Data were analyzed using censored Tobit models with heteroskedasticity
corrections, with frequencies of alcohol-related problems
measured as counts of the number of events in the past year.
Models were run for each of the domains separately (Table 3). All
models included controls for demographic characteristics (age,
gender, race, parental education, weekly spending money), deviance,
and smoking status. For each problem outcome, block tests
were used to determine final models: Models were first run with
only demographic and psychosocial characteristics (Model 1);
next, context-specific frequencies of use were added (Model 2)