Can Social Media Platforms Predict Elections?
On the campaign trail, I’ve heard it said, “you always want the polls to show a close race; you don’t want your supporters to be overconfident or without hope. One way, they don’t show up because they don’t think their vote is needed; the other way, they don’t think their vote will be enough.”
The United States and the world are dealing with the ramifications of COVID-19. Businesses and schools across the country were forced to close, and states postponed primaries and looked for alternative voting methods. Soon the 2020 election season will be one for the history books, but not too long ago, many were still trying to figure out how Hillary Clinton didn’t win in 2016. Looking back at the historical polls, there was every indication that the former Secretary of State would be the new Commander and Chief. It had been said that, “the 2016 presidential election was a jarring event for polling in the United States” (Kennedy, et al., 2016). But, why was this the case? What were the contributing factors that caused pollsters and news outlets to “get it wrong”? Key factors may have been states such as Pennsylvania, Michigan, and Wisconsin who, in most recent history, voted for democrat candidates at the top of the ticket (Kennedy at el needs year). Could the polling debunked come down to changing trends in registered voters? What role did the younger Generation X and older millennials play?
To find those answers, we must look at how pollsters get their numbers. Traditionally, polling companies would call registered or likely voters and ask them a series of questions about the upcoming election. Telephone numbers could be obtained by purchasing lists from local voter registration offices or published telephone numbers. In the early 2000s, as more Americans began to cut the cord and move towards mobile phones, it became harder for polling companies to contact voters.
National polls verse State polls
Most like to think of the United States as a unified nation; we are all Americans. When it comes to elections, we must remember the focus is on what state we reside. This brings us to the argument of popular vote versus the electoral college. We have to remember in 1787, when the U.S. Constitution was written, the fear of a strong central government and allegiance to one’s state was still prevalent. People thought of themselves as a Virginian, New Yorker, or South Carolinian as much as they did an American. The War for Independence ended in 1783 and the new Nation was governed by the Articles of Confederation. People were not as transient as they are in modern America, and the idea of each state speaking with one voice and casting its votes to one candidate fit with the founding father’s goal to provide equal protection for populations of all sizes. So, why does this matter now? It points to why it is misguided to focus on national polling numbers and national results when state polling numbers paint a clearer picture of the outcome — This is why social media support is not a conclusive indicator of how well a candidate may or may not do.
In the months leading up to the 2016 presidential election, many Americans believed that former U.S. Secretary of State, Hillary Clinton, would win and become the first woman elected President of the United States. The Real Clear Politics average had Clinton ahead 46.8% to 43.6% over Donald Trump. Polling numbers consistently indicated that Clinton would succeed her former boss, Barack Obama, and break the glass ceiling for a woman in politics. The October polls showed Trump’s number falling from an October peak and Clinton’s numbers starting to trend upward. So, why did the polls get it wrong? Many election and polling experts have questioned the effectiveness of polling. They wonder if modern lifestyle changes, such as the decline of landlines, have reduced the effectiveness of the power of polling.
The dramatic switch from landlines to mobile devices leaves politicos asking the question, “is there a better way to poll?” That is why we are asking the question, “can social media service as an indicator of who is going to win the presidency?” But before we ask that question, we must examine the power of social media. According to Per Research, in early 2019, 72% of adults were using social media. Diving deeper, we see that 90% of 18–29-year-olds utilize at least one social media platform, 82% of 30–49-year-olds, 69% of 50–64-year-olds, and 40% of individuals who were 65+ were on social media. With this large percentage of the population on social media, is there a way to use the platforms to indicate who will be elected as the next president? The answer to the question is a difficult one; the question comes down to behavior. Do social media users only like and follow individuals they support, or do they connect with politicians on these platforms as a way of obtaining information?
As the 2020 presidential election was heating up, many Americans look toward Super Tuesday as an indicator, I wanted to further examine who had the largest number of supporters and if that support transferred to the ballot box. I reviewed the Democrat Primary only because President Trump faced few challenges in the Republican nomination process. As I focused on the Democrat’s nomination process, I selected only the top 4 candidates and reviewed their Facebook, YouTube, Twitter, and Instagram accounts. I sampled the number of followers and likes on the Monday before Super Tuesday and then compared these numbers to the primary results in the 14 states that make up the primary day bonanza: Alabama, Arkansas, California, Colorado, Maine, Massachusetts, Minnesota, North Carolina, Oklahoma, Tennessee, Texas, Utah, Vermont, and Virginia. These 14 states represent the largest number of states holding primaries in a one-day period. Super Tuesday is a cross sampling of Americans from the South to the Northeast, from the East to the West. Could the Super Tuesday victors be predicted by simply looking at the candidate’s social media accounts?
