The 2016 race for the Republican nomination has already challenged virtually every bit of political conventional wisdom. It would take an entire series of articles to detail how wildly wrong the pundit and commentariat class have been this campaign. The race has also been challenging for pollsters.
With the conclusion of Super Tuesday, 15 states have voted in five separate voting “events.” The first “event” was the Iowa caucus, which is perhaps the most polled individual state on the election calendar. Caucuses can be difficult to poll, but Iowa has been conducting this kind of contest for more than 30 years, offering lots of historical context for crafting a poll.
That said, all of the polls in Iowa predicted a Trump victory in the first voting state. The RealClearPolitics average of polls predicted Trump winning the caucus by five points. Sen. Ted Cruz won the state by 3.3 points, an eight point swing from the polls.
The average of polls, however, obscures the fact that several national pollsters missed the mark widely in Iowa. Quinnipiac, Gravis Marketing, and NBC all showed Trump up by seven points in their final poll. PPP showed Trump up by eight points. On the other hand, Emerson and Opinion Savvy showed a much closer race, with Trump up by just one point. Still, every poll showed Trump winning Iowa.
The New Hampshire primary was the second polling “event” in the campaign. Donald Trump won that primary by a landslide, beating the field by almost 20 points. Unsurprisingly, most polling reflected Trump’s wide lead in the Granite State, although a Gravis Marketing poll seriously underestimated his support, showing him just 11 points up in its final poll.
The New Hampshire polls struggled, though, in handicapping the race for second and third. In general, the polls seriously overestimated support for Sen. Marco Rubio and underestimated support of Ohio Governor John Kasich.
Kasich’s support was also seriously underestimated in Vermont, which voted on Super Tuesday. The only poll of the contest, from Vermont public radio and Castleton Polling Institute, showed Trump winning the state by 15 points and Kasich far back in the pack with just 10 percent support. On Super Tuesday, Kasich won 30 percent support, however, and came within a few hundred votes of winning the state from Trump.
In all, polls in only five states have finished close to the final elections results: New Hampshire, Nevada, Alabama, Massachusetts, and Georgia. Polling everywhere else has generally overestimated Trump’s final support. Even when Trump won those states, he generally won by a much smaller margin than predicted in the polls.
Trump won the South Carolina primary by 10 points. The final poll from Emerson, however, showed Trump winning the state by 17 points. The final Monmouth poll showed Trump winning by 16 points.
In Virginia, CBS had Trump winning the state by 13 points, while Monmouth favored him by 14 points. One poll, from Roanoke University, showed Trump galloping to a 23 point victory in the Old Dominion. Only Christopher Newport University showed a close race, with Trump ahead by six points.
Trump ended up winning Virginia by just under three points.
In Texas, the RealClearPolitics average of polls showed Cruz leading his home state by nine points. Cruz won the state by 17 points, almost double the margin predicted by the polls. The polls generally both overstated Trump’s support and underestimated Cruz’s support.
Polls by Emerson and ARG were the most incorrect, showing a toss-up race, within the margin of error, between Cruz and Trump. For those keeping score at home, those polls were off by more than 15 points. The NBC/WSJ poll was the closest to nailing the actual levels of support for Trump and Cruz, although they missed the final margin by four points.
All the polling in Oklahoma showed Trump with a strong lead heading into Super Tuesday. The RealClearPolitics average of polls showed him winning the Sooner State by 11 points. Monmouth University went a bit higher than this, predicting Trump to win the state by 12 points. Cruz, however, carried Oklahoma by just over six points. This represents a swing of 17 points from the final poll average.
Criticism of polls is both overblown and underappreciated. Polling primary contests is difficult, because you have to predict a subset of the overall electorate. General elections are easier because most people interviewed will be voting. This isn’t the case in primaries.
Harder than predicting the overall turnout, however, is guessing which parts of the electorate, and in what proportions, will participate. Will voters be older or younger than past elections or will a surge of new voters swarm the polls? Which issues will be most important to the voters who cast ballots as opposed the general voting populace?
The other challenge for trying to anticipate results based on polls is that they are increasingly lagging indicators. Even the best poll is simply providing a snapshot of the race from a few days ago and may not reflect the latest developments. In our 24/7 political world, developments arise quickly that aren’t immediately reflected in the polls.
On the eve of Super Tuesday, NBC/WSJ released a round of polls which were comprised, in a large part, by interviews conducted before the final GOP debate before voting. All polling has a very limited ability to capture late developments like this. Four states are voting on Saturday, so no poll will reflect any voting changes that may result from Thursday night’s debate.
Even if a magical polling fairy were able to factor all of these variables into its estimates, it still couldn’t account for the large number of people who don’t make a final decision until the final hours of the campaign.
In Virginia, Marco Rubio beat Donald Trump by almost 30 points among the people who decided whom to support in the 2-3 days before voting. This, more than anything else, accounted for the closeness of the race in Virginia, where every public poll had predicted a double-digit win for Trump.
This will become a particularly large phenomenon as state-specific elections come faster than polling can estimate. Within the next two weeks, 14 more states and territories will vote in the Republican primary. At best, there will be just small handful of polls to estimate their outcomes.
If the past is any guide, though, less reliance on polling may give us a better guess at the final outcome in these states.