Poll Vaulting

By Matthew Kredell

USC makes headlines in opinion research with help from the Los Angeles Times.

Art by Dongyi WuSlideshow IconArt by Dongyi Wu
Art by Dongyi Wu Art by Dongyi Wu

LIKE WEATHER FORECASTING and economic predictions, opinion polling is an imprecise science. So when a polling organization proves accurate, it wins respect and loyal followers.

This was the happy fate that greeted the USC Dornsife College of Letters, Arts and Sciences/Los Angeles Times Poll almost from day one. Launched in 2009, the university-newspaper collaboration wowed observers in its initial election cycle, distinguishing itself by making the most accurate prediction of the California gubernatorial race.

The poll’s final projection that Democrat Jerry Brown would defeat Republican Meg Whitman by 13 percent was just one-tenth of a percent off the actual 12.9 percent margin. Every other poll in California and the nation came in low when predicting Brown’s margin of victory. The USC Dornsife/Los Angeles Times Poll also came within 2 percent on Democrat Barbara Boxer’s victory over Republican Carly Fiorina for U.S. Senate.

The poll has brought tremendous visibility to the university, says Dan Schnur, the political scientist who runs the joint venture from his perch as director of the USC Jesse M. Unruh Institute of Politics. “The survey received more mentions in the news media than any other topic related to the University of Southern California,” he says.

USC Dornsife and the Los Angeles Times collaboration came on the heels of the announcement, right after the 2008 presidential election, that the Times was shutting down its own venerable in-house polling unit. Sensing an opportunity, Schnur approached the Times to propose a joint venture. While there was little precedent for such a partnership, it’s clearly advantageous to both organizations.

“There’s no question that sharing the cost was part of the attraction for the Times,” says Schnur, “but a much greater motivation was the shared credibility that the two institutions brought to the project.” Schnur brings a fair amount of credibility in his own right: He served as director of communications for the 2000 presidential campaign of U.S. Sen. John McCain and spent five years as chief media spokesman for California Gov. Pete Wilson in the 1990s.

Published four to six times a year, the USC Dornsife/Los Angeles Times Poll is the product of a highly symbiotic relationship. “We work jointly with the Times to decide how many polls
will be done each year, the timing of those polls, the topics covered and the questions asked,” Schnur says.

Schnur hammers out the general framework for the year’s polls with Times editor-in-chief Davan Maharaj, who then directs him to specific editors and reporters with a deep knowledge of the state and its politics. To eliminate political bias, two private opinion research firms are used to canvass registered California voters: the Democratic-leaning Greenberg Quinlan Rosner and the Republican-leaning America Viewpoint. Together, the professor and the journalists work with the polling firms to develop and fine-tune the survey instrument and analyze its results. In follow-up sessions, the Times professionals often come to USC to take part in classroom discussions.

In addition to kudos for accuracy, the USC Dornsife/Los Angeles Times Poll quickly gained a reputation for innovation. Leading up to the 2010 election, the poll interviewed 1,500 people. No other state poll queried more than 1,100. (Respondents are randomly selected from a list of statewide registered voters and contacted via landline or cell.)

While Schnur usually works alone, in 2010 he reached out to USC Dornsife political scientist Jane Junn for help. At the time Schnur was on leave from the university, serving as chairman of California’s Fair Political Practices Commission. Junn introduced an important innovation: She decided to give Latino voters the option of being interviewed in Spanish.

“By doing bilingual interviews, we managed to get a better representation of Latino voters,” Junn says. The technique yields higher rates of response and cooperation. It also boosts the poll’s accuracy.

In the May 2012 poll, 38 percent of Latino respondents chose to have the interview conducted in Spanish; 62 percent opted for English.
This spring, Schnur introduced yet another innovation: the USC Dornsife Online Survey. Although Web-based polling is fairly commonplace in the private sector, it’s relatively new in the political world. The approach has several advantages:

• The survey is conducted over just two or three days, limiting the possible impact of breaking news or current events on survey results. (The phone-based poll is conducted over a week.)
• When asking for a reaction to a quote from, say, President Obama, a video clip can be streamed instantly from the Web page, giving respondents a sense of how he delivered the words.
• Before answering a question, respondents can stop the survey and do an Internet search to become better informed on the specific issue.

Online surveys also let pollsters probe more deeply. The typical phone poll lasts 22 to 24 minutes, Schnur explains. “At that point you really are testing the patience of your respondents.” With Web surveys, the respondents don’t have to finish in one sitting, which means pollsters can ask more questions.

In an initial experiment in April, the online survey, for the sake of comparison, repeated all the questions of the telephone poll. The majority of responses fell within the margin of error, indicating that the online poll was accurate.

However, the online poll ran about five or six minutes longer than the phone poll and included a dozen additional questions.

The extra questions teased out some interesting nuances. The telephone poll indicated that 63 percent of voters disapproved of President Obama’s handling of gas prices. The online survey recorded the same 63 percent disapproving, but found that only 13 percent blamed Obama for the higher gas prices. Twenty-one percent blamed problems in the Middle East, and 38 percent blamed oil companies.

Online respondents also were more likely to mark “don’t know” when they had no strong opinion, or to choose “both” when given multiple options. Schnur sees these as more honest answers.

A subsequent Web-only survey, released early in June, leading up to the California primary elections, earned its own stripes for extreme accuracy. It reported that Rep. Brad Sherman had a 9.5 percent edge over Rep. Howard Berman in the newly redrawn 30th Congressional District. Sherman won at the polls by 10 percent. The survey results were obtained after respondents viewed campaign ads for both candidates.

The online survey is expected to run at least two more times before the November election.

If you have questions or comments on this article, go to tfm.usc.edu/mailbag.