A Practical Guide to Causal Mediation Analysis: Illustration With a Comprehensive College Transition Program and Nonprogram Peer and Faculty Interactions

Abstract

Experimental and quasi-experimental designs have been increasingly employed in education. Mediation analysis has long been used to measure the role of mediators. Causal mediation analysis provides a modern approach to evaluate potential causal roles of mediators. Compared with conventional mediation analysis, causal mediation analysis has several advantages, e.g., by enabling us to evaluate necessary assumptions to establish a valid causal role of the mediator of interest. Despite these advantages and the availability of various software programs, causal mediation analysis has not been employed frequently in educational research. In this paper, we provide a step-by-step guide to causal mediation analysis using the free R package mediation in order to promote the more frequent application of causal mediation analysis in education, with an accessible data example from a Comprehensive College Transition Program (CCTP).

Investigating low birth weight and preterm birth as potential mediators in the relationship between prenatal infections and early child development: a linked administrative health data analysis

“…30 Unlike conventional mediation analysis, causal mediation analysis attempts to better infer causality by focussing on nonparametric definitions of causal mediation effects, and is better able to achieve confounder control than conventional mediation analysis. 31 It can also accommodate both linear and non-linear models and both continuous and discrete mediator/outcome variables. 31 The analysis for the present study was specifically carried out using a 'potential outcomes' framework via the Stata/ MP 16 'mediation' package, which estimates mediation effects using Monte Carlo simulations.…”

“…31 It can also accommodate both linear and non-linear models and both continuous and discrete mediator/outcome variables. 31 The analysis for the present study was specifically carried out using a 'potential outcomes' framework via the Stata/ MP 16 'mediation' package, which estimates mediation effects using Monte Carlo simulations. 30 32 33 This works by simulating predicted values of a specified mediator or outcome variable and then calculating appropriate quantities of interest.…”

Investigating low birthweight and preterm birth as potential mediators in the relationship between prenatal infections and early child development: A linked administrative health data analysis

“…Causal mediation analysis, as a method, plays an important role in helping to determine how and why particular treatment effects arise by helping researchers to identify intermediate variables that occur in the pathway between a given treatment and a given outcome [32]. Moreover, causal mediation analysis, in general, has multiple advantages over conventional mediation methods, such as allowing for the evaluation of necessary assumptions to ascertain the causal role of the mediator variable of interest [33]. The analysis for the present study was specifically carried out using a ‘potential outcomes’ framework via the Stata/MP 16 ‘ mediation ’ package, which estimates mediation effects using Monte Carlo simulations [32, 34, 35].…”

The effect of autistic traits on prosocial behavior: The chain mediating role of received social support and perceived social support

“…Based on the principal component factor analysis, the largest variance explained does not account for a majority of the variance (>50%), and in that case, it turns out that the CMB of this study does not pose a severe threat to our results. The analytic approach was then conducted in four steps, as recommended by Chi et al (2022).…”