Why did the opinion polls get it so wrong — again?
Most political observers and analysts agree that the opinion polls got things very wrong in their predictions about the outcome of the 2020 presidential election — just as they did in 2016.
One possible explanation given for why the polls failed to accurately predict the outcome of the contest between Donald Trump and Joe Biden is that pollsters struggled to cope because there was so much going on, including how the coronavirus pandemic was affecting voters, and the effects on the result of increased voting by mail and early voting.
If there is one thing all pollsters agree on it is that there are lessons to be learned from every presidential campaign, and failing to heed them will make it difficult for them to survive in the very competitive market for political consulting.
The discussion about how to improve polling methods must recognize that the needs of the American political system are not adequately being met by polls that continue to miss the mark about who will win elections and by how many votes.
Opinion polls are currently not contributing positively to American democracy, and their consistent failure underlines the challenges in a liberal democracy of gauging the political views of citizens.
Following Trump’s defeat of Hillary Clinton in 2016, pollsters admitted the failings that led to their prediction of a Democratic victory. For example, they conceded that they placed too much emphasis on the opinions of people in big cities and major urban areas, and too little on the views of those in small towns and rural areas.
Another mistake was to rely so heavily on the opinions of better-educated people, while voters without college degrees were underrepresented.
In the 2020 presidential race, pollsters vowed that they would not repeat these and other mistakes, and insisted that their sampling included more people from diverse demographic, ethnic and educational/economic backgrounds.
According to W. Joseph Campbell, a professor of communication at American University in Washington D.C. and author of “Lost in a Gallup: Polling Failure in US Presidential Elections,” in the run-up to the 2020 election “the polls overestimated Biden’s lead by 5.8 percentage points.”
Six weeks before the election, CNN published a story that stated: “If you were to look at the polling right now, there’s a pretty clear picture. Biden has leads of somewhere between five and eight points in a number of states Trump won four years ago: Arizona, Michigan, Pennsylvania and Wisconsin. Those, plus the states Hillary Clinton won, get Biden to about 290 electoral votes. If you add on the other states where Biden has at least a nominal edge in the averages (Florida and North Carolina), Biden is above 330 electoral votes.”
These polls were outright wrong in terms of the gaps between the candidates.
It seems, therefore, that those who attempt to analyze and predict trends and results in the American democratic process need to undertake a process of self-examination to avoid further polling mistakes in the future.
There is no guarantee that people will tell the truth when asked for their political views or voting intentions. There are no laws to punish people who tell a caller from a polling agency that they intend to vote for one candidate but then choose another. There is no obligation to stick with an answer given to pollsters.
This seems to have happened in the 2020 presidential election, as Biden’s lead in key states turned out to be slimmer than the polls predicted. The media in general underestimated the level of support for Trump. He was portrayed as a weak, helpless incumbent whose presidency was collapsing due to his personal behavior, and the consequences of his reaction to the coronavirus pandemic.
Yet Trump proved to be more competitive than expected in the election, so it seems likely that a significant number of his supporters did not reveal they would be voting for him when asked.
This might be explained by the fact that some voters were reluctant to reveal their support for Trump because he was portrayed as racist and sexist, and blamed by critics for the deaths of tens of thousands of Americans due his controversial response to the pandemic.
Moreover, polls are long, detailed and ask too many questions. Increasingly, they can be accessed by individuals online. This favors Democratic voters, whose supporters are likely to be more computer literate than Republican voters, who are less well educated on average.
Also, pollsters can overly complicate their canvassing by splitting people into groups such as “registered voters” and “likely to vote.” People do not like complicated questions. Instead, questions should be as simple as “who are you going to vote for as president this year?”
Another common mistake is to ask: “If the election was held today, who would you vote for as president?” This should be changed to: “When the election is held on Nov. 3 who will you vote for?” This is more likely to elicit greater commitment and accuracy from those being polled.
Polling companies always reveal the number of people who were interviewed in a survey, but they never tell us how many declined to answer questions. Providing these numbers can establish a better pool of the potential voters and who they will vote for as president.
In addition, there needs to be a better balancing of polls conducted in Spanish with those in English. An imbalance might explain why Trump fared much better than expected in places with large Hispanic communities, in particular Florida and Texas where the opinion polls suggested the races were too close to call.
Polling firms were more concerned with measuring Trump’s unpopularity, which turned out not to be an accurate reflection of voting intentions.
Regrettably, Afro-Americans are also underrepresented in surveys, and one reason given for this is that many are reluctant to share voting preferences with strangers.
Another factor is pollsters’ obsession with voting intentions in the suburbs, where populations tend to be more liberal. This means that state-wide levels of conservative support can be underestimated.
Finally, polls of people serving in the US military — who tend to favor Republican candidates — are carried out by military publications such as Military Time. Their findings should be incorporated into large national polls to better reflect the intentions of the electorate as a whole.
Polling companies are failing to accurately predict the outcomes of elections because they are focusing too much on measuring general political opinions in the US and not enough on getting accurate indications of how US citizens intend to vote.
This is why the results of polls ahead of the 2020 presidential election ended up more like hypothetical possibilities rather than real voting intention.
Polling firms were more concerned with measuring Trump’s unpopularity, which turned out not to be an accurate reflection of voting intentions. The end result was predictions that were largely lopsided.
- Maria Maalouf is a Lebanese journalist, broadcaster, publisher and writer. She holds an MA in political sociology from the University of Lyon. Twitter: @bilarakib