A lot of people are walking into Raghavan Mayur’s office in Ramsey these days to ask how he conducts polls.
To answer, Mayur sometimes reaches into a canvas bag and pulls out a college composition notebook with a creased black-and-white cover. He opens it to reveal pages of wide-lined paper, each covered in a hand-drawn chart summarizing the results from the previous day’s polling.
For the last month of the recent election, Mayur spent two hours a day compiling the data. When the numbers were ready, he copied them into his notebook, by hand, using a No. 2 pencil.
This is how Mayur became one of only a handful of pollsters in the United States to successfully predict that Donald Trump would win the 2016 presidential election.
“Doing it this way, day after day, I know the numbers better. So when something seems wrong, I catch it,” said Mayur, a Ridgewood resident and the founder and president of TechnoMetrica, a polling and market research company. “Two-thirds of this work is science. One-third is art.”
Whatever the precise science-to-art ratio, Mayur’s success does not appear to be the result of luck. In 2004, President George W. Bush beat his Democratic challenger, John Kerry, by 2.4 percent; TechnoMetrica predicted the margin would be 2.1 percent, coming closer than any other poll in the country. In 2008, Mayur got it precisely right, accurately predicting Barack Obama’s 7.2 percent margin of victory down to the decimal point. In 2012, The New York Times compared the performance of 23 polling companies over the course of the election season, and found TechnoMetrica was the most accurate of all.
Mayur’s reliance on pencils and composition books makes him seem exotic, but really, he said, it’s just a simple way to check his math at the end. Far more important, he says, is what happens at the start.
“One of the most important secrets in this business is the data. Nothing can compensate for bad data,” Mayur said in an interview in his Ramsey office. “I get top-quality data.”
Perhaps never in American electoral history have opinion polls been the focus of so much anger. During the final month of the campaign Donald Trump made daily sport of slamming polls, virtually all of which predicted he would lose to Hillary Clinton in a double-digit landslide. Trump called the surveys “phony,” and accused pollsters and Clinton’s campaign chairman, John Podesta, of having “rigged the polls.”
Clinton won the popular vote by a narrow margin, but did far worse than most pollsters predicted, leaving her unable to overcome Trump’s big win in the Electoral College.
Democrats alleged bias, too. Unlike the vast majority of polls, surveys conducted jointly by the University of Southern California and The Los Angeles Times regularly found Trump holding even with Clinton. In mid-October The New York Times found that the poll counted a single man’s vote 30 times, and that man happened to be a 19-year-old African-American in Illinois who planned to vote for Trump. Still, Republicans viewed the results as evidence that other pollsters systemically underestimated Trump’s support among African-Americans.
“There is almost 100 percent acknowledgment that the polls were really, really off, and some sincere bewilderment as to why,” said Tim Johnson, vice president of the American Association for Public Opinion Research. “It didn’t matter what their methodology was, everybody got it wrong in the same direction, which tells us there’s an unknown source of systematic bias that we haven’t detected yet.”
Almost everybody got it wrong, except Mayur. By the final days, Mayur reported in Investor’s Business Daily that Trump even had gained a razor-thin lead. His results were so different from the herd that he attracted attention from The Washington Post, which ran a story on Oct. 25 calling Mayur’s methodology “unusual.”
The Post’s polling manager, Scott Clement, wondered why Mayur does not use voters’ education levels in his final tally, especially since the education gap in the 2016 race was so pronounced, with Clinton polling strongly among college-educated voters and Trump finding support among people without college degrees. (Mayur says he doesn’t include schooling in his final tally because so many people inflate their education level, and he doesn’t include income because many respondents report their salaries inaccurately.)
Clement questioned why Mayur used 2014 data from the U.S. Census Bureau as his benchmark, rather than data published more recently. And he wondered why Mayur declines to share his methodology with other pollsters.
Johnson, who is also director of the Survey Research Laboratory at the University of Illinois at Chicago, said that Mayur’s refusal to be transparent is a serious problem.
“In my experience, when people are hiding their methodology they’re doing something they don’t want you to know,” Johnson said. “There’s no defense for it.”
Mayur’s secrecy is especially troubling, Johnson said, since his record of accuracy means he may have something to teach others in the field.
“The fact that he’s been getting it right means the rest of us certainly need to be taking a much more careful look at his methodology,” Johnson said.
Mayur argues that his colleagues are trying to use the association’s transparency policy to steal his ideas.
“Some of these critics are jealous of my work,” he said Thursday. “The proof is in the pudding.”
Looking under the hood of the USC and TechnoMetrica polls may give some hints about how they got it right, however, when everybody else was wrong. In many ways the methodology of the two polls is very different. The California poll was Internet-based. It surveys the same group of respondents once a week throughout the campaign, and it weights responses in part by how the participants say they voted in 2012, according to a description by the university’s public relations department.
Mayur’s method is more old-fashioned. He packs about 35 people into cubicles in his office, and directs them to interview about 250 different people a day. They dial the numbers by hand. Mayur publishes his results daily, using an average of all the data collected over the last week.
His polls were similar to the USC-L.A. Times surveys in one crucial way: Both took a comparatively loose approach when it came to identifying likely voters. Mayur said his first pass at his data includes everyone who says they are “very likely” or “somewhat likely” to vote.
The USC media office said its polls “include voters who did not vote in the prior presidential election and others who may be considered ‘less likely to vote.’ Many traditional polls exclude some or all of these voters.”
This technique may have been more accurate this year, when so many low-income people with high school degrees and spotty voting records turned out in large numbers for Trump, said Ashley Koning, director of the Eagleton Center for Public Interest Polling at Rutgers University. And it may have added to Mayur’s success in 2008, when voter turnout was especially high among African-American voters who typically have below-average election turnout.
Here, then, may lie the true bias of most polls this year: It wasn’t partisan, but rather the habit of defining likely voters as people who have consistently voted in the past.
“Both Obama and Trump mobilized disenchanted voters, infrequent voters,” Koning said. “If you expect people to have substantial past voting behavior, then you may have missed a lot of these infrequent voters.”
Now that the election is over, Mayur is taking a few days to rest. He’s getting more sleep, and on Thursday his phone room was empty of callers. He plans to switch his attention back to the main work of his business, which focuses on conducting polls and focus groups for private companies, and tallying the results using a computer.
Mayur won’t need to buy any more composition books until the summer of 2019, when the next presidential campaign heats up, and he returns to his unique ways.
“I don’t know what other polls do. I know what I do,” Mayur said. “It’s almost like making a pasta sauce. Do you put oregano or not? Some people put sugar. Every house has their own way.”
Click here to read the original article on The Record's website.