Expert Says Polling Is A Snapshot, Not A Predictor, Of What Happens On Election Day
PITTSBURGH (KDKA) -- Hardly a day goes by without a new political poll telling us who is in the lead in Pennsylvania for president. But can we trust these polls to be accurate?
In 2016, political polls in many states, including Pennsylvania, were wrong. National polls correctly predicted Hillary Clinton would win the popular vote. But the polls failed to see Donald Trump's Electoral College victory.
That's led to a healthy skepticism about polling.
"Polls are only as good as their methodology," professor Costas Panagopoulos told KDKA political editor Jon Delano on Thursday. "They are much less precise than people perceive them to be, but they're also snapshots in time. They may or may not predict what happens with Election Day."
Panagopoulos is the chair of the political science department at Northeastern University and a well-known election scholar. He says pollsters have a tough time discerning who is really a like voter for their polls.
WATCH: More With Costas Panagopoulos
"You really don't care about the views and preferences of people who aren't going to vote, and that's more of an art than a science," said Panagopoulos. "Pollsters haven't nailed down exactly how to include only those people who are going to vote on Election Day, and that's one of the biggest reasons why poll results bounce around."
Last week, we had six polls in Pennsylvania, showing everything from a tie between Trump and Biden to Biden up by nine points.
Unless the sample size is large, 1,000 people or more, state polls showing a candidate ahead in the low single digits could just as likely be wrong as right.
"They're much better off looking for trends in the polling, small differences over time that suggest that either strength for a candidate is climbing or eroding for example," Panagopoulos said.
Another challenge is that many voters won't answer their phones for pollsters. Or, if they do, they lie about whom they're supporting. It's a big problem.
"Pollsters have to adjust their estimates sometimes through statistical weighting to pretend they have a representative sample when they actually don't," Panagopoulos said.
Take all polls with a grain of salt.