By Masafumi Fukuda / Yomiuri Shimbun Staff WriterA major theme of the May conference for the American Association for Public Opinion Research (AAPOR) was last year’s failure to predict the result of the U.S. presidential election. A look into the reasons why Donald Trump’s victory could not be anticipated is progressing, but restoring public trust in opinion polls in the United States will be difficult.
Shortly before voting day, U.S. media reported — based on a variety of opinion polls — that the victory of Democratic candidate Hillary Clinton was almost assured. The U.S. newspaper The New York Times calculated that Clinton had an 85 percent likelihood of winning. The Huffington Post news site reported a 98 percent likelihood.
However, Republican candidate Donald Trump secured 306 out of a total of 538 electoral votes, easily surpassing Clinton’s 232. In terms of the total number of votes cast across the United States, Clinton won almost 3 million more votes than Trump. However, the “winner takes all” system that exists in most states stipulates that any candidate who gets even a single vote more than the other candidates wins all of the state’s electors. This system benefited Trump.
AAPOR took the failure to predict the election result seriously, with its president Timothy Johnson acknowledging public confidence in polls was undermined by the perception that they did a poor job with the election. An AAPOR ad hoc committee of experts examined what went wrong, analyzing 39 surveys used to make national forecasts of the voting ratio garnered by each candidate in the 13 days leading up to voting day.
They compared the percentages at which Clinton was predicted to win at the expense of Trump with the actual percentage of the votes received by both candidates across the states and found an average discrepancy of 2.2 points. That meant predictions were more accurate than those made during past elections.
However, regarding state-level opinion polls, which were held separately from the nationwide surveys to predict the total number of electors the candidates would win, the experts found inaccuracies in these polls so large they resulted in a misreading of the overall national result. Looking at more than 400 surveys done in individual states, the experts discovered an average difference of 5.1 points between the opinion polls’ results and the actual election results. The discrepancy was the largest seen since the 2000 presidential election.
Average opinion poll results in 2016 predicted a Clinton win in the states of Pennsylvania, Michigan and Wisconsin, which have consistently gone to Democratic candidates since the 1992 presidential election. On election day, however, these states went to Trump by a narrow margin, handing him the 46 electors that won him the presidency.
Before election day
Why did public opinion polls underestimate support for Trump?
The primary reason was “shifts immediately before election day,” according to the report of the AAPOR committee. An analysis of exit polls found that of the voters who decided who they would vote for in the week before election day, more people chose Trump compared to voters who had made the decision before then. This trend was strong especially in those three states where forecasts turned out to be off the mark.
The experts also provided other explanations. For example, the answers of people with lower levels of education, who were likely to vote for Trump, were not accorded sufficient “weight” in the aggregate results of the polls.
Courtney Kennedy, who chaired the AAPOR committee, expressed her concern that the same mistakes could happen again if the accuracy of predictions at state-level polls is improved.
Immediately after the election, people tried to explain the failure of the forecasts by pointing to “hidden Trump supporters” who had not declared their voting intentions.
However, experts stated in the committee’s report that several analytical methods, such as comparing the results of automated phone surveys, online surveys and the like, had uncovered no evidence to support the “hidden-Trump-supporters” hypothesis.
Even after assuming office, now-President Trump continues to target media outlets that made mistaken forecasts, for example by taking to Twitter to attack opinion polls negative toward him as “fake news” and “rigged.”
One of the media outlets he targets is the TV network CNN. Jennifer Agiesta, director of polling and election analytics at CNN, emphasizes the significance of polls.
“We just have to think carefully about how we can make our wording as neutral as possible,” she said of their polls. She said opinion polls are important as a scientifically meaningful measurement of a democracy, “especially as our country has grown so much more polarized.”
Attracting Japan’s attention
The problems surrounding the failure to predict the results of the U.S. presidential election, and the loss of trust in opinion polls, have also attracted the attention of Japanese experts.
At the annual meeting of the Japanese Association of Electoral Studies in Takamatsu on May 20-21, Prof. Kazuhiro Maeshima of Sophia University, who specializes in modern American politics, gave a presentation on the “Trump phenomenon.” He pointed out that in a public opinion poll conducted by U.S. company Gallup in September 2016 and before the presidential election, 51 percent of Democratic Party supporters said they “trust the media.” By contrast, only 14 percent of Republican Party supporters said the same.
Maeshima said the current climate in the United States will make it difficult for the media and public opinion polls to regain people’s trust.
According to Maeshima, in 1987, the United States revoked the “Fairness Doctrine” — a policy that stipulated political coverage in the media had to be “fair” — in favor of point-of-view coverage that prioritized “freedom of expression.” After that, broadcast media became gradually polarized as either liberal or conservative, and Republican supporters’ dissatisfaction with liberal media outlets grew. It was in this climate that Trump appeared on the scene, attacking the media and amplifying people’s distrust.
Maeshima believes “Trump knows that his supporters rejoice when he attacks the media, so he will probably continue to treat the media and opinion polls as enemies.”
Japan polls add breadth in forecasts
As the aim of conducting polls is to measure public opinion across a country, participants are selected at random. That makes a public opinion poll different from online surveys, which anyone is allowed to take part in. Online surveys can also target registered users of specific websites who applied for the surveys on their own. The scientific approach taken in Japan for public opinion polls are the same as that used in the United States.
However, the nationwide public opinion surveys that Japanese media organizations conduct for national elections are held across all constituencies, in principle. The forecasting methods and the way forecasts are reported are also very different from those in the United States.
For example, The Yomiuri Shimbun conducted a public opinion poll after the day of the House of Councillors election was publicly announced in the summer of last year. Based on the poll results, the newspaper published an analysis of factors such as who the leading candidates were and which ones were “competing” in each constituency by also conducting coverage about the situation rather than forecasting the percentage of votes for each candidate in each constituency.
It also added breadth to its reporting about the number of seats each party was projected to win. Other media outlets use similar methods, and the actual number of seats parties gained in lower house and upper house elections in recent years has generally remained within the range that was forecast.
However, there are people in Japan who hold negative views of opinion polls or confuse them with online questionnaires. There have recently been instances of the Cabinet approval ratings being reported based on online surveys alone. On social media such as Twitter, there are more and more messages reminiscent of Donald Trump’s posts that denounce public opinion polls as “unreliable” or “fabricated.”
Toshiki Sato of the University of Tokyo, who specializes in sociology, suggests a possible explanation for this development.
“As use of social media expanded, it became easier for people to find others who share their opinions, making them easily believe they are the ones who represent public opinion,” Sato said. “They find it hard to accept the results of public opinion polls that contradict their own opinions, and thus assume it’s the surveys that are wrong.”
Sato insisted that “through opinion polls that are conducted in a scientific fashion, we can hear the opinions of people across generations and those with all sorts of ideas. I hope the media will continue working to convey a big picture of public opinion as accurately as possible.”
(From The Yomiuri Shimbun, June 30, 2017)