Mediation - binary variables

new2jasp

Hello, I'm trying to run a mediation model in JASP, which I've previously run using PROCESS in SPSS.

A few things are hindering this.

First, I cannot seem to enter a binary outcome variable - the arrow to select the variable into the 'outcome' box will not highlight.

Nor is it possible to select a binary variable as a mediator.

I can however select a binary var for the predictor.

Now, if I run a separate mediation analysis from this dataset (with scalar mediator and outcome), I get different (n.s.) results to what I've previously got in PROCESS. Why could this be? I've got the same number of boostraps, anything else I need to do to make sure its the same?

Thanks - enjoying the JASP platform, very accessible. Just a few doors I can't open so far!

EJ

Hi new2jasp,

I'll attend our expert to this issue!

Cheers

E.J.

EJ

Also, I am not sure but perhaps this helps:

https://forum.cogsci.nl/discussion/5976/auto-estimator-in-sem-mediation-module

E.J.

evankesteren

Hi new2jasp,

the mediation module is built on structural equation modeling in JASP. The PROCESS module works differently, using regressions. In many cases, they are exactly the same.

Be careful entering your binary variables as scale variables in the mediator or outcome boxes: this assumes they are "just" continuous and the model will behave as such. This will not give appropriate results.

The correct method for entering categorical variables is by changing the variable types to ordered categorical (ordinal) and enter them in the boxes. In this way, the variables will be treated as categorical observations of an underlying continuous variable (which in turn is assumed to follow the normal distribution).

Hope this helps!

Erik-Jan

asb

Hey evankesteren,

thank you for your help. I'm not sure if I understood your comment correctly. But if I scale my dichotome/binary variable as ordinal and insert it as mediator or outcome variable I get the following error message: FIML missing value handling only available when all endogenous variables are of scale type. Ordinal endogenous variables in the model: (binary variable). Do you know what I do wrong? Is it also possible to calculate with a binary dependent variable in a SEM?

Thank you very much and best regards

Alexander 

evankesteren

Hi Alexander,

yes, you should turn off FIML missing variable handling under "missing variables"! Then it should work.

Erik-Jan