% chance of happening, you would expect it to occur about once in six.

Misinformation might influence elections, but not in the way you think

The problem is that, in statistics, it is frequent events that allow researchers to judge whether models are better or worse—in sports betting or weather forecasting, for example, forecasters get daily data and have decades of past records that can be used for calibration. Events that occur every two to four years do not allow for such ratings.

We can use past performance to rule out overconfident predictions, such as those that gave Hillary Clinton a 99% chance of winning in 2016, but it could take hundreds of election years for scientists to distinguish forecasts that remain within reasonable limits. .

So why do I make predictions? First, political science. The fact that US presidential elections are predictable, to within a few percentage points, helps scholars understand US politics. This predictability affects the way politicians and journalists think about elections, the economy and the balance between parties.

Second, as baseball analyst Bill James would say, the alternative to good stats is not “no stats” but “bad stats.” While data-driven forecasts don’t offer the predictive accuracy that would allow forecasters to start the election early, they do provide useful bounds on the contours of the race — however fuzzy they may be.

What election science can reveal in this super-election year

In the absence of predictive models, political observers would be inclined to tell a story around every campaign event and every poll. Forecasting models don’t stop storytelling, but I think they make stories more sophisticated and politically accurate.

Why, then, is the election news so dominated by race and not politics? I have a theory.

If a voter is a committed political follower of the news, they probably already know who they will vote for. They won’t be very motivated to learn more about the candidates’ positions, but they are interested in who will win. This causes news outlets to commission and report on polls, which in turn promote probabilistic forecasts like ours.

Primary elections, where candidates are selected, are another story. Voters have several options to choose from within a single party, and these candidates are likely to have similar policy positions. Even the strongest partisans are motivated to learn more about the candidates and what their positions are on specific issues.

Correct before the presidential electionit makes sense for the media to appeal to the majority who already know where they stand, rather than those who are open to persuasion – and probably less interested in politics anyway. Ultimately, elections will always be uncertain because it is up to the individual to decide how to vote and whether to vote at all.