Cutting corners costs political pollsters credibility.

Byline: ARNOLD ISMACH For The Register-Guard

LAWYERS ARE ABOUT to get a reprieve. Long the chief targets of scorn and ridicule in American society, they're now in danger of losing that distinction. Political pollsters are rapidly taking over as our favorite objects of derision in the professional world. They deserve the distinction.

This month's general election capped decades of decline in polling accuracy. The broad gains by Republicans in Congress caught everyone by surprise - pundits as well as the public. It wasn't long before the blame for the unexpected was placed where it belongs: at the feet of the pollsters. Democrats were widely expected to gain seats in the House and solidify their control of the Senate. Most polls sponsored by the news media carried that message through October.

In the final week before the elections, a few major media polls (including CBS/NYT; USAT/Gallup/CNN and Zogby) reported at least a small GOP surge, hinting at a standoff. But none predicted the large Republican gains nationally, and some polls clearly showed the flaws in election surveys. In Minnesota, for example, one major pollster said Walter Mondale would win by 6 percent and another picked Norm Coleman, the victor, by an equal margin - a difference of 12 points.

In the early years of election prediction - from the 1940s through the 1970s - results were usually on the mark. Why the downward spiral in polling accuracy since then?

Some critics blame bias on the part of the pollsters, such as slanted questions. But there's no motivation for polling organizations hired by the news media to gear the results. Their business success rests on accuracy.

Others fault the concept of opinion research itself, arguing that it's not a science and that accuracy can't be expected. That view runs counter to evidence demonstrating that opinion surveys, conducted according to established methods, can be reasonably precise. We see that in the academic world, where tested theory and rigorous standards govern survey research. The same can be said of polls taken for candidates themselves - polls that the public rarely sees. Why, then, do we see failures in the horse-race political polls conducted for the news media?

In my opinion, it all comes down to one factor: money. Following approved practices for opinion surveys means spending more dollars for larger samples and higher response rates - more than polling firms or news media seem willing to spend. Many political pollsters hide their cost-cutting practices by not revealing those two key elements, sample size and response rate. They merely disclose what they call "error margin."

Error margin refers to sampling. Statistical theory tells how close to reality samples of a certain size will be, just as similar probability theory can tell us how many times the number seven will appear in a thousand rolls of the dice.

Small samples lead to lower accuracy rates. Large samples lead to higher accuracy. It's as simple as that. Thus, a sample of 600 may give a sampling error margin of plus or minus four percent, while a sample of 2000 may reduce that margin to two percent.

The theory underlying sample surveys also dictates that there must be a random, unbiased procedure to select respondents, and that all respondents selected must participate if the results are expected to accurately reflect the larger population.

The problem with news media election polls is that survey organizations rarely follow these procedures. They cut corners to save a buck. The result is almost inevitable: data that don't accurately reflect the population.

Samples for election polls must reflect the voting population, which in this last election was 39 percent of registered voters nationally. That means that the 61 percent who did not vote should not have been included in the pre-election samples.

Typically, political pollsters will randomly select telephone numbers to obtain people for their samples. If someone answers the phone and that person is a registered voter, they become part of the sample. Interviewers will ask if the person is registered to vote and plans to vote. Most people answer "yes" to those questions; it's embarrassing to admit you're a non-voter, so many lie. Thus, answers from non-voters are mixed with those of actual voters.

What they should do is follow a practice of selecting likely voters by checking voting records and picking names randomly only from those who voted in recent elections. That's how samples for candidates' own surveys are usually chosen.

If they use telephone sampling, they should also randomly select respondents from among all likely voters in each household, not just accept anyone who answers the phone. There are simple methods to achieve that randomness - but they take time, and thus increase costs.

Perhaps the biggest source of error in media polls, however, is the low response rate - often below 50 percent because many people today screen out unwanted calls through caller-ID. Even worse, many people today use cell phones exclusively, which means they aren't listed in the telephone directories that many media pollsters use to draw up sample lists.

Good survey research requires samples to be drawn from pools that include all potential members of the population under study. That means that political pollsters should draw their lists from voting records, selecting specific individuals. Or they can draw the numbers randomly from lists that include all possible phones and devise better techniques to weed out non-voters.

It also means they should call back several times if the person chosen for the sample isn't available when first called. The academic standard is usually four callbacks, which produces positive results.

Finally, pollsters should compare the characteristics of their obtained sample with the larger public to see that they haven't under-represented significant voting segments of the public. If the proportion, say, of women or of African Americans is significantly higher or lower than it is in the general population, the data must be weighted to account for that, or the pollster needs to obtain a separate sample for that group. Those steps, in addition to using survey question wording that is logical, clear and unbiased, would produce data that is worth our trust.

Finally, the media should report fully on the sample used: the number surveyed, the response rate, and the margin of sampling error - explaining what those numbers signify. Most reputable survey firms provide that information, as dictated in the conduct code of their national professional organization, the American Association of Public Opinion Research.

If news media organizations and the polling firms they employ started following these tested rules of proper survey research, it wouldn't be long before lawyers regained their hard-won status as our most demeaned professionals.

Arnold Ismach is the retired dean of the School of Journalism and Communication at the University of Oregon, and a specialist in public opinion research, and a former newspaper editor.