Across the multiverse: exploring a diverse set of specifications related to cross-sectional and prospective...

The relationship between adolescent alcohol use and emotional problems remains unclear and contradictory. These inconsistencies may in part be due to differences in the measurement and operationalization of alcohol use and emotional problems across studies, as well as confounder selection and missing data decisions. This study explores the associations between common specifications of adolescent alcohol use and emotional problems in a large sample of adolescents.

A multiverse analysis (also known as specification curve analysis or vibration of effects) was done with 7680 unique model specifications in a large longitudinal sample of 6639 Australian adolescents (aged ~14.7-15.7, 2021-2022).

While alcohol use and emotional problems nearly universally co-occurred in minimally adjusted cross-sectional models (98-99%), the operationalization of emotional problems, temporality of prospective relationships, and choice of confounders substantially impacted findings. Emotional problems appeared to predict later alcohol use more-so than the reverse, depression-focused measures yielded more consistent associations with alcohol use than anxiety-focused measures, and certain confounders (i.e. conduct, ADHD, smoking) explained most of the associations between adolescent alcohol use and emotional problems. Missing data decisions and whether outcomes were modelled continuously v. dichotomously had minimal impact on findings.

While adolescent alcohol use and emotional problems commonly co-occur, inconsistencies in the magnitude, direction, and significance of effects are closely tied to researcher decisions that are often made arbitrarily.