The problem with polling 

Seemingly every day, new polls report that Donald Trump is either ahead of Kamala Harris by 2 points or behind by 2 points.  

Such razor-thin leads are consistent across multiple battleground states, including Michigan, Pennsylvania, Arizona and Nevada. The size of each such poll, typically around 1,000 people or less, leaves margins of error of around 3 percent or more. What does this mean? That these states are simply impossible to call with any certainty prior to the final votes being counted.  

Polls are used to predict what the outcome of an election will be. Pollsters repeat polls for close races, hoping to capture moving trends, while for races that are noncompetitive, like Indiana or California, one or two polls suffice. All such polling numbers have been remarkable steady for weeks, and with early voting and mail-in voting now in full swing, many voters are locking in their votes well before Election Day.  

Yet for polls to be most informative, the people who are polled must be randomly sampled from a representative group of the people who cast votes in the election.  

If a poll of 1,000 people attempts to capture the sentiments of 2 million people who vote, then each person polled is effectively representing 2,000 people. Though this sounds unrealistic, statistical sampling theory makes this possible. 

If the people polled do not represent the people who show up to vote, the polling numbers become nonpredictive, perhaps even misleading. This was observed in the 2016 presidential election, when Donald Trump outperformed the polls in several key states and won an unexpected victory.  

How people are reached can also be an issue. Are robocalls made to landlines, cell phones, or texts? Landlines are disproportionately used by seniors, hence this skew must be accounted for. Emails and pop-up ads can be effective in reaching some sectors of population. Robotexts can also yield voter preferences amongst certain demographics.  

The complexity of voters and how to meaningfully sample their preferences is a challenge for pollsters, who seek to offer reliable indicators for the direction that the election is moving. Lessons learned from previous elections are being applied to improve polling reliability and accuracy, yet each election has its own unique characteristics, making polling adjustments a moving target.  

The campaigns themselves need to understand voters, since they must allocate resources for ads and mailings that will yield the greatest impacts. Indeed, what they spend for each additional vote can be a critical predictor of how well they will fare on Election Day. 

Polls are widely covered by the media, given the appetite voters have for such information. It is common for every campaign to claim that this is the most important election ever. But such statements are nothing more than marketing to encourage more people to vote. 

Despite their widespread reporting, however, polls this year remain mostly uninformative. Why is this the case?  

To paraphrase Yogi Berra, predicting the future is difficult because it hasn’t happened yet. When predicting the outcome of a tight election, it is even more challenging; with people’s behavior and sentiments involved, which are notoriously unpredictable, small changes are amplified in determining outcomes, given how the Electoral College determines the winner. 

Tendencies can be extracted from data. The problem is, with so many states effectively toss-ups, polls until Election Day are unlikely to move much in either direction. They will stubbornly stay within the margin of error, providing media talking points but no meaningful information. 

Polls also cannot account for unexpected events that can disrupt voting. Suppose that Philadelphia, Detroit or Milwaukee gets hit with a freak early-season snowstorm on Election Day, preventing some people from voting. Given that these urban areas are majority Democratic, this could upend all national Democratic candidates on their ballots, shifting elections up and down the ballot to Republicans. 

The same could apply if a hurricane were to hit Republican strongholds in Florida, Georgia or North Carolina (think Helene and Milton), except that here the Democratic candidates would benefit. 

This provides a good reason to vote early and lock in your vote as soon as possible. 

Rare or unlikely events do not commonly occur, by definition. Yet the unexpected does sometimes happen, and predicting where and when is nearly impossible. 

With the election just two weeks away, new polls will say nothing meaningful, even if campaigns and the media widely report them. We will all need to wait until the official vote counts begin and the final winners are announced. This may occur several days beyond Nov. 5, since a number of key states might be so close that recounts are needed to certify the results.   

At that time, we will see how uninformative most polls were, or cherry-pick nuggets of prescience that they offered based on the eventual winners. Indeed, it requires no great skill to play Monday morning quarterback and sound smart after the final results are announced.   

Sheldon H. Jacobson, Ph.D., is a professor of computer science at the University of Illinois Urbana-Champaign.