Kerry Boyd Anderson
In the wake of the US elections in November, pollsters have been reviewing what went wrong with many of their predictions and what it means for the future of polling.
After the 2016 presidential election, there already were accusations that polling was proving inaccurate. The criticisms often exaggerated the errors, but there were certainly some problems in a few states and with specific demographic groups. Many pollsters worked hard to correct those errors in the hope of providing more-accurate forecasts this year.
Instead, the 2020 election raised further profound problems for polling companies. The polls this year were not significantly less accurate than the ones in 2016, especially at the national level. However, the adjustments pollsters had made since 2016 failed to produce better results, which has led to more urgent questions being asked about the value of polling in the future.
On the national level the 2020 polls were off by about four percentage points, in Joe Biden’s favor — this is not a major error, historically, in terms of national polls. However, the US president is not chosen by the national popular vote but by votes allocated by individual states in the Electoral College, and so polls conducted in the key battleground states that determine the outcome are particularly important.
In some of these states, such as Arizona and Nevada, polling tended to be accurate. However in others, such as Iowa, Florida, Wisconsin and Ohio, many of the polls missed the mark significantly.
This led to much soul-searching by pollsters and the political media. There are always challenges involved in polling: members of some demographic groups are more likely to respond than others, for example, and polls always lag somewhat behind election day. Pollsters have developed tools to manage those challenges, such as weighting the data for underrepresented groups and providing transparency about a poll’s likely margin of error.
The 2020 electorate presented challenges that went beyond the usual tools pollsters use to tackle them, however. The analysis is ongoing but some trends are already clear. In particular, many polls failed to sufficiently reflect the voting intentions of Latino and rural voters, and polls struggled in several Midwestern states.
There are several potential reasons for the polling mistakes this year. A fundamental challenge that will extend far beyond this year is technological changes. Decades ago, pollsters knocked on doors to get their answers. Then they used landline telephones, which was an effective way of gaining random samples in specific areas.
The proliferation of cell phones created significant problems for pollsters, including the ease with which people can take their cell phone, and their exiting number, with them when they move. In addition, pollsters must now compete with campaign fundraisers, telemarketers and other callers — many people are weary of all these “cold calls” and so much less likely to answer and respond to a pollster’s call than used to be the case. Some pollsters have employed digital tools, which also present multiple methodological concerns.
The 2020 electorate presented challenges that went beyond the usual tools pollsters use to tackle them
Another potential problem is that Democrats might be more likely to participate in polls than Republicans. There are several possible reasons why this might be true, including the fact that Donald Trump has frequently portrayed polls and the media as untrustworthy. Polling firms work hard to try to ensure that their samples accurately represent Democrats and Republicans, but they may have failed to do so. It is also possible that the Republicans who did respond to polls perhaps were less likely to strongly support Trump than those who did not.
A constant challenge pollsters face is determining who are “likely voters.” Polls stop before election day, so they try to identify who is most likely to actually show up and cast a ballot. This may have been particularly challenging this year.
Another theory is that the polling challenges in 2020 were specific to the pandemic. Many states allowed more mail-in ballots and early voting than normal, because of public-health concerns. Democrats strongly advocated for mail-in voting and early voting, and so Democratic voters were more likely to use those methods than Republicans.
Trump was openly disdainful of mail-in voting, and so Republicans were more likely than Democrats to vote in person on election day. This complicated the ability of pollsters to identify likely voters, and may have favored Democratic turnout in their polling forecasts.
Some of these problems might be fixable. Polling companies might be able to make adjustments. Some are certainly considering new ways to weight populations. However, some of the challenges might fundamentally alter the accuracy of polls.
Furthermore, modern-day US politics might simply be too competitive for polling. If presidential elections frequently come down to a couple of percentage points or less in key states, or congressional races are often tight, they might fall within unavoidable margins of error in polls.
For example, two Senate races in Georgia next month will determine which political party controls the Senate — but polls offer limited insights into the Jan. 5 runoff elections, because the likely outcomes are so close they are expected to be well within the margin of error for most surveys.
Polls will probably continue to have some value in the future, however. They help inform media narratives about politics, for example. They can also identify significant changes in the electorate. In a large, diverse country, polls provide data on what the population is thinking. There is no reliable, alternative source of such information.
Nonetheless, the media and political strategists might need to reduce their reliance on polls, particularly when analyzing tight races. Polls still have an important place in journalism and campaign strategy — however, reports based on them need to clearly communicate their limitations and should include other forms of political journalism.
Campaign strategists need to rethink the extent to which they use polls to guide them on where they focus resources. Meanwhile, pollsters must consider what they can fix and what limitations they must accept.
The writer is a political risk consultant with more than 16 years of experience as a professional analyst of international security issues and Middle East political and business risk.