Using classification trees to profile adolescent smoking behaviors

The purpose of this study was to explore the interactive nature of various predictor variables in profiling adolescent smoking behaviors characterized by intention to smoke, current, situational, and established smoking using classification trees. The data (n = 3610) were obtained from cross-sectional telephone surveys of the Florida Anti-Tobacco Media Evaluation Program. Three classification trees were constructed, namely, intention versus no intention to smoke among non-smokers, current smokers versus non-smokers, and established versus situational smokers. The tree model for the intention model revealed that social and health risks are important in the context of peer smoking. Certain variables such as peer smoking and alcohol consumption retained their relative importance across the tree classifiers demonstrating that smoking intention may be predictable using some of the same variables as in current or more dependent smoking.