SmartPLS 3.0 and small sample size

Frequently asked questions about PLS path modeling.
Post Reply
spangler
PLS User
Posts: 14
Joined: Thu Feb 23, 2006 5:23 pm
Real name and title:

SmartPLS 3.0 and small sample size

Post by spangler » Fri Mar 27, 2015 8:57 pm

Hi all,

I am testing a model in SmartPLS 2.0 and SmartPLS 3.0 The model and data are the same; the program are on
different computers. In SmartPLS 3.0 when I create an
interaction term of two variables I get the message that the sample size is too small, due to having more indicators in the
interaction term than number of observations. In SmartPLS 2.0 this is not a problem, the model runs. What is the
difference between the two versions of SmartPLS?. One reason for using PLS is to explore complex models with small sample size. Also, when I reduce the
number of indicators of one of the variables to reduce the number of interaction indicators, the resulting
coefficient in SmartPLS 3.0 is different from that in SmartPLS 2 though the remaining coefficients and R2s in the model
are the same. (I reduced the number of indicators in the SmartPLS 2.0 model also.) Any help
would be greatly appreciated.

User avatar
cringle
SmartPLS Developer
Posts: 807
Joined: Tue Sep 20, 2005 9:13 am
Real name and title: Prof. Dr. Christian M. Ringle
Location: Hamburg (Germany)
Contact:

Re: SmartPLS 3.0 and small sample size

Post by cringle » Sat Mar 28, 2015 8:08 am

There is no need to take SmartPLS 2 as a reference (it's beta). We fixed some issues in the interaction term computation in SmartPLS 3.2. That's the reason why the results may differ:
https://www.smartpls.com/release_notes

However, basic PLS-SEM results are certainly identical in SmartPLS 2 and SmartPLS 3 since we did not change the basic PLS algorithm.
https://www.smartpls.com/faq/why-do-sma ... lts-differ

So, the difference results from the interaction term computation that we imporved in SmartPLS 3.

What smaple size do you use?

Kind regards,
Christian RIngle

spangler
PLS User
Posts: 14
Joined: Thu Feb 23, 2006 5:23 pm
Real name and title:

Re: SmartPLS 3.0 and small sample size

Post by spangler » Sun Mar 29, 2015 1:31 pm

Hello Christian,

Thanks for the quick response. N = 97. There are four predictor LVs in the model, one with
7 indicators, the others have about 30 - 40 indicators each. We need to create interaction terms using the
seven-item LV and each of the others.

User avatar
cringle
SmartPLS Developer
Posts: 807
Joined: Tue Sep 20, 2005 9:13 am
Real name and title: Prof. Dr. Christian M. Ringle
Location: Hamburg (Germany)
Contact:

Re: SmartPLS 3.0 and small sample size

Post by cringle » Sun Mar 29, 2015 2:14 pm

The product indicator approach does not work since you have 7*30 products. This number is larger than your sample size and cannot be estimated. Try the two-stage approach.

spangler
PLS User
Posts: 14
Joined: Thu Feb 23, 2006 5:23 pm
Real name and title:

Re: SmartPLS 3.0 and small sample size

Post by spangler » Mon Mar 30, 2015 11:30 am

Thanks. I experimented some. It seems that eliminating a few indicators from one or the other of the main effects to allow the product method
yields a larger interaction term than using the two-stage approach with all of the initial indicators. Is this correct in general?

User avatar
cringle
SmartPLS Developer
Posts: 807
Joined: Tue Sep 20, 2005 9:13 am
Real name and title: Prof. Dr. Christian M. Ringle
Location: Hamburg (Germany)
Contact:

Re: SmartPLS 3.0 and small sample size

Post by cringle » Mon Mar 30, 2015 12:50 pm

It often looks like this, but the result may not be correct ;) In line with simulation studies and our own experiments, we suggest using the two-stage standatdized options in SmartPLS 3.

Best regards,
CR

spangler
PLS User
Posts: 14
Joined: Thu Feb 23, 2006 5:23 pm
Real name and title:

Re: SmartPLS 3.0 and small sample size

Post by spangler » Mon Mar 30, 2015 1:24 pm

ok, thanks for the recommendation.

Post Reply