Does lopsided polling make voters complacent? Probably not, according to my research.
In a quantitative lab experiment, I found that exposing people to varying polling numbers about a hypothetical election did not significantly affect their motivation to vote.
With Election Day in the United States only two weeks away, we have seen people become increasingly more nervous about the effect of political polling on voter turnout. Former Vice President Joe Biden, the Democratic nominee, has maintained a consistent lead in national and swing state polls over President Donald Trump, the Republican nominee. Biden is up by approximately eight points nationally in the Real Clear Politics polling average. The FiveThirtyEight presidential forecast gives Biden an 87 percent chance of winning the Electoral College. The Economist presidential forecast gives Biden a 91 percent chance of winning the Electoral College.
Democratic voters and party elites have expressed fears that Biden’s wide polling advantage will lead to complacency and depress turnout among left-leaning voters. These fears have led prominent political figures on the Left to urge people on social media to “ignore the polls” in the weeks before the election:
Much of the fear among Democrats about the negative effect of lopsided polling on voter turnout stems from 2016, when Democratic nominee Hillary Clinton lost the Electoral College to Donald Trump despite being ahead in most polls in the weeks before the election. She was also ahead in many probabilistic forecasts, which average together multiple polls of a political race to come up with a probability of victory for the respective candidates. Clinton had a 42.8 percent probabilistic lead over Trump in the FiveThirtyEight forecast and a 70 percent probabilistic lead over Trump in the New York Times forecast on Election Day 2016 (November 8). This lopsided polling led many to believe that a Hillary Clinton victory was inevitable when it was not.
The idea that Clinton was heavily favored to win spilled over into media and popular culture. In October of 2016, the New York Times published an article titled “Hillary Clinton Presses Her Advantage Over a Struggling Donald Trump.” The Saturday Night Live parody of the 2nd 2016 presidential debate jokingly introduced Trump and Clinton (portrayed by Alec Baldwin and Kate McKinnon, respectively) as “Donald Trump and… can we say this yet? President Hillary Clinton.”
In a 2017 interview, Clinton lamented that “I don’t know how we’ll ever calculate how many people thought it was in the bag, because the percentages kept being thrown at people — ‘Oh, she has an 88 percent chance to win!’” Clinton, along with many others, may have suspected that 2016 presidential polling depressed turnout among those who thought that Clinton was so far ahead in the race that she would be able to win comfortably even if they did not go to the polls. These individuals may have concluded that it was not worth their time to fill out a ballot because the polls showed that a Clinton victory was a sure thing.
There is some evidence that polling may have an effect on vote intention. An experiment done by Sean Jeremy Westwood, Solomon Messing, and Yphtach Lelkes published in The Journal of Politics in 2017 showed that exposing people to probabilities depicting one candidate as highly likely to win over the other made people less motivated to vote in a hypothetical election in comparison to exposing people to probabilities showing the candidates as being neck-in-neck. These results suggest that it is possible that some people who looked at news coverage and forecasts focusing on Clinton’s wide polling lead over Trump in 2016 may have been less motivated to go out and vote for Clinton than if they had thought the race was closer.
The postmortems about the effect that statistical projections had on voting in the 2016 election inspired my honors undergraduate thesis project. Using a grant provided by my university, I conducted a quantitative lab experiment examining whether exposure to probabilistic forecasts in news coverage would significantly affect vote intention.
For my experiment, I used Amazon’s Mechanical Turk Program to expose approximately 2,800 individuals to news articles about a fictitious mayoral race in a mid-sized Wisconsin town*. Subjects were randomly assigned to a control group or one of several experimental groups. In the experimental groups, individuals were exposed to a probabilistic forecast about two mayoral candidates named James Jones and Tom Smith. The first group of subjects was exposed to a mock news article with a forecast showing James Jones as having a two percent probabilistic advantage over Tom Smith (a 51–49 percent chance of winning, respectively). The second group of subjects was exposed to a mock news article showing James Jones as having a 70 percent probabilistic advantage over Tom Smith (an 85–15 percent chance of winning, respectively). After they read the article, I asked subjects to answer a question about how inclined they would be to cast a vote in this race on a scale of 1 to 5, with 1 being extremely unlikely and 5 being extremely likely.
When I analyzed my results, I found no statistically significant difference between the first group and the second group. Varying the probabilistic forecast numbers did not significantly motivate or demotivate my subjects— those who were inclined to vote (or not vote) when the election was close were generally also inclined to vote (or not vote) when the election was lopsided probabilistically.
What did have a significant effect on intention to vote in my experiment was policy stakes — the group of subjects who were told that, if elected, candidate James Jones was going to fix their roads and candidate Tom Smith was not, were significantly more likely to express an intention to vote than subjects who were told that Jones and Smith largely agreed on infrastructure issues. Policy differences mattered more than statistical viability in determining whether my subjects wanted to vote.
There’s no way of knowing for certain why varying polling numbers did not matter to my subjects while varying the policy stakes of the election did. One possibility is that most of the subjects in my experiment realized that no matter how “close” a political race is, it is virtually impossible for a single vote to sway the outcome of an election. It seems that my subjects generally did not use the polling information I provided as their barometer for whether their vote would “count” or not. Instead, they expressed an intent to vote because they wanted to show that they were committed to standing up for their beliefs, making their voice heard at the ballot box, and doing their civic duty.
The conditions of my experiment are not identical to the conditions of a presidential race. Additionally, my experiment tested whether people felt more or less inclined to vote in a hypothetical election, not whether they cast a vote in an actual election. However, some key principles I drew from my results may carry over to the “real world”. In the short-term, my results indicate that fears or hopes that media coverage of Joe Biden’s formidable polling numbers will significantly dampen or increase Democratic enthusiasm are probably misplaced. In the long-term, my results indicate that “horse race” journalistic coverage focusing on polls and forecasts are unlikely to significantly improve or worsen the United States’ problem of low voter turnout.
However, policy and consequence-based journalism just might. If journalists want to play a role in improving voter turnout levels, my results indicate that they should spend more time covering candidates’ stances on the issues and the policy consequences of a particular election outcome and less time on who’s up and who’s down in the latest poll or forecast.
Democratic activists on Twitter and in real life should probably spend less time worrying about whether lopsided polling in the 2020 election will create complacency among a significant share of voters. Issues seem to matter more than polls in determining whether people feel motivated to make their voice heard through their vote.
You can read my full study published in The George Washington University Undergraduate Review here.
*The town I used for my experiment was Kenosha, Wisconsin. This town was picked randomly in the summer of 2019, almost a year before the shooting of Jacob Blake in Kenosha took place. Therefore, there is no way that the Blake shooting specifically could have affected the answers of my respondents.
