2.3. Statistical Analyses

We modeled the two event outcomes—progression to established, current smoking and progression to established, frequent smoking—using a logit model with a menthol use indicator as the key explanatory variable along with other covariates. Based on this framework, we estimated the effect of prior menthol use on the odds of event occurrence rather than a risk of event occurrence. In our model we have 4 periods: (P1) wave 1 to wave 2, (P2) wave 2 to wave 3, (P3) wave 3 to wave 4, and (P4) wave 4 to wave 5. We observe the event for each period at the end of the period, i.e., observe event for P1 at wave 2 (e2), event for P2 at wave 3 (e3), event for P3 at wave 4 (e4), and event for P4 at wave 5 (e5). We observe menthol status at the beginning of each period, i.e. we examine e2=f(menthol at wave 1), e3=f(menthol at wave 2), e4=f(menthol at wave 3), and e5=f(menthol at wave 4). That is, we examine e_t=f(menthol status_t-1).

We examined three model specifications for each event: Model 1 (base model) includes menthol use at a previous wave and age as explanatory variables; Model 2 includes Model 1 variables in addition to gender, race/ethnicity, and a control variable measuring whether the respondent lives with a tobacco user; and Model 3 includes all variables from Models 1 and 2 plus sensation seeking. Observations with missing data were excluded from the models.

We conducted several sensitivity analyses. In Appendix A, we estimated a model to account for youth who progressed to established smoking relatively quickly by progressing from nonsmoking to established smoking within the period between two waves. These respondents experienced the event of progression to established smoking but did not have a measure of menthol use at the beginning of a period because they did not report current smoking at that wave. In this sensitivity analysis, we investigated using same-wave menthol use for these respondents. In Appendix B, we estimated models using two new specifications of menthol use: (1) a set of dummy variables defining never use of menthol, last wave use of menthol, or prior use of menthol that was not last wave use, and (2) a set of indicators measuring never use of menthol, those who switched between menthol and non-menthol (or vice-versa), or those who only used menthol. These analyses assess whether different patterns of menthol use (e.g., for those who use menthol consistently versus who switch between menthol and non-menthol use) are differentially related to progression. All analyses were conducted with Stata version 15.1.