Simone, Melissa (2018). Latent difference score mediation analysis in developmental research: A Monte Carlo study and application.
Developmental and prevention researchers aim to determine how maladaptive health behaviors emerge. Mediation analysis offers a tool to identify the processes through which early risk factors influence later health. Recent quantitative developments offer several longitudinal mediation models through which researchers can examine how these effects unfold over time, rather than modeling all effects simultaneously through the application of traditional mediational models. Among existing longitudinal mediation models, latent difference score mediation stands out due to its unique ability to model changes both within and across individuals, as well as its ability to capture non-linear change over time. However, latent difference mediation lacks empirically supported sample size guidelines, which has resulted in few applications of this method. To address this limitation, the aims of this project were threefold: (1) evaluate the performance of various latent difference score mediation model structures through a Monte Carlo simulation study; (2) use the results from the Monte Carlo study to develop a set of empirical sample size guidelines for future use of latent difference score mediation; and (3) apply one of the latent difference score mediation model structures to real prevention data, to examine the underlying processes between disordered eating among adolescent girls and alcohol misuse among adult women. First, a Monte Carlo simulation study was conducted in which power, parameter and standard error biases, and coverage were examined across several latent difference score mediational model structures and population models to determine the required sample size for each structural and population model. Empirical sample size guidelines were determined in an iterative fashion, in which models with too much power were reevaluated with a larger sample size. The resulting sample size guidelines represent the smallest possible sample size with adequate power, minimal biases, and adequate coverage for each model. Latent difference score mediation was then applied to real prevention data, to examine how disordered eating among adolescent girls exerts its efforts on alcohol misuse among adult women.
Utah State University
City of Publication