Predicting tobacco product initiation from intentions to use: Comparing the validity of item analysis methods

Items measuring tobacco use intentions are used to predict future use. Researchers combine items using different methods; however, no research has compared these methods’ predictive validity. Here, we compare how well six methods of analyzing four intention items predict initiation of cigarettes, e-cigarettes, snus pouches, and other smokeless tobacco one year later. We analyzed youth and young adult never users from the US Population Assessment of Tobacco and Health Study. We compared six methods of analyzing Wave 3 intention items in predicting Wave 4 use: susceptibility scoring (susceptible is not answering “definitely no” to all items); dichotomizing the four-item average using two cut-points on the 1–4 response scale; and dichotomizing one item (next year use intention) with three cut-points. Analyses (1) tested whether each single-item predicted initiation; and (2) compared each method’s (a) true positive rate (rate of correctly identifying future initiators), (b) true negative rate (rate of correctly identifying future non-initiators), and (c) model fit. Results were similar across products and age groups. Averaging items best predicted initiation in regression. Susceptibility scoring had the highest true positive rate but lowest true negative rate. False positives (incorrectly identifying someone as a future initiator) were best minimized by averaging items with a cutoff of 3, or using the single item with a 3 or 4 cutoff. Findings suggest researchers predicting tobacco use initiation using regression should average the four items; and researchers seeking to identify likely initiators should use different analytic methods depending on if they seek to maximize true positives or minimize false positives.