Advancing measurement and scoring methods for testing mediation models of substance use - PROJECT SUMMARY/ABSTRACT
Research on the intergenerational transmission of substance use seeks to identify the various mechanisms or
mediators which connect parental substance use, an established risk factor, to adolescent substance use.
Parent-related mediators are measured using multiple informant data, which allows for a comprehensive
assessment of behaviors and reduces bias from any single reporter. However, current methods for analyzing
multiple informant data in mediation models do not separate the different sources of variance present in
multiple informant ratings, meaning that the variance that is unique to each parent is confounded with the
variance that is common across both parents. As a result, the estimates for mediated effects become biased,
and effects that are caused by one source of variance (e.g., the common behaviors between parents) may be
misattributed to a different source of variance (e.g., maternal-specific behaviors). As a consequence, our
understanding of intergenerational processes and ability to design effective preventive interventions is limited.
To address these limitations, a recently developed psychometric model, the trifactor model (TFM) can be used
to disentangle the sources of variance present in multiple informant data, allowing for more accurate estimates
of mediated effects, and more valid and specific substantive conclusions about the intergenerational processes
therein. The TFM separates the sources of variance in multiple informant data into three independent levels: a
common factor representing the shared variability across reporters, the perspective factors representing the
variability unique to each reporter, and the specific factors representing variance arising from the use of the
same items. Despite the great potential of the TFM, this model has not yet been evaluated within a mediation
model, nor compared to current methods. Moreover, this model is complex and most applications would
require latent variable scores in order to reduce the number of parameters, but there are a number of
challenges to obtaining scores for this model that would need to be addressed. In order to expand the
applicability of the TFM, the proposed research will (1) derive a single mediator model with the TFM and
compare performance to traditional multiple-informant methods, (2) extend the TFM to parallel mediation
model and develop factor scoring approaches, and (3) Apply novel TFM methods to empirical data to examine
parental monitoring as a mediator for the intergenerational transmission of substance use. The current NRSA
proposal presents a novel expansion of the TFM so that it may be used to uncover specific mechanisms that
lead to adolescent substance use. Further, this fellowship will provide the crucial training needed to establish
this applicant's independent research program in developing quantitative methods for the study of adolescent
substance use and prevention through training in statistical mediation, family-based models of substance use,
prevention science, and research dissemination.
