PLS-MGA

[hxin009]

hi everyone,

I got control variables such as GENDER, ETHNICITY in my research model. I want to test the effects of control variables on the Dependent variable. So i do a multi-group analysis, right?

What i did for the multi-group analysis is:
1. split the data set into 2 sub-datasets, one is Female (n=151)and the other is Male(n=150).
2. i import the data sets into the project and run PLS-algorithm.

Here comes the problems. An error message keeps popping up when i tried to run the algorithm. it is a Calculation error and saying that" a singular matrix occurred during the estimation of the path coeffients using the path weighting scheme. setting another weighting scheme could solve the problem."

So to solve this problem, i tried the other two weighting schemes, but unfortunately, another error message popped up saying that " an error occurred during the outside estimation. Maybe there are too few observations".

I don't know what to do now. There are 8 constructs in my research model and female group(n=151) and male group (n=150). I think theoretically it is sufficient for the analysis, isn't it? Can anyone help me out? Thanks in advance!

One more thing is, for the ethnicity groups, i got three groups and they are group 1(n=120), group 2(n=90), group 3(n=70). Are the sample size for those groups sufficient for the analysis?

Please help me! Thank you！

Best regards,
Catherine

[Hengkov]

Hi Catherine,

No, if gender just control variable no test multigroup (but connect to dependent variable).

Yes, if gender is moderator used PLS-MGA.

Best Regards,
Hengky

[hxin009]

Hi Hengky,

Thanks for the help and really appreciated.

one more question about creating dummy variables. i am not sure if it is appropriate to ask this question here, but i am really grateful if you could clear for me.

Please check if i am on the right track:

Say i got four control variables (gender, ethnicity, smoker, years of experiences). I understand for categorical variable( gender, ethnicity and smoker) we should convert to dummy variables, right?

For gender, i used 0=Female, 1=male.
For smoker, i used 0=yes , 1=no.
For ethnicity, i am not sure if i done this right. I created this dummy variable in Excel spreadsheet and used this =if(logical test..) this function and create three variables Asian ( Asian=1, Others=0), European(European=1, others=0), Pacific people (Pacific People=1, others=0).

After i done this, i create a Formative latent variable in the smartPLS called "Ethnicity" (with the dummy variables as three indicators: Asian, European and Pacifici people) and link Ethnicity directly to the Dependent Variable.

Another thing is, for the years of experiences (control variable with values for example, 1.5, 1, 0.5, 10, 5 ) so i don't need to do anything about this variable, right? i just direct link this control variable to the dependent variable.

Thank you for your guidance again!

best regards,
Catherine

[Hengkov]

Hello Catherine,

Yes, right.
Control variable just to control dependent variable with assume accurate estimates. You must have the reason, why control variable is important to include? How to relation control variable and another variable? Basic theory? etc.

For statistical analysis, you cannot test multigroup because just control, no moderator variable.
I suggest you read one article below for clear explain about control variables:
Spector, P.E., and Brannick, M.T. 2011. "Methodological Urban Legends: The Missuse of Statistical Control Variables," Organizational Research Methods (14:2), pp.287-305.

Best Regards,
Hengky

[hxin009]

Hi Hengky, Thank you so much for your guidance!

Best regards,
Catherine

[Choukri]

Hi everyone,
Could I use PLS-MGA with a model contains two variables, independent and dependent, and three moderating variables (three personality traits) to compare between two groups; Female and Male.

I ask this question because after I run PLS-MGA function, the table that shows the results of PLS-MGA does not includ the interaction effects, just includs direct effects

Thank you

[jmbecker]

If you have interaction effects (moderating effects) in your model, they should be included in the MGA results.
But if I understand you correct you mean your grouping variables which serve as moderating variables in your theory (i.e., gender)? These of course are not included as effects in the results, but you should find look at the differences between direct effects between the groups (males vs females).

[Choukri]

Hi,

Thank you for the respond

Look to this phrase "Past research by Chin et al. (1996, 2003) would advise against the use of multiple group models when researchers have continuous moderator variables at their disposal as it could result in inadequate power to detect the moderator/interaction effect".

Regarding to the above, coulkd I make a multi-group analysis by gender (Female vs. Male) using a model contains 3 variables (1 independent and 2 dependent) and three continuous moderator variables?

Thank you

[jmbecker]

Yes, gender is not a continuous variable, but a dichotomous. For such a variable a multi-group analysis is meaningful if you expect all relations to change, because of your grouping variable. If you expect only one relation to change, you can also use an interaction effect with a dummy coded 0/1 variable.