Different results in the outer measurement model

[Katharina]

Dear research community,

I have a question concerning the comparison of results of the outer measurement model. My situation is as follows:
I have one data set with 342 data points. And I have two different models - with the same variables, but in different relationships - that I want to calculate with the same data set.
For the inner structural model there must be differences due to the fact that I am calculating two different models.

But for the outer measurement model, the results are different too, when it comes to indicator reliability for example. For the two models but same constructs I have to exclude different items. When it comes to Cronbach's Alpha or composite reliability, I also have to report different values.

My question now is: Mustn't there be the same results for the outer measurement model due to the fact that I have the same data set and the same constructs (same results for indicator reliability, Cronbach's Alpha, composite reliability etc.)? Or do I have different results due to the fact that my constructs are already built into another model? So to say, is the analysis of the outer measurement model already based on the model relationships or on the constructs solely?

Thank you very much in advance!

Best regards
Katharina

[jmbecker]

The PLS estimation is model dependent. Thus, construct weights and loadings can defer with the same data in different models. Usually, when the construct measures are good and robust, the difference is quite small. If reliability and validity are smaller for constructs, then the results also differ more between models.

One question, however, could be: Why do you need to estimate different models and not analyze all relations in one model?

[Katharina]

Thank you very much for your fast and your helpful reply.

The reason behind my question is that I was testing different models and relationships in order to see differences and similarities.

Best regards
Katharina