Steve Mossop on political polling accuracy and best practices

Steve Mossop, executive vice president of Leger, recently joined our market research podcast. He leads a team of 25 plus researchers serving public and private sector clients with full-service market research solutions. Leger is Canada's largest and best known public opinion polling firm. Steve has impressively correctly predicted the outcomes of 25 elections over his career. During the podcast, Steve shared with us the importance of election polling, why we do it and what the pitfalls and upsides are in the polling world. 

He starts out by sharing a little bit of history - back to telephone and door to door interviews - as well as how he became interested in public opinion polling from working on the “Coke vs. Pepsi” blind taste test wars. He said, “Coca Cola made the biggest mistake ever by changing their formulation without any other consideration than the blind taste testing. And that inspired me over the years as a story to repeat because they didn't ask the right question.” He said people aren’t just influenced by “taste” but also by the marketing and packaging around it. He transferred this knowledge to political polling by examining politics with “brand” in mind. 

Steve says companies that do election polling accurately can use this as a litmus test for the other work that they do. “If you can predict, predict elections accurately, then certainly you can predict things like market share, product selection and public opinion on other things - all those variables that we stand behind as a market research industry.” He also shares that secondary research has shown that election polls are extremely accurate. He speaks with us about some examples of research he is working on now, past case studies, plus important questions and methods to set up for political polling, to illustrate his point. 

Steve shares a story about some polls that were conducted in Canada, during which every pollster got the election results wrong. Leger found during an exit poll that most people decided on how they would cast their vote only 24 hours before they voted. While this is unusual, Steve likens it back to the Coca Cola story and re-emphasizes the importance of asking the right questions. He then talks about some examples in the United States and how a smaller segment of underrepresented voters can sway the vote and panels need to account for that reality. 

Some of the biggest lessons Steve has learned is that last minute changes in voter sentiment and fringe voter segments can overturn mainstream election results. He gives some tips on how to fill in the gaps here to create more accurate outcomes, including panel quality - something behind the recent failure of pollsters to accurately predict recent elections. While there is no single solution to this problem, employing a mix of methodologies and knowing where panel members are coming from can help. Quality control is at the top of Steve’s list when it comes to improving polling outcomes, plus having experts internally who understand political nuances. 

Also during our conversation, Steve emphasized the importance of monitoring competitors and staying updated on industry changes, as this allows companies to learn from each other and collectively improve the polling industry's standards. The presence of peer pressure is identified as a powerful motivator for bettering the industry, as all polling companies are under the watchful eye of their peers, striving to maintain credibility and accuracy in their predictions. He speaks about Leger’s strengths in the space, and some best practices.

No discussion would be complete these days without touching on the impact of artificial intelligence. We explore the positive impact of AI in polling operations as it can aid in various aspects of the business, such as coding, questionnaire design, and data analysis, leading to significant advancements in recent months. However, Steve acknowledges that the rise of AI also brings challenges, particularly in maintaining panel quality. The growing sophistication of AI-powered bots poses a risk, making it increasingly difficult to detect fraudulent responses manually. This raises concerns about the industry's ability to stay ahead of such challenges and maintain the representativeness and accuracy of their panels, plus the potential loss of human control over the decision-making process. 

We conclude by going over all the variables that feed into accurate polling processes (there are a lot!) and some of Leger’s successes.