Mixed effects models were estimated in the R statistical package version 2.13.1 using the lme function in the multilevel package (Bliese, 2008). Models were estimated using full information maximum likelihood estimation to allow for comparison of models with different fixed and random effects. In order to improve the stability of model estimation all outcome variables were centered. The best fitting model was determined via -2 log-likelihood change tests using the anova function with nested models. Estimated models took the form of: Level 1:Craving = β0i + β1i(Hours) + β2i(Hours2) + β3i(Hours3) + eitLevel 2:β0i = γ00 + γ01(Group) + u0iβ1i = γ10 + γ11(Group) + u1iβ2i = γ20 + γ21(Group) + u2iβ3i = γ30 + γ31(Group) + u3i When a significant interaction between Group and one or more growth parameters was observed, a series of post-hoc tests were conducted wherein Hours was successively re-centered at different values to determine the regions of time where moderate smoking and heavy smoking groups significantly differed on the outcome variable (i.e. regions where a significant simple effect of Group was observed at p < 0.05).
