New statistical methods inspired by data collected from alcohol and substance abuse research

This chapter provides a nontechnical review of new statistical methodology for longitudinal data analysis that has been published in statistical journals in recent years. The methodology has applications in four important areas: (1) conducting variable selection among many highly correlated risk factors when the outcome measure is zero-inflated count; (2) characterizing developmental trajectories of symptomatology using regression splines; (3) modeling the longitudinal association between risk factors and substance use outcomes as time-varying effects; and (4) testing measurement reactivity and predictive validity using daily process data. The excellent statistical properties of the methods introduced have been supported by simulation studies. The applications in alcohol and substance abuse research have also been demonstrated by graphs on real longitudinal data.