Many political pollsters took it on the chin in 2016. Their surveys proved to be wildly off in the presidential race, among others. So as the 2018 midterms approach, will we all stop paying attention to election surveys, deeming them unscientific and uninteresting?
Of course not. President Trump himself is among many politicos around the country, including right here in North Carolina, who remain intensely interested in polls. Republicans have their favorites. Democrats have theirs. Typically, the “best” survey is the one that shows your side on top in public opinion, or fated to win the next election.
Because we aren’t really going to tune out political polls — and, when administered and interpreted correctly, they do provide useful information — the best strategy is to avoiding playing favorites. Diversify your survey diet. Fortunately, North Carolinians are well served by a stable of reputable pollsters who take the political temperature of our state on a regular basis.
For the Tar Heel state, at least, the aggregate direction of the publicly available polls in the presidential race of 2016 didn’t prove to be at odds with the result. According to the RealClearPolitics average for the final days of the race, Trump had a one-percentage-point edge on Hillary Clinton. He ended up winning the Tar Heel State by 3.7 points. In Michigan, Pennsylvania, and Wisconsin, Trump trailed Clinton in the final polls — by a sizable 6.5 percent in the latter case — yet eked out victories by Election Day.
Among the pollsters that predicted Trump’s victory in North Carolina was Survey USA, which uses interactive-voice response (no live operator) to deliver polling services both for state news organizations such as WRAL-TV and Spectrum News and for the nonprofit Civitas Institute.
The statistical-analysis website FiveThirtyEight.com periodically rates pollsters on the basis of accuracy and transparency. Survey USA is one of the top performers — and the only non-live-operator pollster to earn an A grade.
Another high-performer is Elon University, whose polling program earned an A-minus. Supervised by political-science professors, it uses student operators to conduct an average of three statewide surveys per semester. Another Triad school with a similar approach, High Point University, earned a B-plus from FiveThirtyEight. It also tends to conduct about three surveys a semester.
Another B-plus pollster is National Research, which has conducted roughly monthly live-operator polls in recent years for Civitas, a conservative think tank based in Raleigh. On the Left, the explicitly Democratic firm Public Policy Polling uses IVR to conduct vast numbers of public and private polls each year. Its grade this cycle was a B. I should also mention that Meredith College has now launched its own polling program. While too young to get a FiveThirtyEight rating, it also deserves your attention.
Many other out-of-state survey units come into the battleground of North Carolina during the homestretch of our general elections. Don’t expect to see much from them in 2018, since we won’t be electing a U.S. senator or governor this year. But in 2020, they’ll be back in droves.
My diversification advice applies to these national outfits, too. Never discount a pollster just because you don’t like the editorial lean of its parent organization. For example, the work that Anderson Robbins Research and Shaw & Co. Research did for Fox News earned an A grade. Their final North Carolina survey had Trump up five points, which wasn’t too far from the final result of 3.7 points. If you’re a progressive who doesn’t much like the commentary on Fox, don’t let that deter you from looking at its polling.
As we learned in 2016, the old saw about survey research can cut you if you’re not careful: the only polls that count are the ones on Election Day (or during the weeks leading up to it, if you vote early). But political polling isn’t going away, and people aren’t going to stop looking at it. So consume it wisely. Track averages, not just single polls. And always remember the inescapable margin of sampling error.