Cannabis Use Prevalence and Trends (2016–2024): Estimating the Impact of Representation Bias in BRFSS Data

A novel method was applied to account for representation bias in U.S. cannabis prevalence estimates derived from the 2016–2024 Behavioral Risk Factor Surveillance System (BRFSS). Although the BRFSS provides national coverage, the cannabis module was only included in areas comprising about 19% (n = 961,366) of the total respondents. To account for differential module adoption across states, we combined observed cannabis prevalence by legal status from module-adopting areas with the distribution of legal status in non-module areas to generate adjusted national estimates. Areas that adopted the cannabis module were more likely to have legalized both recreational and medical cannabis use and less likely to have prohibited cannabis entirely. Specifically, they were 40% more likely to have recreational and medical legalization and 43% less likely to have total prohibition. Cannabis prevalence was substantially higher in legalized settings, with differences ranging from 91% (2016) to 40% (2024) compared to illegal states. Corresponding differences for medical-only legalization were 6% and 10%, respectively. Based on model-derived estimates, prevalence estimates derived from BRFSS cannabis-module data were approximately 7–8% higher than the corresponding model-based national estimates. This upward bias was more pronounced among older adults, women, Hispanics, and individuals with chronic diseases. However, temporal trends and subgroup patterns remained consistent between adjusted and unadjusted estimates. These findings suggest that while BRFSS cannabis module data may overestimate absolute national prevalence due to differential state adoption, they remain useful for tracking trends over time and across population subgroups. Adjusted estimates provide an alternative model-based framework for interpreting prevalence under different assumptions about jurisdictional composition.