SPSS Factor loadings and SmartPLS factor loadings

Frequently asked questions about PLS path modeling.
Post Reply
abangert
PLS Junior User
Posts: 8
Joined: Wed Oct 26, 2016 3:02 pm
Real name and title: Arthur Bangert Associate Professor

SPSS Factor loadings and SmartPLS factor loadings

Post by abangert » Fri Oct 06, 2017 8:09 pm

Hello,

We are conducting research where we have combined items from several instruments and conducted factor analysis in SPSS. We then used the items connected to their respective factors as they loaded in SPSS. However, when we run the SmartPLS , the factor loadings (outer loadings) are different from the SPSS factor loadings. What might be the potential reasons for this. The items did not load as anticipated but we maintained the same factor structure for the constructs in SmartPLS.
Any ideas about the differences would be appreciated.

Thanks,

Art

User avatar
cringle
SmartPLS Developer
Posts: 806
Joined: Tue Sep 20, 2005 9:13 am
Real name and title: Prof. Dr. Christian M. Ringle
Location: Hamburg (Germany)
Contact:

Re: SPSS Factor loadings and SmartPLS factor loadings

Post by cringle » Sat Oct 07, 2017 8:08 am

You find a couple of answers to this question in this discussion forum (e.g., viewtopic.php?f=12&t=2666&p=8088&hilit= ... ings#p8088)

The results differ, that's the expected outcome.

SPSS does not conduct a PLS path modeling analysis. Hence the factors and scores differ. PLS determines the scores from the model estimation of composites. The factor-based estimation in SPSS only focuses on the construct.

These articles give you some additional insights:

Where the bias lies: http://www.sciencedirect.com/science/ar ... 6316304404
Five perspectives: https://rsw.beck.de/docs/librariesprovi ... f?sfvrsn=0

Also, here you find an intro on PLS-SEM:
https://www.researchgate.net/publicatio ... n_Modeling

What you can try it to create a construct Y1 with indicators x1 to x2. Then your create a second construct Y2 with the same indicators x1 and x2. Contect both construct with a path (the direction does not matter) and run the PLS algorithm in SmartPLS. Here, you get outer loadings that are identical to principal component anaylsis resutls in SPSS. Why? The PLS path model does not use an additional conext (i.e. other constructs) and, thus, the loadings are created only by the context of the one construct that your are looking at. Otherwise, the loadings of factor models and PLS differ.

Best
Christian

abangert
PLS Junior User
Posts: 8
Joined: Wed Oct 26, 2016 3:02 pm
Real name and title: Arthur Bangert Associate Professor

Re: SPSS Factor loadings and SmartPLS factor loadings

Post by abangert » Sat Oct 07, 2017 8:57 pm

Hello Christian,

Thank you very much for your explanation. I very much appreciate the resources that you sent and I definitely review them.

Thanks again,

Art

plspetersons
PLS Junior User
Posts: 1
Joined: Thu May 02, 2019 11:12 pm
Real name and title: Sara Peterson

Re: SPSS Factor loadings and SmartPLS factor loadings

Post by plspetersons » Thu May 02, 2019 11:48 pm

Dear Christian,

Can you explain why there is a slight difference between the Construct Reliability and Validity Matrix when we run SmartPLS algorithm for path and then factor analysis. Can you quantify this difference? I see that the loadings change slightly as well. Is this the cause of this difference?

Which do I use when reporting construct validity and reliability?

Thanks for your help.

Post Reply