Chapter 22 Moderated mediation
Mediation refers to a causal model in which the relation between the predictor and the dependent variable is mediated by one or more other variables (“the mediators). Note that mediation by definition entails strong assumptions regarding causality, which require longitudinal designs and almost always experimental independent manipulation of the predictor (and if possible mediator).
Moderation and mediation models can also be combined. The causal path between the predictor and the mediator(s) can be moderated by a moderator. Likewise, the path between the mediators and the dependent variable can be moderated by another moderator. A moderator strengthens or weakens the relation between two variables.
The path from predictor to mediator can be moderated by a moderator designated as”w“, while the path from mediator to dependent variable can be moderated by a moderator designated as”v". The PROCESS macro (SPSS) uses different model numbers to distinguish these models.
In these examples, the dataset/dataframe is called
dat, the predictor (the first variable in the causal chain) is called
predictorVariable, the mediators are called
mediatorVariable2, etc. (the intermediate variables in the causal chain), the dependent variable is called
dependentVariable (the last variable in the causal chain), and covariates are called
covariate2, etc. Finally, the moderators are called
22.1 jamoviFirst download the
jammmodule. Go to the Medmod menu and select ‘GLM mediation model’. Numerical predictor variables are dragged to the window, labeled ‘Covariates’. Mediators are dragged to the window ‘Mediators’. Example 1 is a mediation model without moderator and with two mediators.
By adding variables as mediators, covariates, or moderators, jamovi automatically builds a full model consisting of all terms, including interactions. By removing terms from the model window, all kinds of special models can be defined. In the output pane the defined model is visualised, so here it can be verified whether the intended model is computed.
In R the function is called
gemm(), which is in the package
rosetta. A tutorial can be found at: https://www.academia.edu/40039588/Tutorial_of_moderated_mediation_with_SEM_the_gemm_function or DOI: 10.13140/RG.2.2.22175.10404.
Example 1 is a mediation model without moderator and with two mediators; this model is called as follows, The data are in the R object
In example 2 a moderator of the x-m path is added to the model and only one mediator is used, which is called as follows.
In model 3 covariates are added. It is possible to choose on which variabele (mediators or dependent) the covariates have an assumed effect. In this example both covariates have an assumed effect on both the mediators and the dependent variable. Furthermore, the m-y path is moderated. Example 3 is called as follows.
In example 4 both x-m and m-y paths are moderated. There are two mediators. In this example the covariates have an assumed effect on both the mediators and the dependent variable. Example 4 is called as follows.
The output of the
gemm function can be put in an R object and can then be further inspected. The
plot functions are developed to provide relevant output of this analysis. Plots are only relevant when there is moderation, because they show the simple slopes or index of moderated mediation.
In SPSS the PROCESS macro can be used to analyse a (moderated) mediation model. Version 3 of this macro can be downloaded from http://www.afhayes.com. The SPSS code shown here is based on version 2. Several mediators can be added and also several covariates. A model code must be specified in this version. See Hayes(2013; 2018).
Example 1 has two mediators and no covariates.
PROCESS vars = dependentVariable, predictorVariable, mediatorVariable1, mediatorVariable2 / y = dependentVariable / x = predictorVariable / m = mediatorVariable1 mediatorVariable2 / model = 4.
Example 2 has one mediator, one moderator of the path from the predictor to the mediator (the x-m path) and no covariates.
PROCESS vars = dependentVariable, predictorVariable mediatorVariable, moderatorVariable / y = dependentVariable / x = predictorVariable / m = mediatorVariable1 / w = moderatorVariable1 / model = 7.
Example 3 has one mediator, one moderator of the path from the mediator to the dependent variable (m-y path) and two covariates.
PROCESS vars = dependentVariable, predictorVariable mediatorVariable1,moderatorVariable, covariate1 covariate2 / y = dependentVariable / x = predictorVariable / m = mediatorVariable1 / v = moderatorVariable2 / c = covariate1 covariate2 / model = 14.
Example 4 has two mediators, two moderators one for each path and two covariates.
PROCESS vars = dependentVariable, predictorVariable mediatorVariable1, moderatorVariable, covariate1 covariate2 / y = dependentVariable / x = predictorVariable / m = mediatorVariable1 mediatorVariable2 / w = moderatorVariable1 / v = moderatorVariable2 / c = covariate1 / model = 21.