The replicability of cannabis use prevalence estimates in the United States

Recent changes in cannabis policy in the United States have prompted increased interest in cannabis use estimates, and replicability of these estimates. Here, we compare prevalence estimates from two concurrent approaches: (1) standard social survey with audio computer-assisted self-interviews at respondent's home (ACASI-H), derived from the National Surveys on Drug Use and Health (NSDUH), and (2) standard health survey with ACASI assessment in a mobile health examination vehicle (ACASI-M), derived from National Health and Nutrition Examination Surveys (NHANES), 2005-2012, with essentially the same standardized items in cannabis modules. NHANES ACASI-M prevalence proportions for recently-active-cannabis-use are an estimated 130-140% larger than corresponding NSDUH ACASI-H estimates (p < 0.05). In exploratory stratified analyses, we sought to understand these NHANES-NSDUH differences, and found no differences in lifetime-history-of-cannabis-use. However, for participants living with others in the same residence, moving cannabis assessment out of the dwelling unit might promote larger recently-active-cannabis-use estimates; no NHANES-NSDUH differences were found for participants living alone. The observed discrepancies might be non-ignorable in a policy or program evaluation context. A methods research program will be needed to account for between-survey differences of the type observed here, perhaps with a focus on within-residence versus non-residence assessment as a source of variation.