using the "moderating effect" function available i
using the "moderating effect" function available i
Hi,
I notice the calculation for the interaction effect is done automatically when we choose the "moderating effect" available once we right click on dependence construct. This is a great help especially when we have constructs with many many items.
My question:
1) Is the available "moderating effect" function for reflective construct only?
2) Can I use the "moderating effect" function if my predictor is in formative and moderator is in reflective?
THANKS.
I notice the calculation for the interaction effect is done automatically when we choose the "moderating effect" available once we right click on dependence construct. This is a great help especially when we have constructs with many many items.
My question:
1) Is the available "moderating effect" function for reflective construct only?
2) Can I use the "moderating effect" function if my predictor is in formative and moderator is in reflective?
THANKS.
nazif
-
- PLS Expert User
- Posts: 248
- Joined: Sat Jul 25, 2009 1:34 pm
- Real name and title:
Hey,
The product indicator approach, which is used in SmartPLS, is only applicable for reflective measurements. See for further information following article on p. 198 f:
Chin, W./Marcolin, B./ Newsted, P. (2003): A partial least squares latent variable modeling approach for measuring interaction effects: Results from a monte carlo simulation study and an electronic-mail emotion/adoption study, in: Information System Research, Vol. 14, No. 2, pp. 189-217.
Best regards,
Christian
The product indicator approach, which is used in SmartPLS, is only applicable for reflective measurements. See for further information following article on p. 198 f:
Chin, W./Marcolin, B./ Newsted, P. (2003): A partial least squares latent variable modeling approach for measuring interaction effects: Results from a monte carlo simulation study and an electronic-mail emotion/adoption study, in: Information System Research, Vol. 14, No. 2, pp. 189-217.
Best regards,
Christian
-
- PLS Senior User
- Posts: 28
- Joined: Thu Jun 24, 2010 5:38 am
- Real name and title:
-
- PLS Expert User
- Posts: 248
- Joined: Sat Jul 25, 2009 1:34 pm
- Real name and title:
For that, you need an another approach (two-stage approach).
How to do it, you can see in the following articel on page 86 ff.:
Henseler, J./Chin, W. (2010): A Comparison of Approaches for the Analysis of Interaction Effects Between Latent Variables Using Partial Least Squares Path Modeling, in: Structural Equation Modeling: A Multidisciplinary Journal, Vol. 17, No. 1, pp. 82-109.
Best regards,
Christian
How to do it, you can see in the following articel on page 86 ff.:
Henseler, J./Chin, W. (2010): A Comparison of Approaches for the Analysis of Interaction Effects Between Latent Variables Using Partial Least Squares Path Modeling, in: Structural Equation Modeling: A Multidisciplinary Journal, Vol. 17, No. 1, pp. 82-109.
Best regards,
Christian
Hi Christian,
what should I do if I want to test the moderating effect of a control variable (single measure, continous) on all other connections in my model (some reflective, some formative). Should I include all interaction term into my model at once or better one by one? group analysis is not preferred because of information loss...
best regards,
Carsten
what should I do if I want to test the moderating effect of a control variable (single measure, continous) on all other connections in my model (some reflective, some formative). Should I include all interaction term into my model at once or better one by one? group analysis is not preferred because of information loss...
best regards,
Carsten
-
- PLS Expert User
- Posts: 248
- Joined: Sat Jul 25, 2009 1:34 pm
- Real name and title:
Hey Carsten,
They following statement is only a opinion about this topic. Maybe one other has a better proposal.
Personally I would do it like Liang et al. (Assimilation of Enterprise System, in: MIS Quarterly, Vol. 31, No. 1, pp. 59-87, 2007) do it in there paper. The have tested if the control variables have a significant influence on the target variable. For PLS, in my opinion, that make much sense because it is a prognostic orientated procedure. You can also handle this problem like Huber et al. ( (2007): Kausalmodellierung mit Partial Least Squares – Eine anwendungsorientierte Einführung). At the last chapter of there book, the implemented a „covariate“ to handle that.
Best regards,
Christian
They following statement is only a opinion about this topic. Maybe one other has a better proposal.
Personally I would do it like Liang et al. (Assimilation of Enterprise System, in: MIS Quarterly, Vol. 31, No. 1, pp. 59-87, 2007) do it in there paper. The have tested if the control variables have a significant influence on the target variable. For PLS, in my opinion, that make much sense because it is a prognostic orientated procedure. You can also handle this problem like Huber et al. ( (2007): Kausalmodellierung mit Partial Least Squares – Eine anwendungsorientierte Einführung). At the last chapter of there book, the implemented a „covariate“ to handle that.
Best regards,
Christian
Hi Christian,
thanks for your reply. But I am not sure, if I correctly specified my request.
I checked your literature suggestions. Liang et al test 5 control variables. They check, if they do have a significant (direct) influence on the endogenous variables. If I do that, my three controls show no significant effect.
But I meant something else: I want to check the moderating effects of the controls on the other paths in the model, i.e. has control A an influence on the strength of the relation exogenous varible X to target variable Y. If I include the interaction terms, they show significant results.
for example: firm size has no direct effect on performance, but it does have an influence on the effect of lets say employee involvement on performance: Interaction term is negative and significant => in bigger firms involvement is less relevant for performance than in smaller firms. right?
questions:
1) Do I leave the control variables in the model when using interaction terms, even if they are not significant?
2) If I test more than one interaction term: Should I include all of them at the same time, or one by one in separate models?
3) Generally: Would you include variables in your model to test hypotheses, even if they show no bivariate correlation with the target variable? (they still could have an influence, right?)
