medflex: An R Package for Flexible Mediation Analysis using Natural Effect Models

Abstract

Mediation analysis is routinely adopted by researchers from a wide range of applied disciplines as a statistical tool to disentangle the causal pathways by which an exposure or treatment affects an outcome. The counterfactual framework provides a language for clearly defining path-specific effects of interest and has fostered a principled extension of mediation analysis beyond the context of linear models. This paper describes medflex, an R package that implements some recent developments in mediation analysis embedded within the counterfactual framework. The medflex package offers a set of ready-made functions for fitting natural effect models, a novel class of causal models which directly parameterize the path-specific effects of interest, thereby adding flexibility to existing software packages for mediation analysis, in particular with respect to hypothesis testing and parsimony. In this paper, we give a comprehensive overview of the functionalities of the medflex package.

Does Perceived Stress Mediate the Association Between Workplace Bullying and Long-Term Sickness Absence?

“…Estimation was done using the medflex package for R. 29 Workplace bullying and longterm sickness absence were included as dichotomous variables in the mediation analysis, while perceived stress were included as a 5-level categorical variable (split on the 20th, 40th, 60th, and 80th percentile).…”

Comparing Alternative Effect Decomposition Methods

“…The recently released medflex package in R allows for mediation analyses using imputation strategies for the estimation of natural effects. 38 This study contributes to the growing body of evidence showing the long-term health benefits of years of schooling.…”

Estimation and Inference for the Mediation Proportion

“…We refer the readers to [19], and references therein, for relevant discussions on assumptions needed and analysis conducted in order to avoid, or at least minimize, potential bias due to confounding when conducting mediation analysis. The difference method does not allow for mediator-exposure interaction, and alternative methods to allow for this interaction were previously developed [34, 38, 49]. …”