Male teenagers engage in risky behavior associated with HIV infection more than previously thought, and they are also much more willing to report such behavior if the survey is administered by computer, rather than on paper, a new study suggests.

The research appears to call into question much of the data that has been gathered on sensitive subjects, like drug use and sexual habits, and to confirm what re-sear-chers have long assumed to be under-reporting in such areas. The more socially stigmatized the behavior, the greater the disparity in the results between the two reporting methods.The results of the study, published Friday in the journal Science, have significant implications both for public health and for the understanding of how people relate to computers. The authors said the findings were striking and unexpected because paper questionnaires were believed to adequately address the privacy issues that are known to inhibit honest responses in face-to-face or phone interviews.

"We had the analysis done and re-done, but this is real," said Charles Turner, director of the program in health and behavior measurement at the Research Triangle Institute in Washington and co-author of the study. "It means everything you thought about the risks adolescents face is an underestimate, if you're deriving your perceptions from past surveys."

Among the 1,600 male respondents ages 15 to 19 surveyed nationally, those who listened to questions on earphones and viewed them on a computer screen were almost four times as likely as the paper and pencil group to report some type of male-male sex (5.5 percent vs. 1.5 percent); 14 times as likely to report sex with an intravenous drug user (2.8 percent vs. 0.2 percent); and 5.5 times as likely to report that they were "always" or "often" drunk or high when they had heterosexual sex (10.8 percent vs. 2.2 percent).

The study, financed by the National Institutes of Health, did not test whether the teenagers might be exaggerating in their computer-assisted responses. But the authors and other social scientists said the results were consistent with what is generally believed to be more accurate levels of the reported behavior than has shown up in previous data.