forum.smartpls.com

Please help.

I have a second order formative endogenous construct in my model. But I struggle in validating the formative second-order construct with all formative indicators on the first and second level. Anyone know exact steps. Any good references? Whether we have to check VIF, outer weights etc for second order one?
From the base model using the repeated indicator approach, with latent variable scores I have built the updated model. Whether that updated model needs to be used for formative model assessment and all further structural model assessments? But I found difficulty in analysing the formative second-order construct. Please help.

What exactly are your problems?
Generally, you want to apply all assessment criteria that you use for first-order constructs also for the second-order construct just that the weights there are the path coefficients from the first-order constructs to the second-order construct (and not the repeated indicators).

I think I have a similar question. I have a formative-formative type HCM and I chose the two-stage approach. After the PLS Algorithm calculation in the first stage, there are two VIF values every indicator, and I am unsure which one I should look at. Thanks.

Oh, I am answering my own question now.

Basically one of the two VIF values of every indicator in the first stage is the same as the VIF values of all the indicators pointing to one single construct, so making it HCM does not change the VIF values, but it does change the weights of the formative indicators.

When I estimate the (outer) weights and loadings of an HCM, SmartPLS gives me two values for each indicator: one weight / loading "from" the indicator "to" the 1st order construct (see output: CONS3 -> CONS) and one weight / loading from the indicator to the 2nd order construct (see output: CONS3 -> SE).

If I understand correctly, I should use the weights and loadings of an indicator to the 1st order construct to validate the 1st order construct. Additionally, I use the weights (path coefficients) from the 1st order to the 2nd order constructs to validate the 2nd order constructs.

My question is: Do I need the weights / loadings that go "all the way" from the indicators to the 2nd order construct (e.g. CONS3 -> SE) to validate the 1st or 2nd order?

The following paper nicely summarizes the evaluation of second order constructs:
https://www.researchgate.net/publicatio ... in_PLS-SEM

jmbecker wrote: ↑Tue Jul 02, 2019 9:52 am The following paper nicely summarizes the evaluation of second order constructs:
https://www.researchgate.net/publicatio ... in_PLS-SEM

Dear Dr. Becker,

I have read the following paper that you recommend before. I have found how to calculate the Cronbach's Alpha for higher order construct, but I got confuse about where the value of average correlation between COMP and LIKE come from (0.665)? Moreover that value also will be used in calculating HTMT. I have re-read the paper many times but still I can't find the answer where 0.665 come from. I have tried to download the excel file but the 0.665 is just there no more explanation where it is come from.
Please help me to solve this problem, I really need it for my thesis.

Thank you in advance.

You will get this from the PLS output > Latent Variables > Latent Variable Correlation

jmbecker wrote: ↑Wed Nov 13, 2019 12:11 pm You will get this from the PLS output > Latent Variables > Latent Variable Correlation

Thank you for your explanation Dr. Becker. But I still don't get why we use 0.665, below is the result of PLS Output - latent variable correlation for the data using in your paper (Sarstedt, 2019):

I think 0.665 is the correlation between ATTR - PERF. Why we use that value sir (0.665)? Why we do not calculate the average between COMP - REPU and LIKE - REPU (0.883 + 0.920)/2?

I really hope you would help me by explaining the reason.
Thank you Dr. Becker.

Sorry for the late response, I was also confused in the first moment, but you are using the example data of the corporate reputation model that is built-in to the SmartPLS software. However, the paper uses a replication of the model on a different dataset. I remember that this might have been some nonsense requirement in the review process.

The data from the paper can be found here: https://www.pls-sem.net/downloads/sampl ... d-various/

jmbecker wrote: ↑Wed Dec 11, 2019 5:26 pm Sorry for the late response, I was also confused in the first moment, but you are using the example data of the corporate reputation model that is built-in to the SmartPLS software. However, the paper uses a replication of the model on a different dataset. I remember that this might have been some nonsense requirement in the review process.

The data from the paper can be found here: https://www.pls-sem.net/downloads/sampl ... d-various/

Thank you so much Dr. Becker for your explanation and help. It means a lot to me.