Data were weighted to reflect the complex design of the NSDUH sample and were analyzed with STATA SE version 12.0 software (StataCorp, 2011). We used Taylor series estimation methods (STATA “svy” commands) to obtain proper standard error estimates for the cross-tabulations. Participants were stratified by survey year (2002–2012). For both perceived risk and past year use outcomes, cross-tabulations were performed separately for the total sample, as well separately for men and women; within sex, cross-tabs were calculated separately for each of the four race/ethnicities and age groups. Chi-square (χ2) tests were used to identify differences between 2002–2012 in each one of the following strata: 1) perceived great risk of regular cannabis use; 2) perceived risk of regular cannabis use among past year non-users, past year non-daily users, and past year daily users; and 3) past year non-daily cannabis use and past year daily cannabis use. Multivariate logistic regression analyses were used to determine the association between socio-demographic and cannabis use characteristics with perceived great risk of regular cannabis use. Analyses were adjusted for sex, age, race/ethnicity, education, total family income,
