Although frequently hypothesized, the evidence for associations between affect and marijuana use in everyday life remains ambiguous. Inconsistent findings across existing work may be due, in part, to differences in study design and analytic decisions, such as study inclusion criteria, the operationalization of affect, or the timing of affect assessment. We used specification curves to assess the robustness of the evidence for affect predicting same-day marijuana use and marijuana use predicting next-day affect across several hundred models that varied in terms of decisions that reflect those typical in this literature (e.g., whether to average affect prior to marijuana use or select the affect report closest in time to marijuana use). We fitted these curves to data from two ecological momentary assessment studies of regular marijuana and/or alcohol using college students (N = 287). Results provided robust evidence that marijuana use was slightly less likely following experiences of negative affect and slightly more likely following positive affect. Specification curves suggested that differences in previous findings are most likely a function of the specific emotion items used to represent affect rather than differences in inclusion criteria, the temporal assessment and modeling of affect, or the covariates added to the model. There was little evidence for an association between marijuana use and next-day affect. Overall, our findings provide evidence against the predictions made by affect reinforcement models in college students and suggest that future research should model the associations of marijuana use with discrete emotional states rather than general negative and positive affect.
Supplementary notes can be added here, including code, math, and images.