Polling and Winning on the Margins

By: Heather Farley
November 13, 2024

We are closing the door on the 2024 Presidential election and, once again, polls and the media fell short in predicting the outcome. Since Donald Trump entered politics in 2016, polls have often underestimated his support, and 2024 was no exception.

There is a popular narrative that this is because both the polls and the media are biased toward Democrats. While media bias can be a valid issue, I propose an alternative explanation grounded in polling limitations and the dynamics of this year’s race.

As a social scientist, I have spent a fair amount of time analyzing methods and biases in research and when it comes to researching humans, there is definitely room for error.

You likely noticed in the many Presidential candidate polls that came out over the last sixth months or so, that there is a margin of error listed. This margin of error tells us how much the poll results might vary if the survey were conducted multiple times. It’s shown as a percentage and gives an idea of the uncertainty in the poll’s results. So, if the margin of error were ±3%, we could assume that either candidate could go up to three percentage points higher or lower than the reported number. A “good margin of error” is at or below ±3%. When polls approach ±5%, it is considered too high to draw useful conclusions. In 2024, however, the race was considered so close, particularly in the swing states, that the polling was effectively useless. In other words, with both parties polling a percentage point or so higher or lower than their opponent, margins of 3% didn’t give us much indication of what to expect.

Likewise, the outcome of the election was on the margins. Indeed, Trump swept the swing states because he picked up the one or two percentage points, and sometimes even less, needed to win the electoral college votes in key states. The surprise, I think, was that he earned the margins in almost every case leading to a sweep.

Was this a failure of poll legitimacy? Not necessarily. Leading polls did not exhibit faulty methods or embedded biases. Instead, human factors limited data reliability.

Good data requires good sampling. Sampling in this case can be tricky. Republicans are less likely to engage with media or polling due to distrust, leading to underrepresentation. Conversely, Democrats, particularly college-educated ones, often engage more, skewing data representation.

So where can we turn in future elections to help us predict what might be coming? One new tool released by the Washington Post this year caught my eye and I hoped it might be useful – not in predicting outcomes but levels of support. The tool showed candidates’ online donor counts by zip code, offering insights into grassroots support. In Glynn County, for example, Trump had only 7% more donors than Harris but secured a 26-point lead, demonstrating that donor counts, while indicative of enthusiasm, aren’t great predictors. Not in Glynn County at least.

An interesting, if not controversial, place one might turn for consistently accurate prediction is betting markets. In this election cycle, several prediction betting companies turned out to be successful in accurately predicting the eventual outcome. Companies such as Kalshi and Polymarket put the odds of winning the election in Trump’s favor earlier and more consistently due to the numbers of bets they were receiving. Free market purists may not be surprised by this.

These markets are not scientific in any way and come with potential problems (eg. foreign participation, financial incentive to work against certain election outcomes, low governance and regulation of these platforms). But, when people put money into something, there’s no lying or virtue-signaling to mislead the messaging. In a historical look at Presidential betting markets, Economics researcher, Koleman Strumpf highlights that cutting out the human pitfalls and looking just at the numbers has proven to be remarkably accurate and efficient.

I still believe in the legitimacy and validity of polling. In many cases polling is an excellent tool. When it comes to Donald Trump, or any candidate who can divide the nation so narrowly, I’m not so sure. This season has convinced me to look to the markets for clues as well.

Dr. Heather Farley is Chair of the Department of Business and Public Administration and Associate Professor of Public Management at the College of Coastal Georgia. She is an associate of the College’s Reg Murphy Center for Economic and Policy Studies and an environmental policy scholar. The opinions found in this article do not necessarily represent those of the College of Coastal Georgia.

Reg Murphy Center