A prolonged lead in the polls for the Democratic contender. A projected 90% chance of winning. And then the counting of the votes, with the only certainty being a thin margin for whoever is declared the winner.
As it was four years ago, pollsters are being blamed for a misleading pre-election narrative, even after applying the lessons they learned from 2016: weighting to education in order to capture the non-college-educated white vote and investing more in statewide polling, particularly in battleground states, to provide a better assessment of the Electoral College.
While votes are still being counted, it is premature to provide definitive answers as to whether, why and how something went wrong with the 2020 presidential election polls. But before lampooning the polling industry, we can review what the pollsters were up against during this election cycle, as well as what polls can and cannot do, and how they should be used.
Even in normal election cycles, it is a challenge for pollsters to estimate the electorate. Polls typically sample known populations, but in pre-election polling, the population of voters likely to turn out for any given election is inherently unknown. Any pollster’s likely voter model is an educated guess based on variables like the composition of prior electorates, and individuals’ past voting history, self-professed likelihood of voting and interest in politics.
While pollsters are modeling a future population, they are not predicting future behavior but rather assessing current responses. In reality, some respondents who tell a pollster that they will vote do not ultimately do so come Election Day, and vice versa.
This election added the unpredictability of a pandemic, which brought about new and various voting methods across the country. This made for an even more complex electorate for pollsters to model, given the partisan division over voting methods, the learning curve for many regarding how to vote early or by mail and the sheer volume of those who did so.
We may also once again find the 2020 polls erred in undercounting Trump voters. Given how accurate the 2018 polls were, and how polling has been fairly accurate over decades and across multiple countries, there is some discussion as to whether a large driver of unpredictability in 2016 and 2020 was Donald Trump himself, the common factor between both races. Studies have thus far shown no response bias between those who participate in surveys and those who do not, but underestimating Trump supporters for a second time brings back into question the possibilities of social desirability and response bias.
In addition, we pollsters need to do even more to educate the public about the probabilistic nature of polling, as embodied by the margin of error and other indicators. The press and the public in turn need to better understand and embrace this statistical uncertainty.
Pre-election poll results are often treated as absolutes, a binary representation of who is winning or losing. But these results are simply estimates that represent a range of possible outcomes affected by the sample size of a poll, the numerical difference between the candidates being assessed, and the margin of error – the price we pay for selecting only a sample due to the impracticalities of interviewing everyone.
Our room for error furthermore grows as we drill down into subgroups within a sample. For example, in Florida, the overall sample sizes of various pre-election polls may not have adequately or accurately captured smaller groups like Cuban Americans in Miami-Dade County, who were crucial to Trump’s win.
Nationwide, the polling industry was already suffering from declining response rates, often attributed to the advent of smartphones, caller ID and an exponential increase in spam calls. But response rates have declined even further since 2016, driven at least in part by a hyperpartisan lack of trust in polling. This distrust may be insidiously dismantling one of the most vital tools we have for connecting the people to their policymakers.
Polls are not meant to be predictive but rather explanatory tools, especially for those important issues and policies that are vital to our democracy yet never see their day at a ballot box.
In its purest form, polling is the only participatory act in our democracy that requires few resources of the individual besides a few minutes of their time and, in turn, gives them a voice proportionate to their representation in the population. Few other forms of political participation can claim this.
So let’s put a pause on polling’s obituary. Polling is a science, so let’s analyze the election’s final results and assess where the polls may have gone astray; by definition, the continual assessment, dismantling, and rebuilding of scientific theory will only improve it.
Local journalism is essential.
Give directly to The Spokesman-Review's Northwest Passages community forums series -- which helps to offset the costs of several reporter and editor positions at the newspaper -- by using the easy options below. Gifts processed in this system are not tax deductible, but are predominately used to help meet the local financial requirements needed to receive national matching-grant funds.
Subscribe to the Coronavirus newsletter
Get the day’s latest Coronavirus news delivered to your inbox by subscribing to our newsletter.