SNAPSHOT OF OPINION DID POLLSTERS INTERVIEW RIGHT PEOPLE?

Byline: Arnold Steinberg

Who's on first?

The Los Angeles Times last week said Lt. Gov. Cruz Bustamante. Other public polls, not taken at exactly the same time, said Arnold Schwarzenegger. If Bustamante wins, that doesn't mean the Times poll was right. If Arnold wins, that doesn't mean it was wrong. There is no cooking of numbers here. A poll is not a prophecy. It's a snapshot of reality, at the time the picture was taken.

So what's going on? There's no reason to automatically trust public polls. Some may cut corners. They may make headlines, but do they always make sense? Typically, they use random-digit sampling and then interview anyone claiming to be a voter. They use filter questions (for example, interest in the election) that may or may not define likely voters. In sum, we have no common agreement on how to define recall voters. Thus, the various public polls cannot be compared. It started right after Arnold announced. CNN and Time did a notoriously unreliable one-night poll (Aug. 9). Interviews were done only on the worst interview night, Friday. And Arnold had just announced. So his numbers were pumped.

But CNN already was hedging its bets with another joint poll, this one with USA Today and Gallup. This poll included people not running. Like Republicans Dick Riordan, former Los Angeles mayor, and Bill Jones, former California secretary of state. And Democrats like Loretta Sanchez, congresswoman from Orange County, and John Garamendi, state insurance commissioner. This poll's numbers are useless for two reasons. First, it included these noncandidates. Second, respondents could ``vote'' for more than one candidate. The candidate percentages totaled 169 percent. That didn't even include undecideds!

At least the Field Poll (Aug. 10-13) started after the close of candidate filing. But its sample of 448 likely voters is too small. Not because of the statistical margin of error. But because each candidate's name should be followed by the three-word ballot designation and political party. When you then survey for six or seven candidates in a telephone interview, respondent error may trump statistical error.

Then, there's the Public Policy Institute of California survey. It covers lots of issues. But the recall is an afterthought. Moreover, this survey (Aug. 16-21) was in the field too long. That was like a camera shutter staying open during an action shot. Indeed, this survey, like Field, included candidate Bill Simon, who then quit the race.

How about the very recent ABC Survey America (Aug. 23-25)? This telephone poll uses controversial and questionable automated interviewing and responses. It depends on people to answer honestly and accurately that they (a) are registered to vote, and (b) will vote in this election. Thus, ABC may be interviewing the wrong people.

Pollsters should be interviewing voters who voted in either the March 2002 primary or the November 2002 general election. Then they should add some voters, especially new registrants, who might be drawn into this high-profile election. Finally, they need to adjust data to reflect absentee voters.

A turnout model may need to be updated. For example, the Bustamante candidacy might require greater weighting for Latino voters. On a broader level, the recall itself may impact candidate voting. Thus, an overwhelming recall landslide could wipe out Bustamante.

Suppose this whole campaign turns negative. Once candidates start trashing each other, it's back to the drawing boards. In recent years, negative campaigns have reduced voter turnout. Arnold could draw new people to vote for him. Or Arnold could raise voter interest, paradoxically drawing low-propensity Democrats to support Davis.

Basically, a three-day tracking survey (with callbacks to reach those not at home) probably would be ideal. Even then, events might happen that affect the results. But analysis could adjust for intervening developments. I would ask an open ballot question, in other words, without prompting with choices.

Then, I would ask questions to measure the identification and perception of each major candidate. The actual ballot question would include the five or six major candidates, exactly as they appear on the ballot, but in rotated order. I would also vary the ballot question by several subsamples. Each subsample would have two or three different minor (placebo) candidates. (Remember, there actually are 135 candidates.)

I would push the undecided voters to make a choice. Then I would measure intensity and motivation. I would certainly ask people for their second choice.

Some of these questions would help measure voter propensity. But I would ask other questions to adjust for voter turnout. For example, I might test knowledge of the recall. As we get closer to the election, the people who know the date probably would be more likely to turn out. But simply asking people if they intend to vote is nearly useless. They tend to give the socially acceptable response. You'll soon be dealing with a bogus 96 percent voter turnout.

Remember, the order of candidates on the multipage ballot is not alphabetical. And there are 80 different ballot orders, one for each Assembly district. So a candidate cannot say vote for me, on line 52. This long ballot favors voters who are educated and motivated. Polling must find a way to measure who will negotiate this ballot maze.