FIMIX comparison of segment results

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
JoyFielding
PLS Junior User
Posts: 7
Joined: Thu Jun 08, 2017 9:14 am
Real name and title: Mara Holte

FIMIX comparison of segment results

Post by JoyFielding »

Dear all,

I am currently running a FIMIX analysis in order to test for unobserved heterogeneity in my model. I've done a lot of readings around that topic on how to apply the FIMIX method (Ringle/Hair/Sarstedt etc). However, I am not quite sure on how to interpret the final results with respect to the segments. What I did so far:

1) Determined the preferred number of FIMIX segments (=2 in my case)
2) Run the FIMIX analysis for 2 segments
3) Split the sample into two FIMIX groups according to the segment results of the FIMIX analysis
4) Estimated the structural model for the 2 FIMIX segments
5) Identified an explanatory variable that matches the FIMIX segmentation (overlap of >80%) and split the sample according to this variable into 2 groups
6) Estimated the structural model for the 2 segments originating from the explanatory variable
7) Compared the path coefficients and R^2 for the FIMIX segments with the corresponding ones from the explanatory variable

And here I run into problems. Do I understand it right that in order to show that unobserved heterogeneity is not a problem, I need to show that the model estimations for the FIMIX segments and the segments from the explanatory variable are about the same? I.e., value and direction of the path coefficients, significances and the R^2 should be about the same? In my understanding, this would mean that the explanatory variable is able to reproduce the results from the FIMIX segmentation and hence, there is no indication for unobserved heterogeneity. Is that right?

If my understanding is right, I am a bit confused about the case studies I've read on this topic, as there are hardly any cases were the FIMIX results match the results from the explanatory variable (e.g., Ringle et al. "Response-Based Segmentation Using Finite Mixture Partial Least Squares"). But maybe I am missing something here? Any help would be highly appreciated.

Many thanks,
Mara
User avatar
cringle
SmartPLS Developer
Posts: 818
Joined: Tue Sep 20, 2005 9:13 am
Real name and title: Prof. Dr. Christian M. Ringle
Location: Hamburg (Germany)
Contact:

Re: FIMIX comparison of segment results

Post by cringle »

No, the FIMIX-PLS results, which are based on the probabilities of membership (perfect segmentation), never match the (dichotomous) segment assignment when using an explanatory variable. FIMIX-PLS shows you that there is something; the forming of explanatory groups allows you to explain it as good as possible.

The next step should be to form groups of data based on you explanatory variable (there is an option in the SmartPLS data view to create data groups) and to run a multigroup analysis (e.g., by applying the permutation test).

Best regards
Christian Ringle
JoyFielding
PLS Junior User
Posts: 7
Joined: Thu Jun 08, 2017 9:14 am
Real name and title: Mara Holte

Re: FIMIX comparison of segment results

Post by JoyFielding »

Dear Prof. Ringle,

thank you for your explanation. If I understand it right, the aim is to find a segmentation based on an explanatory variable that best approximates the results of the FIMIX segmentation. Is that correct? Running the MGA with the groups of the chosen explanatory variable would then show whether there are any significant differences between the two groups.

If the above is correct, is there any criteria to decide whether unobserved heterogeneity is an issue in the analyzed data? Or is this a mere subjective evaluation based on the FIMIX analysis?

Many thanks
Mara
User avatar
cringle
SmartPLS Developer
Posts: 818
Joined: Tue Sep 20, 2005 9:13 am
Real name and title: Prof. Dr. Christian M. Ringle
Location: Hamburg (Germany)
Contact:

Re: FIMIX comparison of segment results

Post by cringle »

Correct and yes, it's an evaluation based on the FIMIX-PLS results (information criteria + EN + relative segment sizes). Here is an example:

Hair, J.F., Sarstedt, M., Matthews, L., Ringle, C.M., 2016. Identifying and Treating Unobserved Heterogeneity with FIMIX-PLS: Part I - Method. European Business Review 28, 63-76. http://www.emeraldinsight.com/doi/full/ ... -2015-0094

Matthews, L., Sarstedt, M., Hair, J.F., Ringle, C.M., 2016. Identifying and Treating Unobserved Heterogeneity with FIMIX-PLS: Part II – A Case Study. European Business Review 28, 208-224. http://www.emeraldinsight.com/doi/abs/1 ... -2015-0095

Best regards
CR
JoyFielding
PLS Junior User
Posts: 7
Joined: Thu Jun 08, 2017 9:14 am
Real name and title: Mara Holte

Re: FIMIX comparison of segment results

Post by JoyFielding »

Thank you Prof. Ringle.

As I understand it, the mentioned criteria (information criteria / EN / rel. segment size) serve as indicators to define the appropriate number of segments in the FIMIX analysis. So, if these criteria point towards a one-segment-solution, I would assume that there is no heterogeneity in the data. But how do I decide whether there is an issue with unobserved heterogeneity in case of more than 1 segment? Do I need to compare the weighted R^2 for the FIMIX model estimation with the model estimation of the explanatory variable like it is done in the case study you mentioned?

Many thanks
Mara
jmbecker
SmartPLS Developer
Posts: 1282
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: FIMIX comparison of segment results

Post by jmbecker »

1) If the information criteria etc. point to more than 1 segment as preferable solution, you have a problem with unobserved heterogeneity.
2) If you are able to explain the segment solution with an external variable (as you do), you turn unobserved into observed heterogeneity.

Of course, as Prof. Ringle describes, this process in never perfect and some of the unobserved heterogeneity might not be captured by the observed variable. However, the advantage of the observed variable (heterogeneity) is that it can be tested, for example, with a MGA or a moderation analysis.

A good reference for the process of uncovering unobserved heterogeneity and turning it into observed heterogeneity is also explained in:
Becker, J.-M., Rai, A., Ringle, C. M., and Völckner, F. (2013). "Discovering Unobserved Heterogeneity in Structural Equation Models to Avert Validity Threats." MIS Quarterly, 37(3), 665-694.
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
JoyFielding
PLS Junior User
Posts: 7
Joined: Thu Jun 08, 2017 9:14 am
Real name and title: Mara Holte

Re: FIMIX comparison of segment results

Post by JoyFielding »

Thank you Dr. Becker and Prof. Ringle for your comments. This helped a lot!
Post Reply