Hi
I have some doubt in moderator variable analysis.
I have 4 independent variables(construct) V1 ,V2, V3, V4 all reflective and one dependent variable D
I modelled in PLS like
V1--->D
V2--->D
V3--->D
V4--->D
V2--->V1
V3---->V1
Here V1 is having a direct effect as well as acting as a mediator for variables V2 and V3
I want to know can i use V2 and V3 also as moderators in this model. I want to see their moderating effect on the relation of V1 and D
That is can i model all my variables together like
V1--->D
V2--->D
V3--->D
V4--->D
V2--->V1
V3---->V1
V2*V1-->D
V3*V1--->D
In short
V1-->D
V2-->V1--->D
V3-->V1--->D
V4-->D
V2*V1-->D
V3*V1--->D
Please help me in clearing my doubt.
Thanks
[/img]
moderator variable path diagram question
Mediator variable as a moderator
Hello Stefan,st8472 wrote:Would it also be ok to use a mediator variable as a moderator for another relation?
V1 -> V2 -> D
V3 -> D
V2*V3 -> D
have you got an answer to your question, may be from elsewhere?
I am thinking about using scores of a mediator variable for multi group analysis.
Regards,
Reinhard
-
- PLS Junior User
- Posts: 3
- Joined: Sun Nov 15, 2009 11:38 pm
- Real name and title:
Re: Mediator variable as a moderator
Hi Stefan.
Maybe you've already worked your question out, but here are my humble thoughts on the matter if you haven't.
Most of my work in PLS is exploratory, rather than confirmatory, and I've run across the same question before. If any are dichotomous, you can use the so called PLS-MGA. (See Reinartz, Haenlein, Henseler, 2009 - IJRM, but note that equation 11 is in error, get the spreadsheet mentioned therein from Henseler, or I'll send it to you - if you don't already have it.)
I've always felt a little uneasy about using the same IV twice in product-term interactions (e.g. V1*V2 --> D simulataneously with V1*V3 -->D). Part of my uneasiness is due to the concern of multicollinearity mentioned earlier... Part due to sometimes not having good theoretical grounds (e.g. fishing)... and Part due to a modelling philosophy that values parsimony.
Further to the point on parsimony, it's hard for me to imagine a very useful theory that has V1*V2, V1*V3, and V2*V3... that's every possible pairwise combination of V1, V2 and V3 isn't it? ... Not that I've not done the same thing before ;).
Perhaps you could try running the model with only a single interaction at a time and seeing what works out as the *most* explanatory? It would be a lot easier to write up if you only had a single prominent interaction.
Regards,
Hal
Maybe you've already worked your question out, but here are my humble thoughts on the matter if you haven't.
Most of my work in PLS is exploratory, rather than confirmatory, and I've run across the same question before. If any are dichotomous, you can use the so called PLS-MGA. (See Reinartz, Haenlein, Henseler, 2009 - IJRM, but note that equation 11 is in error, get the spreadsheet mentioned therein from Henseler, or I'll send it to you - if you don't already have it.)
I've always felt a little uneasy about using the same IV twice in product-term interactions (e.g. V1*V2 --> D simulataneously with V1*V3 -->D). Part of my uneasiness is due to the concern of multicollinearity mentioned earlier... Part due to sometimes not having good theoretical grounds (e.g. fishing)... and Part due to a modelling philosophy that values parsimony.
Further to the point on parsimony, it's hard for me to imagine a very useful theory that has V1*V2, V1*V3, and V2*V3... that's every possible pairwise combination of V1, V2 and V3 isn't it? ... Not that I've not done the same thing before ;).
Perhaps you could try running the model with only a single interaction at a time and seeing what works out as the *most* explanatory? It would be a lot easier to write up if you only had a single prominent interaction.
Regards,
Hal
reini_m wrote:Hello Stefan,st8472 wrote:Would it also be ok to use a mediator variable as a moderator for another relation?
V1 -> V2 -> D
V3 -> D
V2*V3 -> D
have you got an answer to your question, may be from elsewhere?
I am thinking about using scores of a mediator variable for multi group analysis.
Regards,
Reinhard