Dear all,
I'm using PLS structural equation modelling (with SmartPLS) to measure a socio-cognitive model. I'm still new to PLS-SEM. I need to add some demographic variables into the model. Since there are many such demographic variables (in both scale and dichotomous), I need to perform bivariate analysis to drop the variables that are insignificant. Since PLS-SEM is a variance based method, how can I achieve this?
Dear Anuradha,
these are the steps usually recommended to perform your SEM analysis, which is a 2-step technique by the way:
1) perform a bivariate correlation analysis (e.g., Pearson "r") to primarily see what are the significant correlations among your variables (both demographic, control, and latent of your conceptual model); this is a very important initial step to realize what are the strongest and weakest variables' associations;
2) next, conduct a CFA (confirmatory factor analysis, 1° step) of your latent variables to assess the model fitting indexes (both relative and absolute); remember to be "parsimonious" (less is better in SEM!) and retain only the indicators with acceptable factor loadings (>0.50, usually); in this way, you'll know the validity of your model, and also if some errors have to be correlated to improve the general fitting.
3) finally, run your SEM model: in this step, you'll have to follow a specific validated or exploratory theory allowing you to "control" the hypothesized regression paths with some of your demographic variables; in this step too, have a look at the model fitting indexes of your structural model.
Have a look at these video tutorials by James Gaskin using SmartPLS:
https://youtu.be/qK05XYx5CwU
Hope this help,
best luck with your study!
- Badges
- Science topic