Nested second-order formative constructs

[angelhernandez]

Hi everyone,

Long time lurker here!

I've come across a problem in a research model I'm analyzing right now, and I hope any of you may help me.

The problem is that the model includes two nested second-order constructs, and I don't know the right way to proceed at this point.

The simplified situation is this (see picture attached):

• We have two second-order formative constructs, U (in green) and V (in red).
• V (red) is formed by 2 reflective constructs, X and Y.
• U (green) may is formed by 1 reflective construct (W), and the second-order construct V.
• We are using repeated indicators for both second-order fromative constructs

The first question here is:

When using repeated indicators for formative second-order construct, would the approach depicted in the above diagram be correct?

(We are especially concerned about the indicators of construct U)

Then, on how to proceed: following a two-step approach, after a first analysis of the models we obtain the Latent Variable Scores and we re-build our model and add the LVS to the list of indicators.

And here comes our big dilemma: which of the following scenarios would be correct?

1. Scenario 1: Subtitute only the part containing the nested second order construct (X-Y-V), include LVS for V, replace the repeated indicators of V by its LVS, but including all of W's indicators (see image), and then perform another two-step analysis with these values (which would then lead something like Scenario 2 next).
   
   
2. Scenario 2: directly replace all the indicators by each construct's corresponding LVS, and then perform a regular analysis (see image).
   
   
3. Scenario 3: a mix of scenarios 1 and 2. In this case, we would replace all the indicators by the LVS of each construct, except for U, where we would use the LVS of V and W as repeated indicators; then we would perform a new two-step analysis (see image).
   
   

Any hints on which is the best course of action? (any help would be much appreciated).

P.S.: Of course, our starting model could be wrongly defined (I don't really know if there is any limit to apply the repeated indicators approach). If so, please tell me.

Best,

Ángel.


EDIT: If this is not the right section to post this question, please move to the right one.

[bradley7680]

are you still working on this problem. you are welcome to contact me directly.

read my publications on researchgate first

thanks

Brad

[bradley7680]

Perhaps a solution may be to use nested CVTA. Confirmatory Vanishing Tetrad Analysis

See my Researchgate.

Bradley Wilson.
https://www.researchgate.net/profile/Bradley_Wilson2