Jim Clark: Polling the 2016 election – an update (opinion)
Election Day 2016. East Coast voters cast final ballots at 8 p.m. (EST); the Clinton campaign and most of the media were ready to pop champagne corks.
Trump had made gaffe after gaffe while the well-oiled Democrat machine with its vaunted ground game was well-ahead in the major polls.
At 10:30 p.m., networks announced that Trump had won swing state Florida. Then Iowa and Ohio went to Trump. Early Wednesday, Pennsylvania and Wisconsin were called for Trump. Clinton conceded.
What happened with the polls? As I wrote in my post-election column, “How did the polls get the election so wrong?”, Bloomberg Politics, CBS News, Fox News, Reuters/Ipsos, USA TODAY, Quinnipiac, Monmouth, Economist/YouGov and NBC News/SM all got this election seriously wrong.
I since found out that Gallup, the granddaddy of all pollsters (covered every election since 1936) sat this one out. So did the respected Pew Research Center. Neither did any polling nor expressed any opinion on the 2016 election. Only three pollsters accurately predicted a Trump victory: Investor’s Business Daily (“IBD”), the LA Times/USC poll and Rasmussen.
How did IBD get it right? They recognized more enthusiasm for Trump by Republicans and independents and predicted that GOP turnout would at least match Democratic turnout despite Clinton’s vaunted “ground game.”
They also recognized a cohort of voters who didn’t vote in 2008 or 2012 but supported Trump in primary elections. These were downplayed by other polls as “unlikely voters,” but were in fact Reagan Democrats looking for someone to “clean up the swamp.”
LA Times/USC also got the outcome right. The newspaper partnered with the University of Southern California to pioneer a different polling method.
They carefully selected their sample of voters and communicated by email every day. They never changed their sample but measured changes in sentiment among the same voters. LA Times/USC showed Trump the winner in every poll from the beginning of the general election campaign.
Rasmussen also correctly forecast Trump’s win in the 2016 election and recently shed more light on their success. In a report titled, “Drilling Deeper into Our State Surveys,” issued Jan. 25, 2017, Rasmussen revealed the story of a 2016 polling subcontract they performed for private clients involving a detailed look at the presidential and Senate races in Nevada, Florida, Ohio, North Carolina, Pennsylvania and New Hampshire.
These six, all “swing” states, cumulatively have enough electoral votes to decide the presidential race. Additionally, Rasmussen’s clients believed that the results in these states’ Senate races would determine control of the upper house.
Rasmussen measured voter sentiment by landline and email contact. To predict the result in each of the six states, they had to apply a voter turnout factor. A “strong Democrat” turnout would be based on Democratic performance in the 2012 election; a “strong Republican” turnout would be based on the GOP turnout in the 2014 election.
The report said: “Creating models was a bit like salting stew — part science and part feel. But by using different (state) models we didn’t bias the analysis with our own turnout expectations.”
In the end, as with IDB, their “strong Republican” turnout models almost exactly forecast the actual presidential and senate election results for each state. But their “strong Democrat” turnout models produced forecasts which were no more accurate than the other pollsters’ forecasts.
Rasmussen’s postmortem of their subcontract performance concluded that polling works even if done by robo-calls. If the sample chosen accurately reflects the characteristics of likely voters pollsters can tell the status of and trends in public opinion.
But to accurately forecast an election outcome they need to use both objective and subjective means to predict turnout. Bottom line: “Get turnout right and … the polls will likely lead you to the right conclusion.”
Let’s see if other pollsters adopt IDB’s and Rasmussen’s attention to turnout modeling and LA Times/USC’s sample selection and communication techniques.
If they get this accurate enough, we can just skip the elections.
Jim Clark is president of Republican Advocates. He has served on the Wshoe and Nevada GOP Central Committees. He can be reached at email@example.com.