To Huber et al: They do a group comparison, not interaction terms? Or did you mean something else?
cheers,
Carsten
thanks for your reply. But I am not sure, if I correctly specified my request.
I checked your literature suggestions. Liang et al test 5 control variables. They check, if they do have a significant (direct) influence on the endogenous variables. If I do that, my three controls show no significant effect.
But I meant something else: I want to check the moderating effects of the controls on the other paths in the model, i.e. has control A an influence on the strength of the relation exogenous varible X to target variable Y. If I include the interaction terms, they show significant results.
for example: firm size has no direct effect on performance, but it does have an influence on the effect of lets say employee involvement on performance: Interaction term is negative and significant => in bigger firms involvement is less relevant for performance than in smaller firms. right?
questions:
1) Do I leave the control variables in the model when using interaction terms, even if they are not significant?
2) If I test more than one interaction term: Should I include all of them at the same time, or one by one in separate models?
3) Generally: Would you include variables in your model to test hypotheses, even if they show no bivariate correlation with the target variable? (they still could have an influence, right?)
To Huber et al: They do a group comparison, not interaction terms? Or did you mean something else?
cheers,
Carsten
-
- PLS Senior User
- Posts: 28
- Joined: Thu Jun 24, 2010 5:38 am
- Real name and title:
-
- PLS Expert User
- Posts: 248
- Joined: Sat Jul 25, 2009 1:34 pm
- Real name and title:
Hi Fusion,
In your case I would use a group comparison for testing the moderation influence of age and gender. For building two groups you can use a median split for the variable age. How to perform the group comparison you can see described here:
viewtopic.php?p=3940&highlight=#3940
I hope that helps!
Christian
In your case I would use a group comparison for testing the moderation influence of age and gender. For building two groups you can use a median split for the variable age. How to perform the group comparison you can see described here:
viewtopic.php?p=3940&highlight=#3940
I hope that helps!
Christian
-
- PLS Senior User
- Posts: 28
- Joined: Thu Jun 24, 2010 5:38 am
- Real name and title:
Hi Christian,
I am really appreciate with your advice and suggestions.
If I want to test the 'age' as moderating effects and I have 5 groups on ages, let say ..below 25, 25 to 30 and so on. Then, I need to test all the 5 groups ?. Is it this a main idea of moderators effect in UTAUT?.
I try to to make all moderators as an aggregative effects, for example as depicted in the Figure below:-
The reason for that is I'm not test the effects of age groups e.g between PE to BI in UTAUT, but only the effect in total.
Thanks
I am really appreciate with your advice and suggestions.
If I want to test the 'age' as moderating effects and I have 5 groups on ages, let say ..below 25, 25 to 30 and so on. Then, I need to test all the 5 groups ?. Is it this a main idea of moderators effect in UTAUT?.
I try to to make all moderators as an aggregative effects, for example as depicted in the Figure below:-
The reason for that is I'm not test the effects of age groups e.g between PE to BI in UTAUT, but only the effect in total.
Thanks
Last edited by iris_afandiphd on Sun Mar 27, 2011 1:04 pm, edited 1 time in total.
-
- PLS Expert User
- Posts: 248
- Joined: Sat Jul 25, 2009 1:34 pm
- Real name and title:
You are welcome!
I cannot say, if this a main idea of moderators effect in UTAUT. But it seems for me reasonable to do it this way because your age variable is a single indictor with an impact of almost all paths in your model. You can split your gender variable in two groups e. g. the first two categories together with the last two categories together. But that is only an idea.
In your picture I see two critical points. One point is that the main effect of age is an antecedent of BI and USE at the same time. For modeling moderation effects I would separate the main effects of age, that means, I would modeling age as a separate variable for BI and for USE. That could have be an impact especially for a multi item construct. The second point is that I am missing the product variable FCs * age in your model if you want modeling the moderating effect of age between FCs and USE. But especially for age I would estimate two models and compare the paths of this two models. If you want to know only the total impact of gender on one variable e.g. USE you can put it as one antecedent variable in your model. But then it is not necessary to build product indicators for modeling moderators.
Best regards,
Christian
I cannot say, if this a main idea of moderators effect in UTAUT. But it seems for me reasonable to do it this way because your age variable is a single indictor with an impact of almost all paths in your model. You can split your gender variable in two groups e. g. the first two categories together with the last two categories together. But that is only an idea.
In your picture I see two critical points. One point is that the main effect of age is an antecedent of BI and USE at the same time. For modeling moderation effects I would separate the main effects of age, that means, I would modeling age as a separate variable for BI and for USE. That could have be an impact especially for a multi item construct. The second point is that I am missing the product variable FCs * age in your model if you want modeling the moderating effect of age between FCs and USE. But especially for age I would estimate two models and compare the paths of this two models. If you want to know only the total impact of gender on one variable e.g. USE you can put it as one antecedent variable in your model. But then it is not necessary to build product indicators for modeling moderators.
Best regards,
Christian
-
- PLS Senior User
- Posts: 28
- Joined: Thu Jun 24, 2010 5:38 am
- Real name and title:
I found that Journal Article from open access informaworldchristian.nitzl wrote:For that, you need an another approach (two-stage approach).
How to do it, you can see in the following articel on page 86 ff.:
Henseler, J./Chin, W. (2010): A Comparison of Approaches for the Analysis of Interaction Effects Between Latent Variables Using Partial Least Squares Path Modeling, in: Structural Equation Modeling: A Multidisciplinary Journal, Vol. 17, No. 1, pp. 82-109.
Best regards,
Christian
have a look at http://pdfserve.informaworld.com/886574__918536521.pdf