2016 Presidential Polling Was Right and Wrong
National Polls Got Close to the Popular Vote but Statewide Polls Missed in Key Electoral College Battlegrounds
--
As a professional pollster for the past 20 years, I’ve absorbed your collective scorn. Never trust the polls. They’re always wrong. I’ve never been called. Everyone lies to pollsters. Who are these people who do online polls? The polls missed on Donald Trump. Did they?
Predicting the National Vote
Any one poll can be right or wrong within its stated margins of sampling error even when you add in other possibilities of why the polls might be wrong. Still, I would say the national polls did reasonably well. Here are the facts on the performance of national polls as a group.
- According to the best non-partisan tally of the popular vote I can find, Hillary Clinton won 65,534,951 votes and Donald Trump won 62,858,281, which is a 2% difference. That’s the benchmark you want to duplicate in national polling.
- The Real Clear Politics (RCP) aggregation of polls taken right before the election had Clinton up by 3.3%. Margins of sampling error in those calculations ranged from 1% to 3.6%. There’s plenty of room in there for most of these pre-election estimates to be correct, even IBD/TIPP’s tracking poll that had Trump winning by 2% in national voting.
- Huffington Post’s aggregation of national polls had Clinton up by much more — 5.3%. Still, factoring in the margin of survey error, the 2% actual ballot tally could show up as 5.3% in some polls with margins of error over 3%. Being wrong within a margin of error is not exactly being right.
Predicting the Electoral College
As Americans learn in civics or social studies classes, the national popular vote does not determine who wins the presidency, the Electoral College does. Each state’s electors are determined by the combined number of U.S. Senators (2) plus the number of members of the U.S. House (varies). According to Gallup, almost half of Americans now support the Electoral College, a significant shift from just over a third in 2004.
The campaign effect of the Electoral College is to force candidates to visit smaller, less populated states, that might help them reach 270, the majority needed to win the presidency. Therefore, we must go beyond national polling averages and look at how statewide polls did in the states that Trump flipped from blue to red.
Florida: This state is a true battleground, which flips back and forth between the parties. It was said, with good reason, that Trump had to win Florida (his second home state) to get to 270. He did. The final RCP average in Florida had Trump up 0.2% and the final official tally had Trump winning by 1.1%. In my book that’s really solid.
Ohio: Trump won by 8% in Ohio, which was well outside the margin of error than the final aggregates that had him winning by 3.5%. The prediction was based on very few polls and none within the final two days. We simply didn’t have enough polls, late, to predict the spread of the outcome very well. But they did get it right.
Michigan: Here is where things went off the rails. The final RCP average had Clinton winning by 3.4% and Trump won by 0.2%. So, the prediction was wrong and it was likely outside of the margin of error. Not good. The only organization that called Trump’s win was the Trafalgar Group, a Republican firm that overestimated the victory by 1.8% — not bad. Why? They were the only pollsters to report data on November 6, still two days from the election but later than the others in the prediction set.
Wisconsin: Again, Trump won by less than a point (0.8%) and Clinton was predicted to win, this time by a much wider margin (6.5%). What went wrong, again, was the final poll reported in the aggregate was on November 2, six days before the election. With so little polling in that average (four polls), you have to wonder why you’d make a prediction at all.
Pennsylvania: This was the surprise, for me, of the Election Day. Republicans hadn’t won the state since 1984 in Ronald Reagan’s 49-state landslide. Polling aggregates had Clinton consistently ahead but predicting only a 1.9% victory the Democrat, which should have been a warning sign to all of us watching the campaign. Trump won by 1.1%, which is still within the margin of error but was wrong. Again, the firm that got it right was the Trafalgar Group, which reported out relatively late on November 5 and had Trump ahead by 1%, pretty much nailing the result.
2020: More Statewide Polls at the Right Times
Despite all the complications of predicting election outcomes, including social desirability bias, national polls correctly foresaw Clinton winning the popular vote on Election Day within a reasonably close margin of error. There were several polls collecting data all through the final days of the election, giving us a pretty solid guess on who was going to win nationally. There was enough good data in the averages to make the right call.
But statewide polling aggregates have a long way to go before we should put similar faith in them. In the most crucial battleground states, we had too few polls and they were conducted too far ahead of Election Day to accurately predict who would win and by how much. Yes, the polls had their problems, but more of them at the right times would have given us a better picture of who would win the Electoral College and the presidency.
If you enjoyed this article, click the💚 below so other people will see this here on Medium. Follow me on Twitter @michaelcohen. You can follow our research on this website or on Twitter @PEORIAProject, which is funded by a generous grant from Mark R. Shenkman. To learn more about the Graduate School of Political Management visit our website or follow us on Twitter @GSPMgwu.
