In 2016 many of the nation’s top pollsters had Hillary Clinton several points ahead of Donald Trump right up until election day. Then, everything turned upside down for these pollsters when Trump shockingly beat Clinton. Many politicians, journalists, and voters asked,” What happened?” and “Can this happen again?” In nine “swing states” many pollsters predicted a better performance for Clinton than what she had. Much criticism was directed at polling organizations which had them scrambling to come up with answers to what happened as well as what voters could anticipate in future elections. Among the many published theories was that there were a number of people who would not admit to favoring Trump yet voted for him. In all actuality, the polls were not that far off as many had Clinton ahead of Trump and she did, in fact win the popular vote. Still, public skepticism of the polling process was greatly heightened going into the 2020 election. We all know what ended up happening, but why and what can we as an industry do about it going forward?
In 2020, early polling pointed to a win for Joe Biden, predicting that he would pick up several states that Clinton had lost in 2016. While that held true the election was again much closer than many polls predicted. Questions resurfaced regarding the legitimacy of our industry’s polls. In many states, polls were very near the mark however according to data from FiveThirtyEight and Pew Research, some states were “not just off, but off in the same direction” while favoring Biden.
One issue might be the “Pandemic Effect” on the election. A disproportionate number of Democrats were concerned about the virus and used early voting by mail. It is possible, as suggested by Pew, that some Democrats who thought they had cast a ballot actually did not - filling out their ballot but forgetting to actually mail it.
Current data analysis also suggests that there were several national and state polling errors. It is interesting to note that polling errors uniformly underestimated Republican candidates’ performance. How could this happen yet again? According to Pew Research “Partisan Nonresponse” played a key factor as Democratic voters were more willing that Republican voters to voice their opinion by responding to polls. Some data suggests that the most loyal Trump supporters were underrepresented in the polling data. Another theory is that the Republicans lack of trust in the news media, particularly those that sponsor some of the polls, led to non-responsiveness of Trump supporters.
In fact, a friend told me that he had received several calls inviting him to participate in a political poll. He refused them all based on the belief that “The polls are wrong anyway.” Assuming that other voters feel the same, this is very concerning and has ramifications for all surveys that measure behavior.
How do we address these perceptions and resolve the methodology issues moving forward? As always, sampling is of utmost importance. In the 2020 instance, capturing the opinions of loyal Republicans would certainly have yielded more precise numbers. But how do you do that? Pew, Gallup and many other organizations suggest increasing efforts to include those unrepresented in the past two presidential elections.
We at Ironwood believe there may be some potential solutions to this issue:
Design the questions in the survey to be able to identify far-right or far-left voters from moderate voters.
Address perceptions by being more transparent as an industry in how polling surveys are conducted and the measures taken to ensure quality.
In the end, market researchers understand that polling is not an exact science. Unfortunately, the general public does not see it this way. Nevertheless, research organizations such as AAPOR (American Association of Public Opinion Research) are devoting a great deal of time in understanding what happened and investigating ways to ensure researchers are gathering accurate, representative survey data. This is critical not just during a presidential election year, but across all market research. We as researchers need to be able to adapt real time to the ever-changing profile of the U.S. voter. Otherwise, study designs and sampling may not yield the projectability necessary to accurately predict outcomes.
For more information, contact Ironwood Insights Group today at info@ironwoodinsights.com or call 801-569-0107.