Investigating the trend of adolescent alcohol use: a Bayesian hierarchical linear analysis

Traditional longitudinal models of adolescent alcohol use often assume that within-person variability-the extent to which an individual's alcohol use fluctuates over time-is the same for everyone. However, in the real world, some adolescents show relatively stable patterns of use, while others fluctuate substantially across different measurement waves. To capture these individual differences, the present study applies a Bayesian hierarchical linear modeling approach that allows between-person differences in within-person variability.

This study compared 16 model variants that crossed the following features: within-person error variance (fixed or random), lag effect (present or absent), linear and quadratic effects (random linear and quadratic terms included or excluded), and transformation (probit-transformed or untransformed).

The best-fitting model (DIC = -2,615; Deviance = -4,362) included a lag parameter, varying error terms, and random linear and quadratic effects for probit-transformed data. Notably, the average of the within-person standard deviation (${\mu}_{\sigma_j}$ = 0.231) was almost twice that of the varying intercept (${\sigma}_{\beta_{0j}}$ = 0.124), indicating substantial within-person variability in adolescent alcohol use across time.

Accounting for heterogeneous within-person variability and modeling alcohol use with a non-normal distribution significantly improved model-data fit and yielded extra insights into adolescents' alcohol use research. This approach allows researchers and practitioners to more accurately identify individuals with irregular or unstable drinking patterns, enhancing early detection and targeted intervention strategies.