We’re always excited to work with tremendous pollsters like Ruth Bernstein at EMC Research, Anna Greenberg at GQR and Daniel Gotoff at Lake Research Partners. We asked them if they had any assessments and conclusions about the 2020 polling cycle. Guess what? They did.
Anna Greenberg, Ph.D. serves as managing partner of GQR. She has over 20 years of experience polling in the political, non-profit and academic sectors. In 2014, Greenberg won the “Pollster of the Year” award from the AAPC for her work with Mayor Bill de Blasio’s campaign for mayor of New York City.
She said about 2020:
“We learned that polling error 2016 was not an aberration, but in fact, a reflection of the challenges of polling in the age of Trump. There are clearly groups of voters who do not take surveys, either because they don’t want to or because we do not have the right tools to reach them. We will need to work hard in 2022 to figure out how to address non-response bias that produced much closer races that pre-election polling suggested.
It is also the case that we do not invest nearly enough in reaching harder to reach respondents. Typical congressional surveys rarely oversample Hispanic respondents or include enough cell phone interviews to adequately measure response among younger and voters of color. This lack of investment may partially explain why polling missed the erosion of support for Democrats among minority voters.
Finally, political polling is a strategic, not a predictive tool. It is only as good as its assumptions about who is going to turn out to vote. Campaigns should continue to feel confident in using diverse research tools (including online polling, text-to-web as well as traditional live surveys) to figure out message and strategy. We should also integrate more qualitative research into campaigns and use non-polling information to help us assess the mood and possible turn out dynamics (e.g., like patterns in early vote in primaries).”
Ruth Bernstein is President and CEO of EMC Research, which she joined in 1994. Ruth is recognized as one of the top pollsters in the field for her ability to keep the research focused on her clients’ goals. She takes the time to understand the overall objectives and provide hands-on practical advice in addition to quality research. Her clients trust her judgment and know her research findings are reliable and will provide a strategic roadmap for success.
Her takeaways about what we learned in 2020 follows:
“Mixed mode methodology! To get a truly representative random sample of likely voters in most communities, surveys will require a mix of methodologies to meet voters where they are. Sampling will likely include a mix of live telephone, text-to-web and email-to-web.
Not all “online” surveys are the same. We must pay careful attention to sampling and how we visually present questions.
The way people vote (early, in person, by mail) along with the significant increase in phone/text/email communication from campaigns has made late polling extremely challenging.
List quality is crucial!”
Joining Lake Research Partners in 1996, Daniel Gotoff has worked for candidates at all levels of the electoral process and on a wide range of issues, including the economy, national security, and government accountability. Sandra Markowitz has been conducting research since 2016, focusing on political campaigns at Lake Research Partners. Previously, McCauley Pugh worked for The Mellman Group and was an intern for US Senator Maria Cantwell.
Their joint take on what we learned in 2020 is this:
“For public polling, there needs to be less emphasis on the horse race itself, and a larger conversation about things like the margin of error and what the numbers presented truly represent. Pollsters also need to continue the conversations they have been exploring about turnout prediction methods, and polling methods in general. Live caller surveys are still the most effective method to reach some demographics, but not all. Online and mixed-mode surveys can reach groups that are otherwise difficult to include in our calling. Despite the relatively modest level of error in 2020, restoring credibility with the public is also important, if for no other reason so that pollsters can once again have greater faith in the quality of their research. But the solution to these bigger questions is not an end to polling, but rather the continued adaptation of an important tool.
Moving forward, it will be more important than ever to devise and report out on multiple turnout models of what the electorate could ultimately resemble on Election Day, rather than settling at the beginning on a single estimate of the likely electorate.
We need to look at who participated and who did not, who terminated part-way through the interviews (where? and why?) and comparing these and other data points to what we know about voters’ partisan voter registration, and/or other such proxies that are appended to voter files(s), the which serve as the provenance of our samples. Additionally, avoiding the types of bias outlined above will require devising new and better ways to account for systemic misses in our interviewing strategies as well as our adjustments of the data (i.e., weighting) once we are out of the field. “
A big THANK YOU to Anna, Ruth and Team Lake Research for sharing these lessons from 2020. I know at CN4, we relish the opportunity to evolve and learn from these lessons. We believe that to provide the best services to our clients and have the best shot to win campaigns, we have to always push forward. Onward to 2021, 2022 and beyond.
– Dean Nielsen and the CN4 Team