Page 1 of 1

MGA at second order construct

Posted: Wed Sep 05, 2018 3:02 am
by Kim78
I am a new researcher and recently I am working on a model that involved second-order constructs (reflective-formative, type ii), I used the two-stage approach. I draw a model with all variables and constructs including both first order and second order constructs and generate the latent variable scores for all the first order constructs. Then I use the first order latent variable scores as the indicators and link them to the second order constructs in a formative way and run the model again.

But I would like to perform MGA on two group of respondents. In my first order constructs, I can select Data Group information. But at the second order construct, the latent variables did not contain this information which is quite logical.

Given this situation, may I know how can I do MGA at the second order construct?

Thank you.

Kim

Re: MGA at second order construct

Posted: Wed Sep 05, 2018 8:55 pm
by jmbecker
You just need to also include the group information for the second-stage model. The observations in the first-stage model are in the same order as in your data set. Hence, you can just include the grouping variable from your first-stage model in your second-stage model.

Re: MGA at second order construct

Posted: Tue Sep 10, 2019 4:25 pm
by lilyv
jmbecker wrote:
Wed Sep 05, 2018 8:55 pm
You just need to also include the group information for the second-stage model. The observations in the first-stage model are in the same order as in your data set. Hence, you can just include the grouping variable from your first-stage model in your second-stage model.
Dear Mr. Becker,
In this case, we have to use unstandardized or standardized LOCs score for the second-stage model ?

Re: MGA at second order construct

Posted: Thu Oct 10, 2019 7:52 am
by jmbecker
Both are working. Usually I would recommend to include unstandardized LOC scores because it enables you to use advanced PLS techniques such as IPMA, PLSPredict, etc. that rely on unstandardized estimates to some degree.
But for the normal analysis both should provide the same standardized path coefficients, loadings, etc.