Moderating Effect in Second Order Construct

Frequently asked questions about PLS path modeling.
Post Reply
PLS Junior User
Posts: 1
Joined: Tue Mar 30, 2021 8:59 pm
Real name and title: Leonie

Moderating Effect in Second Order Construct

Post by Leonie »


I am working in a PLS-SEM model with an endogenous reflective first-order, formative second-order construct.

The exogenous variable is dichotomous (it is an experiment and I have read in an article of Henseler et al. (2016) that this is possible) and additionally I have another "normal" reflective-specified endogenous construct.

I need to analyse moderating effects between the dichotomous exogenous construct and the second-order construct. I want to dummy-code all my moderators (e.g. gender and age are moderator variables), so I have categorical, dichotomous moderator variables as well.

Until now I have used both the Extended-Repeated-Indicator-Approach and the Two-Stage-Approach to assess the second-order constructs and to assess the structural model.

To integrate the interaction term and analyse the moderating effect, and as I am having a categorical exogenous variable and categorical moderator variables, I would use the Two-Stage-Approach.

1) Can I just add the moderator variable in the path model which I am having when I use the Extended-Repeated-Indicator-Approach? Or would that be wrong?

2) When I add the moderator variable in the second stage while having used the Two-Stage-Approach (for assessment of HCM and structural model), the procedure works out and I get better results for the moderator effects than with using the Extended-Repeated-Indicator-Approach. But is there a problem with interpreting the results because I added new latent variable scores in the second stage for the assessment of the second-oder and the structural model? Or can I just treat and interprete it like a "normal" moderating effect?

3) I have one moderator variable which splits my experimental group into two subgroups. But the control group is not affected from this moderator variable. I have two ideas how to handle that: I just add this variable as a moderator as the others, too. (But this gives me kind of weird results, e.g. path coefficients of 5)
Or I just look at the experimental group and use the moderator variable as the only independent variable.
Are these ways to treat this variable? Which one is the better idea? Or would both be wrong?

Happy and thankful for your help,
Post Reply