Multiverse analysis code for
Advancing our understanding of cognitive development and motor vehicle crash risk: A multiverse representation analysis
Neurobiological and cognitive maturational models are the dominant theoretical account of adolescents’ risk-taking behavior. Both the protracted development of working memory (WM) through adolescence, as well as individual differences in WM capacity have been theorized to be related to risk-taking behavior, including reckless driving. In a cohort study of 84 adolescent drivers Walshe et al. (2019) found adolescents who crashed had an attenuated trajectory of WM growth compared to adolescent drivers who never reported being in a crash, but observed no difference in WM capacity at baseline. The objectives of this report were to attempt to replicate these associations and to evaluate their robustness using a hybrid multiverse – specification curve analysis approach, henceforth called multiverse representation analysis (MRA). The authors of the original report provided their data: 84 adolescent drivers with annual evaluations of WM and other risk factors from 2005-2013, and of driving experiences in 2015. The original analysis was implemented as described in the original report. An MRA approach was used to evaluate the robustness of the association between developmental trajectories of WM and adolescents’ risk-taking (indexed by motor vehicle crash involvement) to different reasonable methodological choices. We enumerated 6 reasonable choice points in data processing-analysis configurations: (1) model type: latent growth or multi-level regression, (2) treatment of WM data; (3) which waves are included; (4) covariate treatment; (5) how time is coded; and (6) link function/estimation method: weighted least squares means and variance estimation (WLSMV) with a linear link vs logistic regression with maximum likelihood estimation. This multiverse consists of 96 latent growth models and 18 multi-level regression models.