The US election shock

The entire community (almost) of American pollsters was shocked by the presidential election’s outcome.

After two days, Nate Silver, my favorite summarizer, scrambled: “Why FiveThirtyEight gave Trump a better chance than almost anyone else.”  His final forecast gave Donald Trump a 29-percent chance of winning the Electoral College. “Even the Trump campaign itself put their candidate’s chances at 30 percent.”

Actually, the result was a virtual tie.  A 1 million vote margin among 120 million votes is only 0.8 percent.

Silver pointed out that Hillary Clinton lost because Michigan (16 electoral votes), Wisconsin (10) and Pennsylvania (20) unexpectedly flipped to Trump. The three states’ total 46 votes, plus Clinton’s final 232, would have given her 278, whereas Trump’s final 306, minus 46, would have given him only 260.

Michigan had an average final polling margin of -4.0 percent for Trump; instead, he won by +0.3. In Wisconsin, his margin was -5.4, but he won by +1.0. In Pennsylvania, it was -3.7, but he got +1.2.  Such final margins—which already factored in the FBI investigation incident—are normally big enough to expect Trump to lose those states.

The outlier poll.  The one poll that consistently favored Trump was the University of Southern California Dornsife/Los Angeles Times Daybreak Poll, which started last July.

The respondents to the Daybreak Poll were originally selected by probability sampling, and then invited to join a panel, for repeated polling by internet.  Those willing to join, but not equipped for online response, were provided with internet service and a suitable tablet.  This is essentially the same strategy as that of the 2016Bilang Pilipino SWS Mobile Survey (Opinion, 4/2/16).

The panelists were asked, online, “What is the percent chance: (a) that you will vote in the election; (b) that you will vote for Clinton, Trump, or someone else; and (c) that Clinton, Trump or someone else will win?”  Thus, the answers were not a simple Yes or No, but choices between zero and 100.

The Daybreak Poll was specially weighted to enhance its representativeness, i.e., it did not depend on the demographics of the original participants.  It also took account of how the panelists had voted in 2012.  For instance, it adjusted the sex ratio to 48 percent men and 52 percent women. One-seventh of the original 3,200 participants were questioned every day, and given one week to reply.  In this way, it produced daily reports based on a seven-day rolling average.

By August, the Daybreak Poll had already seen that Trump needed to mobilize white voters who didn’t vote in 2012.  It didn’t know if he could do it, of course, but it apparently anticipated ABC News’ Nov. 9 report, “Huge margin among working-class whites lifts Trump to a stunning election upset.”

Hindsight.  Election forecasters must operate within whatever the survey respondents say. Given the widespread expectation of a Clinton win, many may have taken refuge in Don’t Know or Undecided answers, following the famous theory of “The Spiral of Silence” (by Elisabeth Noelle-Neumann, 1993).

Canadian sociologist Claire Durand, president-elect of the World Association for Public Opinion Research, says she had expected the outcome, by assigning the undecideds two-thirds to Trump and one-third to Clinton. Alan Dershowitz, author of “Electile Dysfunction,” saw the polling as distorted by the raw emotion, populism, anger, nationalism, and class division of the election.

Contact [email protected].