Top Stories

The most likely polling errors in 2020 and how journalists can avoid being misled by them

A deluge of polling data threatens to overwhelm reporters in the run-up to Election Day.

We reached out to two polling experts to help guide journalists in navigating the numbers. Alan I. Abramowitz is the Alben W. Barkley Professor of Political Science at Emory University and Courtney Kennedy is the director of survey research at Pew Research Center.

Alan I. Abramowitz is the Alben W. Barkley Professor of Political Science at Emory University

Can you share some tips for finding hidden insights from polling information?

Abramowitz: Right now I think that the problem that anyone is having is that they’re trying to discern what’s going on with the polls. We’re being swamped by just the vast number of polls, both national and in a lot of the swing states. So, the advice I give in general for any time — but especially now — is that you should go to the polling aggregation websites and look at the averages. Usually the averages are going to be more accurate in the end. 

Kennedy: This year the share of registered voters who are leaning to a third party candidate or who are undecided is significantly lower than in 2016 (about 6% this year versus 15% in 2016). This suggests that more voters this cycle are committed in their vote.

Courtney Kennedy is the director of survey research at Pew Research Center.

What are the polls not telling us and how can journalists fill in these gaps?

Kennedy: Polls often cannot reach robust samples of groups like older Latinos or young Black Americans. Journalists can help fill that gap through some in-person reporting.

What do journalists need to know about misleading poll figures right now? 

Kennedy: A large national lead may be misleading because the Electoral College is still very competitive (even with some small to modest apparent Biden leads in a few states). Focusing on a candidate’s lead nationally can, thus, be misleading when the real contest is still very competitive. Similarly, discussion of a landslide outcome is premature and could affect the likelihood that some people will vote.

Abramowitz: I would just say that you should discount polls that I’d call outliers.

A lot of reporters in 2016 were misled by polling data. How can journalists avoid making the same mistakes in their coverage today?

Abramowitz: It’s pretty hard to know in advance if there’s going to be any kind of systematic error in the polls. The problem we had in 2016 is that, especially in some of the swing states … we saw that there was a consistent underestimate of Trump’s votes here. And subsequently, there were studies that found some of the reasons for that. It had to do with late vote changes; it had to do with underestimating turnout among non-college white voters in states like Wisconsin, Pennsylvania and Michigan. So, some of the polling organizations are trying to correct that. One of the things we’re doing now that we were not doing then is weighing by education to ensure that you get the right proportions — approximately the right proportions of college and non college voters. They probably won’t make the same mistakes or have the same errors they had four years ago, but you might have different errors. 

Eight years ago in 2012, the polls in the swing states — and nationally — underestimated Obama’s margin. Four years ago, they underestimated Trump’s margin. So, there’s no way of knowing in advance what the errors might be, so I would just say, take it with a grain of salt and understand that there will almost certainly be polling errors. The polls, on average, still do pretty well. … There are a lot of challenges to doing polls right now, with low response rates being a big one.

In the 2018 midterm elections, the polls were pretty good in terms of looking at some of the key races and looking at the overall outlook for that for the congressional elections.

Kennedy: It’s important to always remember that polls are a snapshot in time (not predictions) and they always have a margin of error, even when averaged together. Also, be mindful that basically anyone can release a poll result these days, so reporters should carefully vet polls and only report on polls conducted by established, reputable firms that have a track-record of high quality, nonpartisan polling. I would discourage reporters from focusing on forecasts/predictions and instead encourage them to focus on what issues are motivating voters and how are voters planning to cast their ballot this year.

What are the best resources for journalists who are new to covering election polls?

Abramowitz: Nate Silver‘s website FiveThirtyEight does a good job of putting polls together into averages and they’ve weighted the polls according to their own estimates of the quality of the individual polls, the sample size, any kind of known partisan lean that certain polls have on it, and they try to filter that out. So, it’s not perfect, but I think if you follow that, you can look at both where the national average is and you can look at where the averages are from day-to-day for some of the states. 


Notify of
Inline Feedbacks
View all comments