Political polls face significant challenges, but they still hold value

Should Americans take any interest in political polls?

High-profile failures to predict election outcomes in recent years have given citizens pause when it comes to heeding poll results — but polling expert Christian Grose says the numbers still deserve attention.

“Polling is really important from an electoral perspective for a campaign or even for voters,” he says, because campaigns make strategic decisions based on polling, and voters might decide to get involved if the election is close.

As professor of political science and international relations and public policy at the USC Dornsife College of Letters, Arts and Sciences, Grose recently collaborated on the California Elections and Policy Polls. The joint effort between USC, Cal Poly Pomona and California State University, Long Beach, aimed to gauge outcomes among California races for key congressional seats and for governor.

Grose acknowledges political polls like these face a number of challenges, including getting enough people to participate.

“The biggest challenge right now is getting high enough response,” he says. “It’s really hard to get people to take surveys,” and response rates to poll participation requests have dropped precipitously, from 25% to 3% or less, he says.

Further, pollsters also want to be sure participants represent those who will actually vote during the election. “It’s sometimes really challenging to know exactly who’s going to turn out to vote,” he explains.

Further complicating matters, different groups are often disproportionately represented in polls due to differing interest in participating. For example, voters of color are less likely to participate in a political poll than white voters. The same is true for younger voters versus older.

Grose says these challenges can be addressed in a number of ways. To get enough participation, pollsters send invitations to far more people than they need. “For a survey of a thousand people, you often have to ask 10,000, sometimes 100,000 people to take your survey,” he says.

Incentives can also help. “I do that in some of my own work, incentivizing people by giving them small payments or a chance to enter a lottery if they take the survey,” Grose says.

To ensure groups with lower turnout for polls are properly accounted for, pollsters can “weight” responses from participants in those groups more than participants who turn out in higher numbers. “That’s relatively easy to solve, … and you end up getting an accurate count,” he says.

“Another solution is to rely on what is called a panel, where voters who are representative of the broader public take the survey and changes can be tracked within that representative group of survey takers,” he added. (The Understanding America Study administered by USC Dornsife’s Center for Economic and Social Research is an example of such a panel.)

Grose says that the public’s waning trust in polls may also be due to a lack of understanding of how to interpret the poll results. He points to the margin of error as particularly confounding for some. “Understanding the concept of margin of error — understanding uncertainty around the numbers that come out of a poll — is really important.”

For example, if a candidate is ahead in a poll by two points, but the margin of error is five points, then in effect the candidates are tied.

Grose also cautions people against focusing on only those polls they like. “Don’t just cherry pick one poll that happens to find a result for the candidate you prefer.” It’s better to look at a variety of polls to see if they are trending in any direction.

Finally, Grose says that even a poll that turns out wrong in the end wasn’t necessarily in error when it was published, nor was it rigged, as some have claimed.

“It’s a snapshot in time. So, like right now, when the poll is being conducted, and if it’s been done using scientific practices and appropriate methodological practices, that’s what the result is,” he says.

But people can change their minds later, or events can sway opinions after the poll, he adds. “A good poll is only accurate the day it is taken.”