- Published on Tuesday, 25 November 2008 00:00
- Written by Carl Bialik, WSJ.com
I wrote the week before the election about the struggles of one pollster to reach the elusive demographic group of 18- to 24-year-olds and measure their collective opinion. Despite those challenges, TechnoMetrica Institute of Policy & Politics, the polling arm of the market-intelligence company TechnoMetrica, came very close to predicting president-elect Barack Obama's winning margin. In its final poll for Investor's Business Daily before Election Day, TIPP gave Obama a lead of 7.2 percentage points over Sen. John McCain. Obama is leading by 6.8 percentage points in the national popular vote, with nearly all votes counted.
"I did a pretty good job," Raghavan Mayur, president of TechnoMetrica, told me. His method to allocate those voters who identified themselves as undecided to each candidate helped him get close to the final margin - Mayur assigned two-thirds of these voters to Obama, expanding his lead in the poll by 2.1 percentage points. Unlike other pollsters, who may split the undecideds evenly or based on the candidates' standing among decided voters, Mayur used a complex algorithm to assign each of these voters. The method uses 10 variables, such as the respondent's party, gender and 2004 vote, to predict the 2008 vote.
More than some other pollsters, when publishing his poll Mayur breaks down results by a wide range of demographic variables. Among some of these groups, he was right on target, at least compared to exit polls: Obama won by 13 percentage points among women, as Mayur predicted, and won the group of voters whose annual income ranged from $30,000 to $50,000 by 12 points, compared to the nine-point margin predicted by Mayur. Among other groups, Mayur's predictions weren't as strong. Obama did better among Catholics and young voters than predicted. Mayur said the occasional miss could be attributed to a small sample size among certain subgroups. "If you see the same pattern over three or four days, you can feel comfortable about the thing," he said.
He also predicted that candidates other than Obama and McCain would get 4.2% of the presidential vote; their actual vote share was 1.3%. Mayur explained that he names these candidates explicitly in the polls over the final days before the election, which may boost their share. He includes them "just because I want to be more careful," he said.
Like all pollsters, Mayur will face fresh challenges in 2012, including an expected increase in the proportion of households without a landline telephone. "It's a tougher economy, and people are looking at every expense," he said. Young people may also get harder to reach: "The age-group problem is nagging me all the time."
Because cellphone users can be averse to unexpected phone calls, particularly when they cost valuable minutes, Mayur said he would consider assembling an opt-in panel of cellphone-only users, and call on a portion of them in each poll. "The world is changing, and it's not anymore the days when you can rely on a random-digit sample," he said.
In other retrospectives on election polling:
- The polls in Alaska missed the mark badly on three major races, indicating a much-bigger Democratic vote share than materialized on Election Day. The Anchorage Daily News explored the polling failure. "I'm not happy how that came out," said one pollster who predicted a 22-point victory for the Democratic Senate candidate, who was declared winner only last week, and by just a percentage point. That aberrant survey didn't make it into the pollster's selected recap of his "most ACCURATE final national and state election polls."
- Nationally, as I wrote the week of the election, the pollsters did well. "While we worried about the many challenges, the telephone survey again defied the odds and delivered mostly accurate results," former pollster and Pollster.com blogger Mark Blumenthal wrote in the National Journal. He offers more context on his blog, and ends with a question about whether all post-election analysis should center on pollsters' final polls, when the vast majority of polls came earlier and helped shape coverage.