This study began as an assignment for Beau Kilmer's Drug Policy class at Pardee RAND Graduate School. Over a year before designing and implementing this particular study, I had had a general familiarity with marijuana use questions from the NSDUH data. I noticed two trends: general prevalence of dependence in the population was stable, while the number of daily/near-daily was greatly increasing; therefore, I could infer without touching the data that prevalence of dependence among daily/near-daily users was almost definitely falling. Originally, the paper had a more broad scope: dependence prevalence among past-month users of different use intensities. However, to simplify and sharpen the research, the scope was narrowed to examine only daily/near-daily users. I then had planned to plot time trends to confirm these trends and check their robustness to demographic covariates, and further to examine trends in the constituent dependence symptoms (e.g., needing more to get the same effect). The statistical procedure was not designed all up-front before viewing the data, as might have been preferred. But instead, it was designed in a more abstract sense, with the details developed as required to bolster the findings from the analysis. I first plotted these time trends, with subgroup analysis by age and gender group. Later, to confirm the significance of these results, I chose a quasi-binomial regression design, which found significant results; a Cochran-Armitage trend test was added at request of a reviewer and also found signficant results.