Audience: High School Students
Almost 245 million people were eligible to vote for the 2024 U.S. presidential election. However, 90 million people did not vote, making the voter turnout rate only 65%. Photo by visuals from Unsplash.
On November 5th, 2024, the entire nation watched the nail-biting race between presidential candidates Kamala Harris and Donald Trump. The air was thick with anticipation as people stood on edge, holding their breath, waiting for the final call. News networks flashed results in real-time, while social media buzzed with predictions and reactions. Each new update felt like a step closer to shaping the future of the country. With Vice President Harris earning roughly 48% of the votes, and 50% of the votes going to former President Donald Trump, Trump came out victorious, winning 312 out of the 538 electoral votes. Only 270 are needed to win. Though Trump won by quite a large margin, the poll results leading up to election day predicted that the results would be much closer than they actually were. ABC News used polling, economic, and demographic data to explore likely election outcomes. Last updated on November 5th at 6 a.m., they predicted that Harris had a 50% chance of winning, while Trump had 49% chance. Where did these predictions come from? What caused the discrepancy between polling predictions and the actual result?
Election outcomes are predicted through election polling, which is the process of surveying a sample of voters to gauge public opinion, preferences, or predictions about an upcoming election. Polling is done in a variety of ways:
Telephone polling: surveyors contact a random sample of voters via landline or mobile phone
Online polling: conducted on the internet using survey tools
Door-to-door polling: interviewers go directly to people’s homes to ask respondents questions
Mail-in polling: surveys are sent via the mail to a random sample of respondents; often have lower response rates.
Text message polling: survey organizations send questions to respondents' phones via text messages
Exit Polling: conducted immediately after people vote at polling stations to help predict election results after people have voted but before the result is finalized.
While many people may feel strongly towards one candidate or another and indicate their preference in polls, not actually end up voting. Voter turnout refers to the percentage of eligible voters who actually cast their ballots in an election, called the electorate. In 2024, roughly 64% of eligible voters voted, which is a slight decrease from 2020 in which 66% of eligible voters voted. However, in 2016 and 2012, the voter turnout was much lower, at 60% and 59% respectively. Voter turnout can be affected by a variety of reasons, including election type, voter engagement, voting accessibility, demographics, and/or other external factors. Generally, presidential elections have a higher turnout than local or midterm elections. Voter turnout is also higher when people are more interested in the candidates or issues present, especially if voting registration and process is easier. The age group with the highest voter turnout is people aged 65 and older. Poor weather and political scandals may discourage voters to vote and therefore reduce voter turnout.
All these statistics are determined through data science and statistical modeling. Data science in election polling is the use of advanced computational methods, like machine learning and statistical modeling, to analyze voter behavior, predict turnout, and forecast election outcomes. Statistical modeling is a huge aspect of election polling, with regression analysis playing a key role. Regression models examine relationships between demographic elements, such as age, income, and education, and voter behavior. These models help estimate turnout rates and candidate support by analyzing patterns in past and current survey data. For example, a regression model might reveal that older voters are more likely to vote, aiding in the prediction of turnout across age groups.
Another important statistical method is weighting, which adjusts survey results to better reflect the population. Pollsters – people who conduct polls – often oversample certain demographic sections to ensure there is representation of all groups, then apply weights to balance the final results. Time-series models are also commonly used to track voter preferences over time, incorporating external factors such as economic trends or political events. These models help anticipate shifts in voter attitude as election day approaches.
As technology is advancing, modern polling is moving beyond traditional statistics by utilizing advanced data science techniques. One such prominent example is machine learning, which allows pollsters to analyze vast datasets from past elections, surveys, and social media platforms. Algorithms can detect patterns in voter behavior, separate the electorate into likely and unlikely voters, and predict turnout or candidate support with stronger accuracy. Machine learning models are especially valuable for grouping voters into categories based on shared characteristics, such as political preferences. Polling organizations also use natural language processing (NLP), which analyzes text data from sources like social media, news articles, and public debates to understand how voters are feeling and identify trending issues. By interpreting the language used by voters, NLP can detect shifts in public opinion and issues that voters are concerned about like healthcare or immigration.
While there exists a vast range of techniques and tools to predict election results as reliably and accurately as possible, it is unlikely for any of these methods to be 100% perfect. For example, just a few days before the presidential election, a Des Moines Register/Mediacom Iowa Poll showed Vice President Harris leading former President Trump 47% to 44%. Though the polling organization was generally reliable, this was extremely unusual as Iowa has historically been a Republican state. However, though these polls revealed Democrats to be up by 3% points, in the actual election, Iowa was once again Republican. Polling results can be inaccurate for several reasons, such as:
Sampling error: the sample does not represent the entire population
Nonresponse bias: occurs when some people are more likely to respond to polls, while others do not
Question wording and phrasing: extreme or leading questions may influence a person’s response.
Mode of the survey: some polling methods may be easier for some people but harder for others
Timing: public opinion can change quickly, leading to fluctuations in poll results
As technology continues to evolve, with Artificial Intelligence growing smarter and more integrated into everyday life, it's likely that AI will revolutionize the statistical and technological methods behind polling. While recent polling results have faced challenges and inaccuracies, the next four years promise to bring new, more precise methods. In the midst of a thrilling presidential election race, where hundreds of polls captured the nation's attention, all eyes are now on January 20th, 2025, when Trump’s inauguration will mark the next chapter in American politics. The future of polling, driven by innovation, holds a new kind of excitement and possibility.
Bibliography:
Berger, A., Schneier, B., Gong, E., & Sanders, N. (2024, June 7). Using AI for Political Polling. Harvard Kennedy School. https://ash.harvard.edu/articles/using-ai-for-political-polling/
Croucher, S., & Perry, D. (2024, November 8). Did the 2024 Election Polls Get it Wrong? Newsweek. Retrieved November 27, 2024, from https://www.newsweek.com/were-2024-election-polls-wrong-pollsters-give-verdicts-1982170
Kronenberg, A., & Cline, S. (2024, November 15). How Many People Didn't Vote in the 2024 Election? U.S. News & World Report. Retrieved November 27, 2024, from https://www.usnews.com/news/national-news/articles/2024-11-15/how-many-people-didnt-vote-in-the-2024-election
Low Cost MRP Regression. (n.d.). Electoral Calculus. https://www.electoralcalculus.co.uk/services_polling.html
Mercer, A., Lau, A., & Kennedy, C. (2018, January 26). 1. How different weighting methods work. Pew Research Center. Retrieved November 27, 2024, from https://www.pewresearch.org/methods/2018/01/26/how-different-weighting-methods-work/
O'Connor, K., & Sabato, L. (2009). American Government: Roots and Reform. Pearson Longman.
Rodrigues, D. F. (2009). Modelling election poll data using time series analysis. Kent Academic Repository. https://kar.kent.ac.uk/94615/
2024 President Results: Trump wins. (n.d.). NBC News. https://www.nbcnews.com/politics/2024-elections/president-results
When is Inauguration Day 2025? Trump set to be sworn into office. (2024, November 18). USA Today. Retrieved November 27, 2024, from https://www.usatoday.com/story/news/politics/elections/2024/11/18/when-is-inauguration-day-2025/76328252007/
Who Is Favored To Win The 2024 Presidential Election? (2024, November 5). FiveThirtyEight. Retrieved November 27, 2024, from https://projects.fivethirtyeight.com/2024-election-forecast/
Комментарии