FIMIX result VS ex post FIMIX result (dilemma)

[oemar99]

Dear PLS users,

I've applied FIMIX analysis on my sample and appears that K=2 to be the best solution. Now i'm facing problem with the best approach to identify the variables that best characterized the two-segment as recommended by FIMIX result. I followed step-by-step suggested by Ringle et al. (2010) "Response-based Segmentation...." and Rigdon et al. (2010) "Structural Modeling of Heterogenous Data...." to identify the explanatory variable for segmentation.

Based on demographic variables using CHAID and QUEST algorithm (SPSS answer tree v.20), 'user's internet experience' happened to be the suitable one among others. However, when I compared between the FIMIX-based segmentation and Answertree-based segmentation, the results are not similar, only few path coefficients are consistent (about 60%). My questions:

1. How to conclude if the FIMIX result vs ex post FIMIX result do not tune each other? I believe the FIMIX vs ex post FIMIX result from both above articles are just nice, but what if they are different?
Do we need to choose other variables (if yes what is the best approach besides Answertree)? or Do we need to choose different segmentation from the FIMIX result?

2. Still related to no.1, if this is the case, I'm wondering to start with Answertree first and match with FIMIX result later, can it be done instead?

3. Knowing FIMIX only assessed the inner model, can we simply present the result for outer model after we run the ex post FIMIX analysis?

Thank you in advance for the answers. Im anxiously waiting for the reply.

regards,
Umar

[cringle]

Hi

No worries!

1. How to conclude if the FIMIX result vs ex post FIMIX result do not tune each other? I believe the FIMIX vs ex post FIMIX result from both above articles are just nice, but what if they are different?
Do we need to choose other variables (if yes what is the best approach besides Answertree)? or Do we need to choose different segmentation from the FIMIX result?

FIMIX-PLS gives you the best possible segmentation based on observations' probabilities of membership. You can never get an exact match when you segment data ex post based on a good expalnatory variable. It's good enough if the key characterisitcs match. Take a look at some of my FIMIX-PLS applications:
http://www.tuhh.de/hrmo/team/prof-dr-c-m-ringle.html

2. Still related to no.1, if this is the case, I'm wondering to start with Answertree first and match with FIMIX result later, can it be done instead?

I never did this. But if you know a good explanatory variable, you may want to use the PATHMOX approach: http://www.stat.unipg.it/iasc/Proceedin ... 0Aluja.pdf

3. Knowing FIMIX only assessed the inner model, can we simply present the result for outer model after we run the ex post FIMIX analysis?

Sure, use the segmentation based on the explanatory variable and run a PLS multigroup analysis (http://www.emeraldinsight.com/doi/abs/1 ... 0000022012) for the entire path model.

Best regards,
Christian

[oemar99]

Thanks Prof. Ringle for the prompt reply, it's relieving indeed.

I was thinking to make a comparison between a multigroup analysis based on a priori information and ex post FIMIX analysis, see if the segment will differ.

In IS studies, I think setting up a priori segmentation suffer from a solid ground theory, thus in my opinion, FIMIX or other similar algorithm should be the appropriate one to detect unobserved heterogeneity and later can be termed as observed heterogeneity, am I right Prof.?

However, is the new version of SmartPLS available free for student? :)

regards,
Umar

[cringle]

Hi

In IS studies, I think setting up a priori segmentation suffer from a solid ground theory, thus in my opinion, FIMIX or other similar algorithm should be the appropriate one to detect unobserved heterogeneity and later can be termed as observed heterogeneity, am I right Prof.?

Perfectly right! You may want to checkt this MIS article on the need to uncover unobserved heterogeneity:
http://misq.org/discovering-unobserved- ... s1rqq80n85

Best
Christian

[cringle]

However, is the new version of SmartPLS available free for student? :)

Yes, at the moment, for a limited period of time. Just click on the Purchase-Button.

Best
Christian

[oemar99]

thanks a lot Prof.

rgds,
Umar