A Monte Carlo Sensitivity Analysis Approach for Omitted Confounders in Latent Growth Curve Mediation Models

• In Event: Data Analysis and Statistics

Abstract

Parallel process latent growth curve mediation models (PP-LGCMMs) are frequently used for longitudinal mediation analysis. An important assumption in PP-LGCMMs is that there should be no omitted confounders of the relationships among treatment, mediator intercept and slope, and outcome intercept and slope. This assumption is often challenging to satisfy in practice. In this study, we analytically examined how the omission of pretreatment confounders impacts mediation inferences from the PP-LGCMM. With the analytical results, we developed a Monte Carlo sensitivity analysis approach for assessing the robustness of mediation inferences from this model. R functions and a user-friendly Shiny web app were developed for the implementation. A real-data example was included to illustrate the application of the developed approach.

Authors

• Xiao Liu, University of Texas at Austin
  Presenting Author

• Zhiyong Zhang, University of Notre Dame
  Non-Presenting Author

• Kristin Valentino
  Non-Presenting Author

• Lijuan Wang
  Non-Presenting Author