Short- and long-term prediction of self-reported cigarette smoking in a cohort of late adolescents: report of an 8-year follow-up of public school students.

The purpose of this study was to assess accuracy in predicting adolescent smoking status using attitude, knowledge, behavioral, and sociodemographic variables. A cohort of 4,641 schoolchildren was tested in 1975 (grades 4-6), in 1978 (grades 6-8), and, finally, in 1983 (grades 10-12). From 1978 to 1983, eight variables were used to account for 25% of the variance in smoking status. Over the 8-year period, six variables accounted for 13% of the variance. In both models, prior experience with cigarettes, peer and parental smoking, sex, and student intention were used. Tests of the Ajzen-Fishbein model and social learning theory indicate that social factors were more important than attitudinal ones in predicting future smoking, but these relationships were relatively weak. However, for both long- and short-term prediction, previous behavior proved to be the best predictor. Implications for the design of prevention programs are discussed.