multigroup: median--> what to do with the borderline case

Frequently asked questions about PLS path modeling.
Post Reply
PLS User
Posts: 17
Joined: Wed Apr 24, 2013 7:38 am
Real name and title:

multigroup: median--> what to do with the borderline case

Post by Julen » Wed May 28, 2014 3:09 pm

Dear all,

I want to conduct a MULTIGROUP analysis, but I have not a categorical variable. Instead, I have 4 items (Likert 1-5) which define a variable called "moment of entry (ME)". This variable would allow me divide my sample (200) into 2 groups: pioneers (P) and followers (F).

The mean value of the 4 items for the hole sample is 3,16875. I have read (Esposito et al., 2010, "Handbook of PLS", chapter 30: Hensler and Fassott) that the MEDIAN SPLIT can be done, without loosing any respondants. The problem is that the median for those ítems is 3,25 and 20 respondants have just that value (3,25).

I Have decided to take those 20 respondants out of the analysis, and I still have 90 P and 90 F. Is it right? any other ideas?

Thank you very much in advanced.

User avatar
PLS Super-Expert
Posts: 1618
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia

Post by Hengkov » Fri May 30, 2014 12:34 am


Just have two conditions to run PLS-MGA:
1. Non-metric variables play role moderator in your cases. This is mean your have non-linear model and know how many groups.
2. Unobserved heterogeneity with metric latent variables. This is mean your not know how many classes.

In your cases, your mention have two groups. But, I confuse where you know two groups?. I think option two possible for you, because you says not have categorical variables in models. Now I will give my opinions about Median Split your mention according to Henseler and Fassott (2010). This approach for handle option two, but I think this is bad method. Some resons I will give for you:
1. This approach just results two groups
2. This approach will results extrime different group "high and low" with extrime different sample size (your mention 20 sample above mean, so 20:180)
3. The result thi approach may be produce missleading conclutions.
4. Not guide or detail explanation for interpretation this result approach.

So, I recommend you to run REBUS-PLS first, before run PLS-MGA.

Best Regards,

Best Regards,

Post Reply