Despite unforgivable slips in the 2016 U.S. presidential race, the polling industry must be strengthened, not discredited. It remains crucial in an era in which markets are hypersensitive to political outcomes.
By 11:00pm EST on November 8, 2016, after commercial breaks allowed the world to swallow the unexpected reality of a Donald Trump presidency, pundits pinned the blame on public opinion polls. Electoral experts firmly renounced major polls for miscalculating the electoral outcome by biblical proportions into the morning hours. Mr. Trump, who according to the superstar electoral statistician Nate Silver had a 28.6% of winning the election, ended up flipping the states of Michigan, Wisconsin, and Pennsylvania. He also took the key battleground states—Florida, Ohio, and North Carolina.
Mr. Silver is not the only expert who was wrong. Every major publication, think tank, and agency able to forecast the election predicted a Trump loss by margins making Silver’s forecast look optimistic for the Trump campaign. Gallup published an article on November 2 (less than a week before election day), stating that the Trump campaign’s ratings were the “worst in recent election years”, with a 29% national approval rate.
News broadcasting stations, however, were the furthest off the mark. The day before election day, all major cable stations had Clinton winning by a couple points: Fox, ABC, and CBS had Clinton up by 4 points; while NBC had Clinton ahead of Trump by 6 points. Polls funded by other publications and agencies also forecasted a Clinton win, albeit by a smaller margin: The Economist and YouGov had Clinton up by 4 points while Reuters and Bloomberg had her up by 3 points.
The following headlines from The Economist articles provide a glimpse into the level of disproportional forecasting going into Election Day:
This sequence, which starts cautiously optimistic, then defensive, and finally accepting, is representative of the sentiment among the pundits. Going forward, experts should not jump to discredit the entire polling industry but rather allow academia to reassess the methodology for future national polls. A technical review of polling companies should address the following stress points.
Concerning the research design, national polls usually have a sample size of 1,000 people. In a geographically massive and culturally heterogeneous country of 320 million people such as the United States, sample sizes may need to be significantly larger to cover more counties, including rural areas, to have a deeper footprint among the electorate.
Polls were somewhat accurate when analyzing the overall national popular vote, but highly inaccurate at the state and local level. A miscalculation at the state level can make a large difference under the Electoral College’s points system. The disproportionate tally mechanism of the Electoral College brings into question how polls can correctly predict a national election in a highly decentralized electoral system, especially with an untraditional candidate such as Donald Trump.
Mainstream polls may also want to revisit how to capture politically marginalized groups. One unifying characteristic among three unexpected electoral outcomes in 2016—Brexit, the Colombian peace plebiscite, and the U.S. presidential elections—is the undocumented strength of a resentful anti-establishment silent majority. The polls may be missing this significant chunk of the electorate, which is composed of diverse demographic and income groups. The silent majority’s level of distrust with the establishment may have spilled over to independent institutions such as polling agencies, leading this important cohort to reject polling requests en masse.
In the three electoral cases presented, there was no stark contrast in the options available and voters were left to choose between a menu of suboptimal scenarios. In Brexit, Leave supporters were willing to sacrifice macroeconomic stability for bureaucratic sovereignty. In Colombia, No supporters sacrificed the demobilization of the most enduring guerrilla in the western hemisphere for the possibility of tougher sanctions. In the United States, Trump supporters turned down political and diplomatic experience for a systemic shock to the establishment.
These cases are not traditional or simple. Voters had to logically process a very rough menu of choices, which only increased popular dissatisfaction. Given the context of broad public distrust of government, polls may need to readjust their methodology to more effectively capture the pulse during untraditional electoral patterns.
Polls are adjusted to a dichotomous party-based political model, when perhaps, the establishment and party do not have the influence over voters they once had. Paul Ryan, Speaker of the House of Representatives, said “Trump pulled off an enormous political feat”, meaning that Trump won mainly with his persona and without the full backing of the party machine. The Republican Party remained fractured over Trump’s candidacy until the end.
Maybe this anti-establishment wave of electoral politics in established liberal democracies has not been grasped by big data. Pollsters are hanging on to an old party-based model in a context in which parties are mistrusted, as they represent a decaying governing elite.
The polling blunders of 2016 cannot be taken lightly. In the era of ultra-low interest rates and thin yields, the markets—particularly currencies—have become hypersensitive to political outcomes. The recent market politically induced volatility also transcends borders.
More than ever, electoral outcomes have a direct implication on global markets, even if the policies promised in campaigns are unfeasible in the short-term. The U.S. election, for example, has severely altered a large number of currencies, regardless of the country’s current account balance or general economic standing. The currency market’s reaction to Trump’s election is symptomatic of the deep economic interconnectedness of the global economy.
If political stakeholders have placed so much trust in polls, it is because they have worked successfully in the past. Yes, 2016 has been an unconventional year for democracy and therefore polls as well, but this is no excuse to discredit the industry as was done by pundits on Election Day. Rather, firms should reassess how polls are structured in unorthodox political contests and recalibrate the qualitative methodology to treat voters as complex social beings instead of robots.
This article was originally published by Global Risk Insights and written by GRI Senior Analyst Daniel Lemaitre.