The study also has several limitations. First, our study cannot determine causality. For example, while some changes in perceived risk occurred around 2009, it does not mean that growth of the medical marijuana industry in Colorado was a main cause. States are not randomly assigned to a specific medical marijuana policy. It is likely states with more lenient attitudes toward marijuana are targeted for policy change. Using currently available data we cannot completely disentangle the causal direction from attitudes to policy. Second, the number of medical marijuana registry cardholders increased over the course of both 2009 and 2010. Ideally we could examine a more fine-grained temporal change in our outcomes of interest. However, the NSDUH design requires prevalence estimation using 2-year periods of data for Colorado. Third, examining years 2011-2012, 2013-14 and 2015-16 (instead of 2009-10), once those data are made available, will provide further clarity on temporal trends. However, policy positions are fluid. For example, in 2011 policy statements from the Justice Department (Grim, 2011) contradicted the earlier Ogden memo (Ogden, 2009), and in 2012, as noted earlier, Colorado
