Substance use is one of the most commonly occurring health risk behaviors in adolescence, making it significant from both developmental and public health perspectives. Adolescent alcohol use alone has been estimated to cost the United States around 60 billion dollars in a single year (Miller, Levy, Spicer, &Taylor, 2006). Perhaps even more concerning are the associations among adolescent substance use and psychological impairment, injury, and even death (USDHHS, 2007). These serious public health issues have propelled substantial growth in the theoretical conceptualization of pathways to substance use during adolescence such as theoretical risk models that identify potential developmental markers specific to the internalizing pathway to substance use. However, researchers'ability to rigorously evaluate these developmental theories has been restricted by the standard statistical models currently available. In adolescent substance use research, findings are frequently based on linear statistical models fitted to count or ordinal drug use data because alternative methods are not well studied or readily available. Statistical theory predicts that this type of model misspecification can lead to unreliable and invalid statistical tests of substantive theory (Agresti, 2010;Long, 1997), but the extent to which this occurs in practice is currently unknown. Additionally, adolescent substance use research faces zero-inflation, a phenomenon that arises because there are an extremely large number of zeros (e.g., adolescent who are not currently using drugs). Standard practice in adolescent substance use research is to ignore this zero-inflation, which limits the breadth of possible theoretical models that can be tested and can lead to incorrect statistical tests of substantive theory (Tu, 2002). Fortunately, drawing on academic fields outside of the behavioral sciences, these shortcomings of standard models can be overcome by introducing a new class of statistical models that are well suited for the empirical characteristics of adolescent substance use data. More specifically, these innovative models have the potential to reliably and validly assess multifaceted theoretical models of the etiology and development of drug use in ways not currently available (e.g., by accounting for theoretically meaningful unobserved heterogeneity among adolescent not currently using drugs). However, this new class of statistical models still needs to be rigorously studied in the context of adolescent substance use. The three core aims of this project are to (1) empirically examine existing statistical models used in substance use research;(2) empirically examine novel alternative models;and (3) apply these novel models to existing adolescent substance use data. Thus, the proposed project will fully integrate advanced quantitative methods with substantive theory so that researchers can reliable and validly test intricate developmental theories of adolescent substance use.
The proposed project will fully integrate advanced quantitative methods with substantive theory by investigating how researchers can optimally model developmental theories of adolescent substance use. This project will evaluate existing statistical models and innovative statistical models for adolescent substance use data. Secondary data analysis will show the utility and feasibility of innovative statistical models for testing novel hypotheses concerning the internalizing pathway to substance use.
|McGinley, James S; Curran, Patrick J (2014) Validity Concerns with Multiplying Ordinal Items Defined by Binned Counts: An Application to a Quantity-Frequency Measure of Alcohol Use. Methodology (Gott) 10:108-116|