Has Technology Made Election Polls More Accurate? Not Yet
Statistical models, simulations, social media outreach, and the capacity to analyze big data sets—you’d think by now that technology would have delivered election forecasts that more reliably predicted Donald Trump’s victory in the 2016 presidential race.
But most pollsters—and the people who relied on them—were gobsmacked by Trump’s upset wins, not only in battleground states but also in regions thought to be safely in the Clinton camp.
There were exceptions. But the more accurate pollsters may have succeeded by making judgment calls on the design of their polls, rather than by relying on the power of technology.
Politico gives credit to a handful of small Republican polling organizations that gave Trump the edge after conducting automated surveys of people answering landline phones. This is a traditional method that proved so unreliable for venerable opinion research firm Gallup that it decided not to conduct polls for the 2016 presidential race. The rap on the method is that many people—especially younger people—communicate via smartphones, and that landline users are also less likely to participate in phone surveys these days. Even so, the Republican pollsters apparently pegged the race.
By contrast, NBC News teamed up with Palo Alto, CA-based tech firm SurveyMonkey to conduct online polls. SurveyMonkey, which helps companies and individuals poll groups on a variety of issues, tapped into its network of online users to pose questions about the presidential race. Like the Republican pollsters, NBC/SurveyMonkey kept on polling up until a few days before election day. But on Election Eve, NBC News reported that Clinton held a stable 6-point lead over Donald Trump.
Another news organization’s polls, also conducted online, broke away from the pack and forecast Trump’s win. The USC/Los Angeles Times Daybreaker poll, which had shown Trump leading for months, put the Republican 3 points ahead on the morning of the election, according to the newspaper’s recap published today. That success relied on decisions about the way survey responses would be evaluated.
The Daybreaker pollsters interpreted its raw data with a complex “weighting” adjustment that, among other things, took into account Trump’s track record of mobilizing disaffected voters. The pollsters also considered survey data indicating that Trump voters were reluctant to discuss their choice with telephone pollsters—possibly indicating a hidden reservoir of support for the Republican candidate.
Back in January, experts interviewed by VOA News reporter Steve Baragona predicted polling failures for the 2016 race. They pointed not only to the sinking response rates by landline phone users, but also to the fact that the use of online polling is in its early days. Further research may refine the method, they say.
“While online polling may be the way of the future, it’s a future that many experts say is still a long way off,” Baragona concluded.