For years, polling showed a common pattern in democratic societies and stood as a way of figuring out public opinion such that candidates could form campaign strategies. However, this form of analysis has seen considerable adjustment and challenges in the last years, especially with the change in technology and social trends that dramatically alter voter behaviors. The 2016 presidential election marked a huge turning point in which important flaws in polling methods were illuminated, forcing polling companies not only to reevaluate traditional techniques, but also to consider how dependent polling is on reaching a large enough cross-section of people to predict election outcomes reliably. To understand modern polling methods, however, knowing how they were formed is essential.
Presidential polling dates back to the 1800s, specifically the election of 1824. The Democratic-Republican Party, running unopposed after the Federalists disintegrated, failed to unite around a single candidate, resulting in a contested election with four candidates. This led to a lot of confusion among voters, who began to take straw polls to see who their fellow citizens were voting for. These primitive polls were fairly rudimentary in design and often took place during community events, at churches, and in other social hubs. They were seen as a simple gauge to determine which political figures were popular locally.
People had different reactions to polls now compared to then: “I think people took this as interesting and informative, but not in any sense definitive,” says Tom W. Smith, director of NORC’s Center for the Study of Politics & Society. Essentially, straw polls were more about engaging citizens in the political process and sparking discussion than accurately forecasting outcomes. This trend still caught on, however, as newspapers started to conduct their own straw polls and make predictions about later elections.
The shift to modern scientific polling became a well-known practice in the 1930s with George Gallup’s scientific polling methods. At first, polls used door-to-door canvassing and telephone questionnaires to achieve a random sample from the public. Pollsters assumed that with these methods they could reach a diverse, representative cross-section of the population to make reasonably accurate predictions of the broader trends. The success of Gallup in predicting the 1936 U.S. presidential election and others after it proved this method reliable, and polling became a staple in American politics—not just for campaigns, but also for media organizations seeking to predict the outcome of elections.
As fast as technology was evolving, so were polling methods. By the 90s, telephone polling had become the primary method. But the reliance on landlines introduced its own challenges. As more households switched to cell phones and the internet, response rates declined, which made it harder to get a sufficient random sample. This trend shows how dependent polling is on reaching enough people to predict election outcomes reliably. At this point, when the demographic makeup began to adjust, pollsters started experimenting with online surveys and changing around statistical weighting to account for these changes. Unfortunately, these adjustments were not enough to surmount the challenges that emerged in 2016.
The 2016 U.S. presidential election is arguably one of the most significant moments in modern polling history. Most national polls agreed that Hillary Clinton was ahead of Donald Trump, although by a slim margin, and forecasters predicted her victory with confidence. Yet, on election day, Trump managed to win some key battleground states, and thus took the Electoral College—a result few had thought possible. The huge gap between the forecast from the polls and the actual outcome forced a lot of pollsters to change their polling practices radically.
A known cause of polling errors in 2016 is how many state-level polls among rural and working-class voters underestimated support for Trump, as these demographics were particularly hard to sample. In any case, the weighting of the polls omitted educational attainment, a variable that would later appear as one of the critical predictors of the vote. College-educated respondents were overrepresented in polls and were more pro-Clinton, which produced an artificially high prediction of support for her campaign. Another confounding variable was the “shy voter,” in which a reluctance to show public support for Trump stopped some respondents from being honest about their intention to vote for him. What resulted from these errors was a response bias that largely underestimated Trump’s support in many urban and suburban polls. These shortcomings revealed fundamental flaws in modern polling: traditional methods could no longer keep pace with the complexities of today’s election demographics.
In response to this revelation, Pew Research found that more than half of national public pollsters (61%) used methods in 2022 that differed from those in 2016. If features like weighting protocols were included in the analysis, that rate would be even higher. These adjustments were put in place to make polling more reflective of today’s electorate and better equipped to capture the randomness of voter behavior. Likewise, polling methodologies used in the 2024 election have included better demographic weighting for educational attainment: many now take into account the degree of formal education of potential voters, an adjustment that will yield a more accurate representation of voters across educational backgrounds. Furthermore, as response rates in phone surveys continue to fall, a growing number of pollsters are turning to online panels to reach more significant numbers of younger voters. They are also using hybrid models combining online, phone, and text surveys to reach further into rural and minority communities. This helps to better account for under-represented minorities, but can also make gathering the data more complicated. A final example is poll aggregates like FiveThirtyEight, which weights polls on characteristics such as accuracy and sample size; advanced statistical modeling then produces probabilistic forecasts rather than point predictions. Using big data and social media analysis-tracking to see the changes in what the public thinks is notably still experimental. Yet, it gives a good indication of how enthusiastic voters are and what issues are important.
While polling will remain an essential tool in democratic societies for shaping elections and informing public debate, the lessons of 2016 put into perspective how measuring public opinion changes over time. Moving forward, the evolution of polling must be characterized by adaptability and commitment to continuous improvement, ensuring that polls serve as a reliable reflection of society’s diverse voices rather than a predictive certainty.
***
This article was edited by Alexa Davidson and Angeline Wu.
