Second-order formative construct (exogenous)

[SHU]

Dear all,

in order to model my first-order reflective second-order formative construct, which is an exogenous variable with 5 dimensions with equal amount of items in the structural model, I followed the guidelines of Becker, Klein & Wetzels (2012), Hair et al. (2014) and Ringle, Sarstedt & Straub (2012), and applied the repeated-indicator approach.

As I was interested in the difference between the repeated-indicator approach and the two-stage approach, I have also tested my model with the latter method. The paths of the structural model are both quite similar. What I found irritating is that all paths coefficients / weights between the LOC and HOC are positive and significant when using the repeated-indicator approach, but they are flipping the signs and are non-significant when using the two-stage approach. On the other hand, the HTMT values of the LOCs are all good when using the two-stage approach (checking them in stage 1) and partially violating the HTMT threshold when the repeated-indicator approach.

I am aware that both approaches would yield in “non-identical results”, but I have expected them to be somehow more close…

Does the repeated-indicator approach tend to show somehow perfect positive and significant paths coefficients / weights between the LOC and HOC because it explains all the variance of the higher-order (exogenous) construct?

Looking forward to you reply.

[cringle]

In the advanced book on PLS-SEM (www.pls-sem.com), we tackle this issue more closely (it will be available next May). Usually, the 2 stage approach works fine, especially when you select mode B for the second-order construct (double click on the LV in SmartPLS).

Kind regards
Christian

[SHU]

Thank you for your reply. I am looking forward to the detailed insights in the topic in the upcoming book!
Bests Susanne