Advisory Committee Chair
David L Roth
Advisory Committee Members
Inmaculada Aban
Cynthia J Brown
Leslie McClure
Gerald McGwin
David T Redden
Document Type
Dissertation
Date of Award
2011
Degree Name by School
Doctor of Philosophy (PhD) School of Public Health
Abstract
The use of mediation modeling is quite prevalent in a number of disciplines to answer questions about how or why one variable exerts its influence on another variable. Although mediation can be assessed in the context of several types of study designs, the use of cross-sectional data and a single-mediator model tend to be the most commonly reported features in empirical tests of mediation. There are several limitations associated with assessing mediation with cross-sectional data, perhaps the most significant is that mediated effect estimates are biased in the case of true longitudinal mediation. For this and several other reasons, there has been a greater emphasis on the development of longitudinal mediation models. There are several classes of models for evaluating longitudinal mediation with the collection of three or more waves of data. These models are increasingly utilized in the applied literature and methodological research continues to evaluate them, as well as extensions and new approaches. Despite their use in substantive research, the preponderance of mediation hypotheses are still tested with cross-sectional data. Furthermore, consensus on the optimal implementation of longitudinal mediation modeling methods is largely lacking and there are many unanswered questions. In the first paper of this dissertation we sought to demonstrate the application of one approach to longitudinal mediation modeling, namely the autoregressive model, and build on a set of steps recommended for testing such models. In the other papers we attempted to address significant methodological questions related to different methods of longitudinal mediation modeling. Paper two explored the effects of unreliability and the failure to account for shared method variance in the autoregressive mediation model. Paper three evaluated the statistical performance of methods used to test mediation in a two-stage piecewise parallel process latent growth curve model and examined the impact of misspecifying the true piecewise model as a single-stage parallel process model. Although mentioned in the literature as a possible method to address a substantial criticism associated with parallel process mediation models (i.e., the inability to delineate temporal ordering), piecewise growth models are rarely utilized in substantive research and lack a full elaboration in the literature.
Recommended Citation
Bentley, John Paul, "An Examination of Statistical Methods for Longitudinal Mediation Modeling" (2011). All ETDs from UAB. 1145.
https://digitalcommons.library.uab.edu/etd-collection/1145