keywords Path Diagram, Mediation, mediated moderation, moderated mediation, conditional mediation

In this page we show how to set up some mediation models without interactions. For models with interactions (conditional mediation, moderated mediation, etc) see jAMM: Building conditional models.

Simple mediation

To set up a simple mediation model, simply define one dependent variable, one independent variable and one mediator.

jAMM will define the models for you, as shown in the model info table.

Categorical independent variables

The same logic applies for categorical independent variables.

However, for categorical independent variables, the path diagram shows only one box for the IV, even though the estimation requires dummies (contrast variables) to estimate the effects. This is signaled in the notes.

Thus, since fact1 variable has 4 groups, in the results we will find three mediated effects, one for the contrast variable fact11, one for fact12, and one for fact11.

Multiple independent variables

You can add as many independent variables as you need, the model is updated, by default, accordingly with the simple rule that each IV is mediatated by the mediator.

Nonetheless, one may have in mind a different model (not necessarily sensible in all applications). Assume we want the mediated effects of x1 and x2 through m1, but we want the effects on y (and only them) to be computed after partializing x3. First, we go to Mediators models and remove the effect of x3 from the list of predictors. In doing so, the jAMM removes x3 also from the Full model. We go there and re-include x3 in the list of predictors. So we obtain the model we had in mind.

Notice the green path. That path is not estimated, but jAMM suggests to estimate it. That is because the model we just set up is not complete in terms of x3 mediation (recall that jAMM assumes you want to estimate mediation models).

If seeing the suggested paths disturb you, you can suppress them in the Mediation options tab.

Multiple mediators

Parallel mediators

By defining two or more variables as mediators in the variables input panel, we obtain a parallel mediators model by default.

You can put as many mediators as you want

Mediators chains

Several mediation models require building some sort of mediators chain. For instance, we can have that variable x is measured 4 times,and we want to see if x1 affects x2, x2 affects x3, and x3 affects x4. To build such a model, first we select x1 as the independent variable (it is the only exogenous variable in the model), x2 and x3 as mediators and x4 as dependent variable.

While we do so, jAMM builds a parallel mediators model (the default), but we ignore it and move on to change that. In fact, as compared with a parallel mediators model, the model we have in mind requires also that x2 be a predictor of x3. We go to Mediators models tab and add x2 in the field Mediator=x3. There we have it.

Multi-layer model

Assume you have x1 as independent, that influences x2 and x3, that in turn affect y through m1. We can think of this model as a multi-layer model in terms of layers of mediators: first the x2 and x3 layer, then m1 layer. To set it up, first we put all endogenous predictors in the mediators field and x1 and y in the independent and dependent field, respectively.

Again, we should change the automatically created parallel mediators model. We should add the paths from x2 and x3 to m1. To do so we go to Mediators models tab and add x2 and x3 in the field Mediator=m1.

Suprised by the ugly picture? Well, the model is correct but the picture can look better if one re-arranges the position of the terms. First, remove the mediators (jAMM does not like re-arranging terms already defined, so it is better to start anew). We want the last mediator m1 to be in the center of the diagram, so we should put it as the second mediator of the list. Then we change the Mediators models as we did before. Here is the input: a bit better looking.

Here is the output.

Multiple IV and multiple mediators

One can combine the multiple IVs and the multiple mediators models to better suit the research aims. By defining multiple IV and multiple mediators in the variale input tabs, one obtains, by default, a full parallel mediators model.

Obviously, one can then change the structure of the model, for instance by making the mediators a chain of mediators, or by removing some path from one IV to one mediator.

Notice that the previos model is not equivalent to two independent simple mediation models, because in the present model all the effects on the dependent variable y are computed keeping constant both independent variables. This feature is reminded by the notes of the path diagram.

Mis-specified models

There are models that may make a lot of sense in general, but are not mediation models. These models will not be estimated by jAMM, because they cannot produce mediated effects (that are the effects jAMM requires).

The first example is a model without mediators effect.

In this model I removed the path from m1 to y in the Full model tab. That creates a model in which the mediator has no role, and thus the path diagram shows a red path indicating that that path is required. It also signals the problem in the Models info table.

Another possible error is to have each mediator to predict all the other mediators. That would make the model recursive, and not a mediation model.

jAMM warns you of the problem with a purple path in the diagram, and with some explanation in the Models info table.

Examples

Comments?

Got comments, issues or spotted a bug? Please open an issue on GAMLj at github“ or send me an email