application of IPMA - complicated but important issue

[widefov]

Hi guys,

I am conducting an research with a PLS model and the result is fine. But what I really want to do is to apply the model result to a specific company by using IPMA.

In sum,

1. I have tested an research model having 10 hypotheses. It is about a firm capability called X. 150 samples were collected from 50 companies in industry level. 4 hypotheses were rejected so I got a modified model with 6 significant paths (hypotheses). The result is okay to debate.

2. I have a 50 samples from a specific (single) company. What I want to do is to identify the weak points of capability X of this company. IPMA is a good tool to identify the opportunities to improve (weakness). Thus, I thought I need to compare the 2 IPA matrix from industry level (150 samples from 50 companies) and the firm level (the specific company with 50 samples).

3. In this stage I confront some problems:

To create IPA matrix of the company, which path model I should use? 1) original research framework having 10 hypotheses? or 2) modified model with 6 hypotheses?

In case of using 1) for the company, the problem is that one of the hypotheses i have rejected from the original research model is significant in this company's case, implying that the original model test is not valid to some extent. (interestingly if I combine the two data sets, the result is the same as the original model test result)

In case of using 2) for the company, the problem is that I need to compare the IPA matrix of the modified model having 6 hypotheses (the firm level) and the matrix of the original model having 10 hypotheses (industry level) which is a bit strange as 'comparing' needs to be conducted in the same conditions for me.

Thus, what I thought is that to use option 3.

First, test the original model (10 hypothesis) with 150 samples from industry and get the modified model (6 hypotheses).
Second, test and assess the modified model (6 hypothesis) again by using 150 samples like alternative model testing and create IPA matrix.
Third, use 50 samples (from the single company) with the modified model and create IPA matrix of the company. Now I can compare the two matrix in the same condition.

Do you think that option 3 is reasonable? In other words, I test the original research model with 150 samples and get modified model. And I test the modified model with 6 hypothesis by using the same 150 samples again. Of course all the paths are significant.

Actually, If I do not compare the IPA matrix between industry and the company, the same issue arises. Because creating an IPA matrix should be based on a proper path model if the matrix want to be robust.

I am looking forward to hear from you guys soon.

BW,
widefov

[jmbecker]

1)
Refining a model based on a sample is not a good idea in my opinion.
Heterogeneity could be a problem. On average over all companies in an industry it may be true that not all relations are significant (important), but for single companies (as your example shows) the situation might be different.
Hence, I would use the full (unrefined) model to compare the company to the full industry. Otherwise, you also may get strange results, as you are only comparing them on dimensions that not fully represent the situation of the company.

2)
Your sample sizes (especially that of the company) seem to be quite low for the complexity of model and in general to get robust and generalizable conclusions. I would be extra careful with the conclusions from your analysis.

[widefov]

Dear jmbecker,

Thanks very much for your reply. It was very helpful.

If I use the full model for the two data sets, a path rejected in the industry level test is significant in the firm level test.

Do you thinks that this is acceptable?

Actually, I have seen some research papers which apply IPMA followed by PLS assessment but they do not look at extra case but only their original model with IPMA.
May I ask you recommend any articles using PLS and IPMA (which are not recommended in this forum)?

Have a great day.

Regards,
Widefov

[widefov]

Any other thoughts, guys?

[Alizdh]

Hi everyone
I've conducted an IPMA but now... I got confused
Q1: Shall we use standardized total effects or unstandardized?
Q2: what is the uses of "Quality Criteria" section? I've read A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) but it has no information about QC section.