It seems the test suggested by Kock (2015) relies in a less straightforward procedure. Actually, the concept of a "full collinearity test" comes from the paper of Kock and Lynn (2012), where in their Appendix F suggest to create a new variable with random values. Then, they create a new model where all of the latent variables point to a construct with this new "random" variable (a single-indicator LV). Next, you can run the PLS procedure, and check the "inner model" VIFs. These VIFs are the ones relevant to assess the existence of Common Method Bias
Thank you so much for these videos. Is it a good idea to run a Confirmatory Tetrad Analysis to be sure which constructs are reflective and which formative?
I prefer to just rely on the wording of the measures to imply which type it is. However, if your factor fails to load well as reflective and you want to see if it is possibly formative, you might run an CTA. I have only ever used CTA in demonstration, and never in a published manuscript (as it was never needed to confirm the type of factor measurement).
For the full collinearity approach, we are concerned with method bias implied by collinearity between factors, rather than anything in the measurement model.
How can I provide solution for reflective variables. For example, in my VIF, I had NE8. Natural Setting -> WRI11.Revist 10.498 NOV7.Novelty -> WRI11.Revist 10.600 in here, is it okay to remove NE8. Natural Setting and NOV7.Novelty.
If it is for the inner model, then one of the predictors is redundant with the other. You can decide which to keep. If it is for the outer model, then it is probably fine to just keep all the indicators since reflective indicators are supposed to be redundant.
These should help: CMB: ruclips.net/video/cjLldBXIIX4/видео.htmlsi=HKL5RN3LcUm9dgss Invariance: ruclips.net/video/L2LdAgMKmBo/видео.htmlsi=vXC3ka-HnkTEb6pm
After learning this video, I found all the values in the "inner model" is lower than 3.3. However, in the outer model, some items are higher than 3.3. In this case, is it a big issue? May I have your suggestion about how to deal with it, please?
@@GaskinationI am writing my masters thesis and this is the case I faced. But to be accepted the committee probably will ask for a published reference. I have cited your video here but not sure if this ok. Also searched for a reference but couldn't reach any. So, any hints would be appreciated.
I ADDED A RANDOM VARIABLE IN MY DATA BUT WHEN I IMPORT INTO SMART PLS AND BUILDING A MODEL THERE RANDOM VARIABLE IS NOT SHOWN. HOW TO SORT THIS PROBLEM?
The only reason it might not show is if you unchecked the box in the data preview in SmartPLS. The only other reason that I can think of is if the number of decimals is truly enormous (like more than 32 places).
I seems that the way to test CMB in SmartPLS 4 is different from that in SmartPLS 3 (ruclips.net/video/J7eeu4O80_M/видео.html). Could you please explain why? Thank you.
It seems the test suggested by Kock (2015) relies in a less straightforward procedure. Actually, the concept of a "full collinearity test" comes from the paper of Kock and Lynn (2012), where in their Appendix F suggest to create a new variable with random values. Then, they create a new model where all of the latent variables point to a construct with this new "random" variable (a single-indicator LV). Next, you can run the PLS procedure, and check the "inner model" VIFs. These VIFs are the ones relevant to assess the existence of Common Method Bias
Good catch! I'll make a follow up video to show how to do it with a random variable. Thanks!
Here is an updated video. You make a cameo around 11 seconds in: ruclips.net/video/cjLldBXIIX4/видео.html Thanks again!
Could you tell me please, if we have reflective formative measurement model and we consider only VIF to check CMB is it considerably good.
The assessment of CMB is still debated. VIFs is currently an acceptable method for mixed models like this.
@@Gaskination thanks for sharing knowledge with generosity
Thank you so much for these videos. Is it a good idea to run a Confirmatory Tetrad Analysis to be sure which constructs are reflective and which formative?
I prefer to just rely on the wording of the measures to imply which type it is. However, if your factor fails to load well as reflective and you want to see if it is possibly formative, you might run an CTA. I have only ever used CTA in demonstration, and never in a published manuscript (as it was never needed to confirm the type of factor measurement).
@@Gaskination Noted with thanks
I want to ask that why we should look at the inner model rather than the outer one?
For the full collinearity approach, we are concerned with method bias implied by collinearity between factors, rather than anything in the measurement model.
How can I provide solution for reflective variables. For example, in my VIF, I had
NE8. Natural Setting -> WRI11.Revist 10.498
NOV7.Novelty -> WRI11.Revist 10.600 in here, is it okay to remove NE8. Natural Setting and NOV7.Novelty.
If it is for the inner model, then one of the predictors is redundant with the other. You can decide which to keep. If it is for the outer model, then it is probably fine to just keep all the indicators since reflective indicators are supposed to be redundant.
Professor how to do CMB TESTING FOR FIRST ORDER FORMATIVE CONSTRUCTS. ALSO TELL HOW TO DO MEASUREMENT INVARIANCE BEFOR MGA
These should help:
CMB: ruclips.net/video/cjLldBXIIX4/видео.htmlsi=HKL5RN3LcUm9dgss
Invariance: ruclips.net/video/L2LdAgMKmBo/видео.htmlsi=vXC3ka-HnkTEb6pm
After learning this video, I found all the values in the "inner model" is lower than 3.3. However, in the outer model, some items are higher than 3.3. In this case, is it a big issue? May I have your suggestion about how to deal with it, please?
The outer model is expected to have higher VIF values because the factors are probably reflective.
@@Gaskination Thank you so much, Professor. Before your explanation, I worried about my data so much.
Is there any reference to this? I mean accepting higher VIF outer loading for reflective items @@Gaskination
@@Esalni We would expect reflective indicators to have high redundancy, and therefore higher VIF.
@@GaskinationI am writing my masters thesis and this is the case I faced. But to be accepted the committee probably will ask for a published reference. I have cited your video here but not sure if this ok. Also searched for a reference but couldn't reach any. So, any hints would be appreciated.
I ADDED A RANDOM VARIABLE IN MY DATA BUT WHEN I IMPORT INTO SMART PLS AND BUILDING A MODEL THERE RANDOM VARIABLE IS NOT SHOWN. HOW TO SORT THIS PROBLEM?
The only reason it might not show is if you unchecked the box in the data preview in SmartPLS. The only other reason that I can think of is if the number of decimals is truly enormous (like more than 32 places).
I seems that the way to test CMB in SmartPLS 4 is different from that in SmartPLS 3 (ruclips.net/video/J7eeu4O80_M/видео.html). Could you please explain why? Thank you.
Even this approach is not very good. Currently there is not a good approach as far as I know. The best method then is prevention.