mixed model and bootstrapping

[rpalrecha]

Hello,

I am comparing three models, two are reflective (Model a and model b) and one is a mixed model (model c). Model c has reflective dependent variable and formative independent variables.

My research test is as follows
1) run full models a, b, and c. Compare the R2.
2) Bootstrap the models to determine the significant factors.
3) Reduce the models based on significant factors. Run the reduced models and do a F test to determine the unique contribution of each model.
I did this by running reduced model a + reduced model b. Noted the R2. Added reduced model c and calculated change in R2.

My question is related to the mixed model c. I can not bootstrap my mixed model due to low variability in the indicators. I used a path coefficient criterion of >0.2 to include the factors in reduced model. This is based on my observation that most path coefficients > 0.2 are also significant. Is this an acceptable work around?

Just as an additional note, if I make model c as reflective, I can bootstrap the model. In this case, I will have the same three factors in my reduced model as with the criterion of > .2 if I did it reflectively. Model c is formative based on theoretical reasonings.

Please let me know if I need to provide any additional details.

Thanks
Rita