Prediction-Oriented Segmentation in PLS-SEM (PLS-POS)

[janschreier]

I would say so :)

[cringle]

Yes, your understanding is correct. The group assignment shows which observations belongs to a certain group. More details on PLS-POS:
Becker, Jan-Michael, Arun Rai, Christian M. Ringle, and Franziska Völckner. 2013. "Discovering Unobserved Heterogeneity in Structural Equation Models to Avert Validity Threats." MIS Quarterly 37 (3): 665-694.
http://www.researchgate.net/publication ... ty_Threats

Best
CR

[spiking]

Screen Shot 2015-09-30 at 8.17.10 am.png (111.95 KiB) Viewed 5844 times

I am getting this error message frequently when I run PLS-POS on SmartPLS 3.

I tested various combinations of the settings based on a sample size of 1230:

FAIL
1. Groups = 5
2. Maximum iterations = 2000
3. Search depth = 300

FAIL
1. Groups = 5
2. Maximum iterations = 600
3. Search depth = 300

FAIL
1. Groups = 5
2. Maximum iterations = 500
3. Search depth = 500

FAIL
1. Groups = 5
2. Maximum iterations = 500
3. Search depth = 200

FAIL
1. Groups = 5
2. Maximum iterations = 400
3. Search depth = 100

PASS
1. Groups = 5
2. Maximum iterations = 400
3. Search depth = 400

PASS
1. Groups = 5
2. Maximum iterations = 300
3. Search depth = 300

PASS
1. Groups = 5
2. Maximum iterations = 300
3. Search depth = 200

PASS
1. Groups = 5
2. Maximum iterations = 300
3. Search depth = 100

PASS
1. Groups = 5
2. Maximum iterations = 200
3. Search depth = 200

PASS
1. Groups = 5
2. Maximum iterations = 200
3. Search depth = 100

PASS
1. Groups = 5
2. Maximum iterations = 100
3. Search depth = 100

As you can see below, my iMac specifications have 8GB of memory which is sufficient to run the program.

Why does it fail when I run higher maximum iterations and search depths?

Thank you.

Screen Shot 2015-09-30 at 11.19.06 am.png (48.2 KiB) Viewed 5841 times

Screen Shot 2015-09-30 at 11.19.15 am.png (26.15 KiB) Viewed 5841 times

[cringle]

The error message occurs if in certain segment, all obersvations of a variable (e.g., x1) have the same values (e.g. 4). Then, no variance exists and the PLS algorithm cannot run.

This hapens when your responses are very similar and when there is a very small group of data.

Similar data can occur when you have many missing values and use mean value replacement.

You may want to check your data, use smaller number of segments and use case wise deletion instead of mean value replacement.

Best
CR

[spiking]

Dear Professor Christian

Thank you for your feedback.

For my data, they do not contain missing values. As such, casewise deletion or mean value replacement does not apply.

I took heed of your advice and re-run the PLS-POS based on 4 segments:

FAIL
1. Groups = 4
2. Maximum iterations = 800
3. Search depth = 800

FAIL
1. Groups = 4
2. Maximum iterations = 600
3. Search depth = 600

PASS
1. Groups = 4
2. Maximum iterations = 500
3. Search depth = 500

The above trial and error process is to determine the optimal number of segments to work with based on getting the highest possible combination of maximum iterations and search depth.

Will update again if there are interesting findings.

Thank you!

clemen

[cringle]

Thanks!

You can also use FIMIX-PLS to determine the number of segments. Then, run PLS-POS only for that numberof segments.

Best
CR