forum.smartpls.com

Hi everyone!

[*] I have a latent, formative construct ("technological complexity") with 4 indicators.
[*] I would like to compare two groups of data; i.e. "high" techn. complexity vs. "low" techn. complexity.
[*] I am thinking about a median split.

a) Which steps are needed to split the data?
b) Do I use the latent variable score or the indicator values to split the data?
c) Could you suggest any literature on that?

Kind regards,
Frank

The only source I found so far:
Testing moderating effects in PLS path models: An illustration of available procedures; J Henseler, G Fassott - Handbook of partial least squares, 2010 - Springer

Can anyone help?

Dear Frank,

I have the same question. I also have 4 ítems (Likert 1-5) which would allow me divide the sample(200) into 2:

I have done this: calculate the mean value of the 4 ítems for every respondant. Then calculate the median (with Excel). The problema is that the median is 3,25 and there are 20 respondants who have exactly this number.

I am thinking about taking them away. Doing that I would have 90 respondants in "high", 90 respondants in "low". Still good enough. But I am not sure if this could be done in a better way...

I also have the same source as you do (Henseler and Fassott), I haven´t found any other interesting one yet.

Hi,

If dichotomous variables is "natural" you can easy to run PLS-MGA. Example: Gender have two groups (male and female). But, if you not know how to split the data, your have problem unobserved heterogeneity. So, I recommend you to run REBUS-PLS for know how many sample in each groups. This is simillar with cluster analysis.

Best Regrads,

Dear all,

In my case it is not natural (like gender), and I want to split the data into two with my own criteria.

I guess I have to keep in mind that I need the minimun simple size in both sub-samples.

I will read chapter 9 of Gastón Sánchez free e-book (Frank, that might help you too), it talks about Rebus-PLS.

The problem is that it says Rebus is used for detecting classes...and I want to have only two classes (I want it to be determined by me)...

Thank you Mr. Hengkov.

Hi,

If you know two groups in your cases, why your confuse to run PLS-MGA? This mean your have non-metric natural variables, such male and female, manager and employee, firm with high complexity tech and low complexity tech, so your just split based on sample groups "high and low" (Note: this is simillar with discriminant analysis). But if you not know how many classes in your cases (maybe in your cases possible with very high, high, moderate, low or very low) you must run REBUS-PLS for detection number of class and each sample groups with result. After that, your can easy to run PLS-MGA (this is simillar with cluster analysis). I think this clear and I hope your understand it.

Best Regards,

Hi

Thank you. No, I don´t have non metric natural variables (like sex).

I have a variable which is called "moment of entry" (4 items), and I want to have only 2 groups if possible (pioneers and followers). That´s why I want to "force" the analysis to have only 2 groups, splitting the sample into 2.

May be it´s not possible to force it beforehand and I need to conduct what you say, a Rebus-Pls analysis. But I think I wouldn´t like to do it, since it might complicate the theorical basis of my work.

Thank you!

Why would you want to do this in the first instance.

you should keep it as a continuous variable and not lose the power in trying to detect the effect

see PLS Handbook chapter Wilson. on researchgate

Brad