Treating number of H.S. peers who smoked and age as continuous variables in logistic regression models makes the assumption that they are linearly associated with nicotine dependence. Testing this assumption using fractional polynomial analysis [28], no higher power polynomial fit the association of peer smoking or age with nicotine dependence better than the linear model (p ≥ 0.49 and p ≥ 0.45 for models up to 4 power terms, respectively). Treating these variables as continuous is appropriate. The resulting odds ratios can be interpreted as the increased risk of nicotine dependence for every additional H.S. peer who smoked (26%) and for every year older a participant was (3%).
