Leading up to the 2020 presidential election, many major pollsters predicted a landslide victory for democratic candidate, Joe Biden. Yet, days after Nov. 3 the vote is still too close to call, and countless states performed far differently than popular polls predicted.
Regardless of who wins, it is clear from the past two election cycles that the current polling techniques face serious challenges – namely, they are not providing accurate estimates. This is not merely the case within the United States, but also in the UK where polls before the 2019 general election showed support of the Labour Party, while Conservatives won a majority.
What’s wrong with today’s polling?
The challenges with polling results have been widely discussed, but few can agree on what the actual problem is. Given my deep background in the data science and measurement field, here are my thoughts with the shortcomings of our current polling methods.
One challenge with current polling methodology is the continued reliance on the good ol’ telephone. With the advent of caller ID, more and more people are not picking up calls from random strangers. A 2018 Pew survey showed that telephone survey response rates dropped from 9% in 2016 to an abysmal 6% in 2018. I wouldn’t be surprised if it has fallen even further over the past two years. Let me ask you: do you answer phone surveys, particularly during election season?
Think about what you are missing out on with that 6% response number – that’s 94 out of 100 individuals not sharing their preference. No matter how sophisticated your statistical algorithms are, it will be hard to make up for the missing sentiment.
Another challenge with current polling is the reliance on personal disclosure. Many individuals are reluctant to share their political preferences due to increasing divisiveness and incivility. After the 2016 election, the theory emerged that many Trump or conservative voters simply didn’t want to share their opinions (e.g., “the shy Trump voter”). That concept continues to be debated among statisticians like Nate Silver, but I would argue that the aggressively combative discourse around politics online and in person makes votes on both sides on the aisle more hesitant to share their true options. Of course, this reduces the informativeness of polls and can lead to increased uncertainty in the estimates.
A third significant polling challenge is that poll responses are merely snapshots in time and cannot calculate actual sentiment come election day. This is especially the case for undecided voters. The percentages of swing voters appear to vary with each election cycle, but even with small percentages, they can matter tremendously when margins are slim, as in the case of our current presidential election.
My proposed solution
I know we cannot solve polling in a day, but I do have thoughts for how to right this critical ship. The three challenges I outlined above - reliance on phone outreach, reluctance to share opinions, and inaccurate snapshots in time - can all be resolved with modern data capture methodology.
Our pollsters need to shift their approach in several ways:
Technology platforms like ExpiWell, which I created initially to improve academic research, offer these types of capabilities by focusing on ongoing digital and mobile capture of opinions, surveying people anonymously over time. Importantly, ExpiWell allows data capture within curated online communities in which people can anonymously discuss their thoughts over the ExpiWell app, and researchers can leverage qualitative data to augment quantitative numbers. Researchers and pollsters can also provide incentives directly through the platform.
America and the world are in dire need of accurate polling methods, particularly for critical elections.
With newly emerging technology like ExpiWell, I am hopeful that we can enhance our current polling technology and practices and ultimately produce results we can rely on once more.