How polls got the primaries wrong

March 9, 2016
Senator Bernie Sanders thrusts his fists in the air as he speaks to supporters on the night of the Michigan, Mississippi and other primaries  on March 8, 2016.    REUTERS/Carlo Allegri

Senator Bernie Sanders thrusts his fists in the air as he speaks to supporters on the night of the Michigan, Mississippi and other primaries on March 8, 2016. REUTERS/Carlo Allegri

 

It’s been a tough few days for pollsters. First, they missed Senator Ted Cruz’s surging popularity in the Louisiana primary. Then they were totally wrong in the Michigan Democratic race, consistently predicting a double-digit margin of victory for former Secretary of State Hillary Clinton, not U.S. Senator Bernie Sanders of Vermont, who narrowly won Tuesday night.

This week, no doubt, anyone who’s in the business of predicting races is busy double-checking their methods.

As the U.S. political polling editor for Reuters, I should say at this point that the Reuters/Ipsos poll doesn’t predict state primaries, in part for the simple reason that they’re really hard to get right. Turnout is usually low at state nomination contests, and it’s tough to determine who’s going to show up. In Michigan’s case, for example, it’s possible that polls of “likely voters” focused too heavily on politically active Democrats and not enough on new voters who were inspired by Sanders.

That said, we’re always looking for ways to improve. This year may turn out to be the year of the frustrated American — when people who were turned off by politics in the past finally showed up to vote. The challenge, then, is to make sure that we’re fairly accounting for all of these people into our poll.

The Reuters/Ipsos poll is different from many others in that we interview everyone before we decide who’s a “likely voter”. We then consider their reported voting history, registration status, interest in election news and certainty about showing up on Election Day. That won’t change, though it’s possible that we’ll fine-tune this analysis before November after checking our results with primary returns in large states. We’re also looking at several other ways to improve the way we predict turnout in the general election like cross-referencing voter files.

Come November, our poll will be put to the test. The Reuters/Ipsos poll was considered one of the most accurate in 2012, and we hope to keep that going this year.

 

No comments so far

We welcome comments that advance the story through relevant opinion, anecdotes, links and data. If you see a comment that you believe is irrelevant or inappropriate, you can flag it to our editors by using the report abuse links. Views expressed in the comments do not represent those of Reuters. For more information on our comment policy, see http://blogs.reuters.com/fulldisclosure/2010/09/27/toward-a-more-thoughtful-conversation-on-stories/