Political pundits and politicos look at the 2016 election as a wake-up call — needing a systematic change in the way polling is viewed and understood. To understand the need to examine alternative ways to conduct polling in the modern-day election cycle, which is often dominated by digital advertising and less by traditional forms of advertising, we need to examine the polling process and who is polled. When addressing the 2016 election cycle, one researcher stated that the 2016 election cycle was challenging because of the “nonrandom measurements bias.” To put it more bluntly, people are willing to lie and not tell pollsters whom they will choose. Voters may select to say they don’t know who they are going to vote for or say they will vote for a specific candidate when they have no intention to vote for that candidate.
An additional reason that the 2016 polls may have failed to predict the presidential winner may have been oversampling or sampling populations that did not accurately reflect the U.S. voting population’s makeup. I chose to look at two social media platforms — Instagram and Twitter. These platforms were selected because of the level of the campaign’s activity on the platform and the perceived appeal these platforms had to primary and caucus voters. As can be seen in Chart A, Senator Bernie Sanders of Vermont had the highest number of Instagram followers
at 4.5 million. Some may ask if these numbers correspond to the social media platform’s targeted age group, which may coincide with Senator Sanders’s key demographic. Senator Elizabeth Warner was in a distant second with 2.2 million followers, followed by former Vice President Joe Biden with 1.4 million followers and former New York City Mayor Michael Bloomberg with 429 thousand followers. The stark contrast between Senator Sanders and Mayor Bloomberg of more than 4 million followers, could be an indicator of lack of popularity between Bloomberg and the users of Instagram or Mayor Bloomberg’s late entrance into the presidential primary race. Super Tuesday was Bloomberg’s first serious attempt to make a run for the White House. As shown in Graph 2, Warner had the least number of Super Tuesday votes out of the top 4 candidates. Once again, multiple factors may have contributed to Bloomberg’s low numbers, but the former New York City mayor’s social media ranking did match his overall standing in social media followers. As seen in Graph 5, Senator Sanders had 10.9 million followers on Twitter, and Mayor Bloomberg had 2.7 million. One key factor was the length of time that Senator Sanders had been a national political factor (multiple Presidential campaigns) and the far less time that Mayor Bloomberg had been a national political factor. As for the other key candidates, former Vice President Biden had 4.2 million followers, and Senator Elizabeth Warner had 3.8 million Twitter followers. Graph 3 and Graph 4 indicate that of the 4 major Super Tuesday challengers, Senator Sanders clearly had the largest social media following on the 2 platforms but as demonstrated in Graph 2, Senator Sanders was number 2 in the total number of votes received on Super Tuesday. Sanders won 4 of the 14 states casting a primary or caucus votes. One of those states is the Senator’s home state of Vermont where he had been the U.S. Senator since 2007 and was the U.S. Representative at-large from 1991 to 2007.
Former Vice President Joe Biden won 10 of the 14 Super Tuesday races as shown in Graph 2. A review of Biden’s Twitter and Instagram (Graph 4) shows that the former Vice President had the third-largest number of followers on Instagram, out of the presidential contenders, with 1.4 million and the second largest number of followers a close second, with 4.2 million followers on Instagram. It is important to remember that the Former Vice President was the U.S. Senator from Delaware from 1973 to 2009. Biden also ran for President in 1988 and 2008. His long and extensive political career had not led to increased social media followers, putting his political career into perspective. Individuals born when Biden was first elected to the U.S. Senate are turning 47 years old this year. The average age of Twitter users is 40 years old. The average age of Instagram users is between the ages of 25–34.
The premise of using social media as an indicator of polling had received attention since the 2008 election. In 2008, political campaigns used the social media site Facebook to connect with voters and a way for supporters to show support for their chosen candidate. The general election between Senator Barack Obama and Senator John McCain, in all fairness, was the first general election to see social media become the nation’s town square, a place for people to come together to receive and share information and opinions with social media becoming the citizen’s megaphone. During this time, school media sites were still primarily being accessed by desktop and laptops. The iPhone’s 2007 release had not had enough time to become as pivotal as it was in the 2016 election cycle. The world had not forgone the landline. 2008 was still experiencing the rise of the blog and traditional news outlets like CNN, Fox News, and the major broadcasting networks which continued to dominate the information chain. Why is this background important? The 2008 election cycle saw traditional means of collecting polling information beginning to fail. The 2004 election saw a heavy reliance on conventional means of polling, which was focused on calling a sample of registered or likely voters. Even with the controversy surrounding the 2000 election exit polling, traditional means of polling were more reliable. The election cycle of 2008 was the start of the shift in election forecasting.
This was the first-time researchers examined Twitter’s usage as a gauge in the 2010 United Kingdom elections. The increased level of interest came after the 2008 United States election cycle, in which researchers looked into the usage of the social media giant, Facebook. It was noted that Facebook, and to a more considerable extent, all social media, was the new home of political discussions and social engagement. A review of this research regarding political activity on social media sites indicated that Twitter was the social media space most often viewed platform that elected officials connected with the broader electorate. Today, Twitter has grown into a venue to directly communicate and bypass traditional media outlets, which can skew the original statements and messaging. Elected offices such as Donald Trump and Great Britain’s Boris Johnson are prime examples of political figures bypassing the media by utilizing Twitter.
