Hi Mr. Gaskin, I checked for Common Method Bias like you described. Turns out that one connection to the random variable has a VIF-value > 5. The variable building the connection to the random variable in the CMB-test is the one that has no outgoing connections in my model. Can you tell me how to handle this?
If it is the dependent variable, then this might be reasonable since we are trying to predict it with the other variables anyway (i.e., we expect it to share substantial variance with the other variables).
Control variables can be added just like any other predictor. If they are categorical (such as religion, marital status, etc.) then they should be separated into binary dummy variables (with a reference group omitted). If they are ordinal or continuous, then they can be included as latent factor (if they belong to latent factors) or as single item "factors".
Thank you so much for your video. However, I am wondering if some indicators have VIF value which are > 3. How can I handle it (reflective model)? And what is the proper range (good range) for VIF value ?
Greetings, so one of my variables is at second order while other i am taking dimension wise. Do i align all variables as they are in final model with the random dependent variable?
Pardon me Dr. Gaskin. May I know which common method bias I should follow? Is it this video or the previous video that you have demonstrated? Hope to hear from you soon. Thank you 😄
@@JyotiKaur-qj6hl It should be metric. Make sure it is checked in the data view. If it won't allow this, check the excel file to make sure the random number is just a number, and not a formula. Also perhaps limit the number of decimals to five.
VIF with random outcome is a way to show the shared variance explained in an unrelated outcome, rather than in a potentially related outcome. So, this just avoids attributing shared method variance to what might actually be shared trait variance. i.e., the extent to which all variables share variance with a random outcome is not likely due to any true trait relationships.
Hello James, If I have a model with second-order constructs (reflective- formative and formative- formative), how should I use this method? Is there any video that explains this?
Method bias is primarily a problem with reflective scales. So, I don't think I would worry too much about method bias tests for the formative model. forum.smartpls.com/viewtopic.php?t=15341
The papers are: Kock, N., & Lynn, G. (2012). Lateral Collinearity and Misleading Results in Variance-Based SEM: An Illustration and Recommendations. Journal of the Association for Information Systems, 13(7), 546-580. doi.org/10.17705/1jais.00302 Kock, N. (2015). Common method bias in PLS-SEM: A full collinearity assessment approach. International Journal of E-Collaboration, 11(4), 1-10. doi.org/10.4018/ijec.2015100101
From my understanding, VIF values are calculated from independent variables only; thus, any variable can be a dependent variable. You don't have to manually create a random variable for a dependent variable.
Hi James, I am very grateful for your videos！It helps me a lot. The prior procedure in the before video (link: ruclips.net/video/pp-2dKCFrWo/видео.html) is not correct, right? And, the procedure in this video that all variabls are regressed against a new variable with a item valud by a random number is correct. Or, both of ways you showed are right？
Hi Mr. Gaskin, I checked for Common Method Bias like you described. Turns out that one connection to the random variable has a VIF-value > 5. The variable building the connection to the random variable in the CMB-test is the one that has no outgoing connections in my model. Can you tell me how to handle this?
If it is the dependent variable, then this might be reasonable since we are trying to predict it with the other variables anyway (i.e., we expect it to share substantial variance with the other variables).
hi, thank you for your tutorial. I have a question, how to add my control variable to my smartpls model? thank you
Control variables can be added just like any other predictor. If they are categorical (such as religion, marital status, etc.) then they should be separated into binary dummy variables (with a reference group omitted). If they are ordinal or continuous, then they can be included as latent factor (if they belong to latent factors) or as single item "factors".
Thank you so much!
Thank you so much for your video. However, I am wondering if some indicators have VIF value which are > 3. How can I handle it (reflective model)? And what is the proper range (good range) for VIF value ?
For reflective indicators, there is supposed to be some multicollinearity, as they are intentionally redundant. So, we only check the inner model.
@@Gaskination Thank you for your response. I am sorry for late reply
Greetings, so one of my variables is at second order while other i am taking dimension wise. Do i align all variables as they are in final model with the random dependent variable?
You can do it at either level, though it probably makes more sense to do it at the first-order level.
Shall i do this process just once for all the data set, or i do it for each group in my data set , which i will further analyse using mga??
Just with all data.
Hii Prof. James, how to access the data set available for practices purpose and run it as shown by you
It is the Sohana dataset available on the homepage of statwiki.gaskination.com/
Professor, Is the procedure same for a model with 2 formative constructs and 1 reflective construct?
Yes, it should be. Method bias is an item-level phenomenon.
@@Gaskination Thanks Prof
Pardon me Dr. Gaskin. May I know which common method bias I should follow? Is it this video or the previous video that you have demonstrated? Hope to hear from you soon. Thank you 😄
If using SmartPLS, then this is the way.
@@Gaskination thank you Dr. Gaskin 😄
Can you please tell how the random variable was calculated.. Was that variable a totality of the earlier.. Or that formula onky =RANDOM()
I just did it in Excel prior to importing the data into SmartPLS.
As always perfect!
I followed the same steps, and receiving the classic error that "this variable is part of an incoherent graph"....any idea what went wrong?
Make sure all latent variables are blue (connected to other latent variables), and try to avoid circular paths.
Sir, i am selecting the correct data file but random variable is not visible left habd side while analysis. What to do now??
In excel import when I am looking, it's showing random as none in scale option and not having min and maximum values.....
@@JyotiKaur-qj6hl It should be metric. Make sure it is checked in the data view. If it won't allow this, check the excel file to make sure the random number is just a number, and not a formula. Also perhaps limit the number of decimals to five.
Okay thank you sir
Sir,,what is the difference between this VIF value and the VIF value from the previous video..which is more straight forward?
VIF with random outcome is a way to show the shared variance explained in an unrelated outcome, rather than in a potentially related outcome. So, this just avoids attributing shared method variance to what might actually be shared trait variance. i.e., the extent to which all variables share variance with a random outcome is not likely due to any true trait relationships.
@@Gaskination Thank you so much sir. BTW this video is excellent, I manage to get the results
by following your explanation. thanks again!
Hello James, If I have a model with second-order constructs (reflective- formative and formative- formative), how should I use this method? Is there any video that explains this?
Method bias is primarily a problem with reflective scales. So, I don't think I would worry too much about method bias tests for the formative model. forum.smartpls.com/viewtopic.php?t=15341
@@Gaskination Thanks James
Would you be so kind to add the link to the paper referring to this approach?
The papers are:
Kock, N., & Lynn, G. (2012). Lateral Collinearity and Misleading Results in Variance-Based SEM: An Illustration and Recommendations. Journal of the Association for Information Systems, 13(7), 546-580. doi.org/10.17705/1jais.00302
Kock, N. (2015). Common method bias in PLS-SEM: A full collinearity assessment approach. International Journal of E-Collaboration, 11(4), 1-10. doi.org/10.4018/ijec.2015100101
From my understanding, VIF values are calculated from independent variables only; thus, any variable can be a dependent variable. You don't have to manually create a random variable for a dependent variable.
The random variable is recommended so that you can test all other variables together.
thanks
Hi James, I am very grateful for your videos！It helps me a lot. The prior procedure in the before video (link: ruclips.net/video/pp-2dKCFrWo/видео.html) is not correct, right? And, the procedure in this video that all variabls are regressed against a new variable with a item valud by a random number is correct. Or, both of ways you showed are right？
yes. The random dependent variable is considered more valid for this test.
@@Gaskination Thanks a lot！！Having a good day！