It’s high season for political polls, and if you listen to the people who run the surveys, television’s Dr. House is wrong when he says everyone lies at least some of the time. The pollsters contend few voters ever lie to any of them.

“I believe they tell us their opinions honestly,” says Mark DiCamillo, director of the prestigious Field Poll. “But they sometimes change their opinions.”

Adds longtime pollster Mark Baldassare, president of the Public Policy Institute of California, “People don’t lie to us.”

And yet…California political history is replete with examples of polls being wrong. There was the 1982 race for governor, won by Republican George Deukmejian after polls indicated a last-week surge toward Democratic rival Tom Bradley, then the mayor of Los Angeles. There was 1994, when most polls showed voters split virtually 50-50 just before the vote on the anti-illegal immigrant Proposition 187. Only the automated Survey USA, then a rookie outfit, showed 187 winning big just before the vote.

It carried by a 59-41 percent margin. There are more such cases, but those two illustrate the problem this year’s Prop. 19 poses for the polls, as it aims to legalize the use and growing of marijuana.

Polling results have been consistently inconsistent on this measure, as they were with 187.

Early on, it seemed legalizing pot would pass in a cakewalk. The two major polls measuring initial sentiment on it, Field and the automated Survey USA (which polls for TV stations in Los Angeles, San Francisco, San Diego and Fresno) showed it winning handily. But then a midsummer Field survey indicated a major shift. Suddenly, Prop. 19 trailed by 48-44 percent, looking like bad news for proponents since few initiatives ever pass when their support dips beneath the 50 percent level at that point in a campaign.

But just a few days later, here came Survey USA with the pot proposition up by 50-40 percent. Exactly one month later, another robo poll by the same company got the same result. Then, more recently, Survey USA had Prop. 19 up by only four points.

Why such differences, when the major polls generally are pretty close together, even if a couple – Survey USA and Rasmussen Reports, which is also automated – often tend to show Republicans running slightly stronger than other surveys?

One possible explanation is that some voters lie to live pollsters, as they apparently did in the Deukmejian-Bradley contest and about Prop. 187.

How Often Are People Candid?

“It’s not that people consciously lie,” said Jay Levy, president of New Jersey-based Survey USA. “What happens sometimes is that people think they will behave in a certain way, and then they don’t.”

This can happen, Levy suggests, in contests where voters feel they ought to vote one way, but find they can’t do it at crunch time.

“On 187,” he said, “people were asked if they would vote to throw children of illegal immigrants out of schools and deny them healthcare. Who would say ‘yes’ to another person on that? But in the end, that’s what they did.”

It only takes 10 or 15 people doing this in a polling sample of a few hundred to make things come up wrong.

In the Deukmejian-Bradley contest, with Bradley the first African American ever to win a major-party nomination for governor, it’s apparent that inaccuracy followed when enough of those polled didn’t want to admit they would vote against a black man – but still did it.

When campaigns involve issues where shame might be involved, it may be easier for some voters to reveal their true intentions by simply punching a button than confiding in a cold-calling stranger.

Reasons to Doubt Polls

“It might be a matter of not ‘confessing,’” Levy said. “If you tell a live person you’re voting ‘yes’ to legalize marijuana, some voters might believe they’d be seen as admitting they use it.”

There’s also the matter of how questions are asked and who answers them. The live polls read voters the full ballot title and summary just as they’ll see it in the voting booth; Survey USA and other robo polls use tighter summations of propositions.

“We know the ballot title and summary language mean a lot,” says Baldassare. “There’s also the question of who’s pushing the buttons on the phone with automated polls. It could be a teenage kid. We don’t really know. But I don’t believe shame or fear of exposure ever play into our results.”

Adds DeCamillo, “Nobody is asking if a voter smokes pot when we poll on Prop. 19. If we asked that, a live interviewer could understate its support, but we don’t.”

But it’s possible the issue itself implies such a question, just as asking about 187 or Tom Bradley also bore obvious but unspoken implications.

There’s also the fact that the live-operator surveys reach the 30 percent of Californians who have only mobile phones and no land lines, while robo polls don’t.

It adds up to a lot of question marks about which polls and which kinds of polls are most likely to be accurate.

Mr. Elias may be contacted at tdelias@aol.com.

His book, “The Burzynski Breakthrough,” is available in a soft cover, fourth edition. For more Elias columns, visit www.californiafocus.net