Twitter as a guidepost of elections.
A study of the 2011 Irish election cycle looked at Twitter as an indicator of forecasting election outcomes. The study by Bermingham and Smeaton used a program called “Twitter Tracker” to examine the connection between the volume of tweets for a candidate or a party and the election results. They found that they could predict the election outcome when using their system of tracking. A similar review was conducted when examining the 2009 German national parliament election. The review process uses linguistic software to analyze the emotional, cognitive, and structural components of the text of the tweet. The tweets were then placed in categories based on the quality of the tweets’ author. The results of that information was connected to leading candidates from Germany’s top 5 political parties. The political parties were reviewed based on the 12 emotions and the volume of tweets per emotion mirrored the outcome of the 2009 national parliament election. In this case, tweets could have been an indication of the election’s outcome. Interestingly, the study found that a large percentage of political discuss is driven by a small number of individuals.
Unlike the original question, “could the social media support replace or supplement polling?”, the German study indicated that the subject and the emotions of the tweets may serve as a better indicator than a candidate’s popularity. The one factor that needs to be remembered is the political leaning of those using the social media platform.
As shown in Graphs 5 and 8, active users of Twitter are more likely to be identified as a Democrat than any other political party. This is a factor that will need to be examined as pollsters continue to develop new procedures to accurately forecast election results.
Additionally, it is important to note, as shown in Graph 6 and 7, a majority of tweets in the U.S. are coming from a small segment of the platforms users. A review of the entire U.S. Twittersphere shows that the top 10% of Tweeters have less than 500 followers and tweet on average 70 tweets a month. Graph 7 shows that the other 90% of Tweeters may only tweet once a month and have less than 100 followers. What does this mean? Those few tweets could place a higher value on the items that they do tweet but this is only speculation.
Data showcasing a relationship between Instagram and polling is still unsubstantiated. Mireille Lalancette and Vincent Raynauld’s research revealed a way to communicate the characteristics that the voter is looking for in their candidates: qualities such as “friendliness, intelligence, honesty, trustworthiness, and sincerity” (Lalancette & Raynauld, 2019). This particular research examined the relationship between Justin Trudeau, Instagram, and Celebrity Politics. Instagram, the second largest social media platform allows the candidate to create and smooth out their image and become relatable to the electorate. Yes, we are a long way from determining the platform’s direct connection to the polls or votes but a 2014 review of Swedish political parties noted that the party’s activity on Instagram afforded organizations the opportunity to communicate with a targeted demographic, manage their public image, and mobilize supporters (Lalancette & Raynauld, 2019). This has shown that Instagram is an important tool for any candidate
So, is social media better than the polls?
As we have witnessed, social media platforms may not provide 100% accuracy in determining which political candidate will win, but can offer valuable insight. A group of researchers recently published their findings that social media users who provide their location can help pollsters and election forecasters determine how specific regions or states may vote. These locations, combined with keywords and phrases, help pollsters identify the rhetoric that supports or opposes political issues, candidates, and political parties. The method is called Opinion-Oriented Word Embedding; this method can provide a higher level of success. One reason for that high level of success may be found in the voluntary nature of the use of social media platforms such as Twitter. Platform users are free to express their thoughts and ideas without the prodding of pollsters via calls or emails. One issue that may harm the credibility of this method may be found with the uncontrolled sample population; those collecting information are limited to individuals that utilize the social media.
Platform and key demographic information such as age, education, and voting likelihood may not be available. However, the researcher can receive real-time data and determine which topics are being discussed and prompting users to engage others on social media platforms.
The answer to the original question is yet to be determined and needs a substantial amount of research. We know the number of individuals who like, subscribe, or follow a candidate’s social media page does not indicate if that candidate will win a primary or general election. When we started, we asked the question, “will the person with the most social medial followers be a predictor or who will be successful on Super Tuesday?” The answer to the question is “no,” the number of followers or subscribers may not correspond to the most successful candidate.
We know there are many ways to examine the content that social media users produce, and the information that is harvested can be very beneficial to political campaigns and their ability to be successful at the ballot box.
· Information for this article was retrieved from the following articles:
Kennedy, C., Blumenthal, M., Clement, S., Clinton, J. D., Durand, C., Franklin, C., . . . Wlexien, C. (2016). An Evaluation of the 2016 Election Polls in the United States. Public Opinion Quarterly, 1–33.
Lalancette, M., & Raynauld, V. (2019). The Power of Political Image: Justin Trudeau, Instragram, and Celebrity Politics. American Behavioral Scientist, 887–924.
Vepsaiainen, T., Li, H., & Sucomi, R. (2017). Facebook Likes and Public Opinion: Predicting the 2015 Finnish parliamentary elections. Government Information Quaterly, 524–532.
Wojcik, S., & Hughes, A. (2019, April 24). Sizing Up Twitter Users. Retrieved from Pew Research Center Internet & Technology : https://www.pewresearch.org/internet/2019/04/24/sizing-up-twitter-users/
