I am fairly new to SEM in general and use SmartPLS to test my reflective-formative second-order construct Perceived Transparency as a result of enterprise system implementations for my master thesis.
I use the repeated-indicator approach to model the second-oder construct (HOC) and its 4 first-order constructs (FOCs).
When I run the model consisting of only these constructs, I get satisfying results for indicator loadings, AVEs, cronbach's alpha, CR,Fornell-Larcker Criterion, HTMT, and cross-loadings for the FOCs. All FOCs also had significant wheights on the HOC (between 0.3 - 0.5) and multicollinearity was not a plroblem since the inner VIFs were all below 3.3
Now I wanted to test the impact of the HOC on several dependent variables (DVs). However, when I add them to the model, the wheight of one of the FOCs (FOC A) on the HOC increases drastically, while the wheights of the other three FOCs decreases in the same manner. The item loadings of the three FOCs also decreased, which is why convergent validity was no longer satisfied. I checked the correlation matrix and found that FOC A correlated higher with the DVs than the other FOCs. I excluded FOC A from the model and found out that it explained almost all of the variance in the DVs alone and that the path coefficient to the DVs are almost as high as of the complete HOC.
Now I have two questions:
(1) Can someone explain to me why the weights to the HOC change so much because of the addition of dependent variables and why the outer loadings of the other FOCs are decreasing? I could not find an answer to this in the literature and I am fairly unexperienced in PLS and SEM in general. I also tried the two-stage approach and had the same problems. The outher wheight of FOC A increased to over 1, while the other wheigts were around 0.1.
(2) I now want to assess the multidimensionality of my construct because it seems like my HOC is not valid. MacKenzie et al. (2011) recommend to test if the direct effects of the FOC on the DVs are higher than the direct effects of the HOC. For testing the significance of this path, thy recommend a chi-square difference test. However, this is not possible with PLS. Can anyone recommend my how to assess the multidimensionality of my construct in PLS?
Thanks a lot for your help.
Frequently asked questions about PLS path modeling.
1 post • Page 1 of 1