Help! Higher order construct model with mediation analysis?

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
olivia.lee
PLS User
Posts: 11
Joined: Thu Nov 26, 2015 12:18 pm
Real name and title: olivia F.LEE

Help! Higher order construct model with mediation analysis?

Post by olivia.lee »

Dear all,

I have a bit complicated model as below.
Independent variable A, a mediator B, dependent variable C(so, A affects B and C, B affects C). A is a second-order construct. A has 3 first-order constructs.

My question is

1. For A, there are some issues raised regarding the validity of the reflective-reflective type of second-order construct around 10 years ago. However, Sarstedt et al. 2019 explain the use of reflective-reflective type variables in a detailed manner. Is the concern regarding reflective-reflective resolved?

2. Using the second-order construct A, do I allowed to conduct a mediating analysis for the target variable of C? So I will compare the direct impact of A to C and the indirect impact of A to C via B.

Best regard,
Oliva from somewhere
jmbecker
SmartPLS Developer
Posts: 1282
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Help! Higher order construct model with mediation analysis?

Post by jmbecker »

Regarding your question 1:
I personally have problems finding good arguments for reflective-reflective higher-order models in general, because the main issue here is that your have to argue why you have distinct sub-dimensions (lower-order components) of a reflective (and thus uni-dimensional) higher-order construct. Yet, I have also seen papers where the authors have quite convincingly argue for why this model makes sense. Thus, I would not generally rule out the model, but always critically evaluate whether this setup actually makes sense. I think in this sense, we also mention such problems (at least in our latest paper where we cover higher-order model: https://www.emerald.com/insight/content ... /full/html). Yet, our main focus is to provide guidelines for how to model and evaluate the different models, without making definite judgments about the applicability of them, because that is always a case-by-case decision.

Nevertheless, I personally also believe, that in many situations a reflective-formative setup makes much more sense, because as soon as you argue for distinct sub-dimensions you somewhat approach formative measurement. Yet, you also have to convincingly argue for such type of measurement.

Then I think you should also ask yourself, why not modelling the sub-dimensions directly? Of course, parsimony is always an argument, but the differential effects of your lower-order components on the mediator and final outcome could also be interesting.

Overall, I think it all depends on the specific research model and not general recommendation can be given here. Good arguments and theoretical justifications are always key to choosing the right approach to measurement.

2: why not?
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
olivia.lee
PLS User
Posts: 11
Joined: Thu Nov 26, 2015 12:18 pm
Real name and title: olivia F.LEE

Re: Help! Higher order construct model with mediation analysis?

Post by olivia.lee »

jmbecker wrote: Fri Dec 16, 2022 1:32 pm Regarding your question 1:
I personally have problems finding good arguments for reflective-reflective higher-order models in general, because the main issue here is that your have to argue why you have distinct sub-dimensions (lower-order components) of a reflective (and thus uni-dimensional) higher-order construct. Yet, I have also seen papers where the authors have quite convincingly argue for why this model makes sense. Thus, I would not generally rule out the model, but always critically evaluate whether this setup actually makes sense. I think in this sense, we also mention such problems (at least in our latest paper where we cover higher-order model: https://www.emerald.com/insight/content ... /full/html). Yet, our main focus is to provide guidelines for how to model and evaluate the different models, without making definite judgments about the applicability of them, because that is always a case-by-case decision.

Nevertheless, I personally also believe, that in many situations a reflective-formative setup makes much more sense because as soon as you argue for distinct sub-dimensions you somewhat approach formative measurement. Yet, you also have to convincingly argue for such a type of measurement.

Then I think you should also ask yourself, why not modelling the sub-dimensions directly? Of course, parsimony is always an argument, but the differential effects of your lower-order components on the mediator and final outcome could also be interesting.

Overall, I think it all depends on the specific research model and not general recommendation can be given here. Good arguments and theoretical justifications are always key to choosing the right approach to measurement.

2: why not?
Oh my godness!
My research guru has replied to my question! Very nice to hear from you!
I just printed out your guideline paper and extensively am reading it.

For the first question, I have designed the reflective-formative variable but i do not have the global scale in my questionnaire. This means I do not have a chance to go to the reflective-formative one. Am I right?

With the second one, do you have any good example paper doing mediation analysis with a higher-order dependent variable? I notice it is not a widely used approach.

Hope you can help me.

XXX
Olivia (phd is almost finished!)
Malley
PLS Junior User
Posts: 1
Joined: Thu Sep 07, 2023 6:51 am
Real name and title: Hilda Murphy

Re: Help! Higher order construct model with mediation analysis?

Post by Malley »

olivia.lee wrote: Mon Dec 19, 2022 8:13 am
jmbecker wrote: Fri Dec 16, 2022 1:32 pm Regarding your question 1:
I personally have problems finding good arguments for reflective-reflective higher-order models in general, because the main issue here is that your have to argue why you have distinct sub-dimensions (lower-order components) of a reflective (and thus uni-dimensional) higher-order construct. Yet, I have also seen papers where the authors have quite convincingly argue for why this model makes sense. Thus, I would not generally rule out the model, but always critically evaluate whether this setup actually makes sense. I think in this sense, we also mention such problems (at least in our latest paper where we cover higher-order model: https://www.emerald.com/insight/content ... /full/html fnaf). Yet, our main focus is to provide guidelines for how to model and evaluate the different models, without making definite judgments about the applicability of them, because that is always a case-by-case decision.

Nevertheless, I personally also believe, that in many situations a reflective-formative setup makes much more sense because as soon as you argue for distinct sub-dimensions you somewhat approach formative measurement. Yet, you also have to convincingly argue for such a type of measurement.

Then I think you should also ask yourself, why not modelling the sub-dimensions directly? Of course, parsimony is always an argument, but the differential effects of your lower-order components on the mediator and final outcome could also be interesting.

Overall, I think it all depends on the specific research model and not general recommendation can be given here. Good arguments and theoretical justifications are always key to choosing the right approach to measurement.

2: why not?
Oh my godness!
My research guru has replied to my question! Very nice to hear from you!
I just printed out your guideline paper and extensively am reading it.

For the first question, I have designed the reflective-formative variable but i do not have the global scale in my questionnaire. This means I do not have a chance to go to the reflective-formative one. Am I right?

With the second one, do you have any good example paper doing mediation analysis with a higher-order dependent variable? I notice it is not a widely used approach.

Hope you can help me.

XXX
Olivia (phd is almost finished!)
I also wait for the answer for these question.
Post Reply