I am trying to perform a partial least square model (PLS) using one response variable and 33 explanatory.. Before the analysis I have standardized all my data so i am expecting that the regression coefficients of my model will be also standardized-meaning that the higher regression coefficients will have a higher relative importance to my model and so on...
However, when I calculated the variable importance in projection (VIP values), the most important variables (the ones with VIP >1) do not coincide with the ones that have higher absolute values of regression coefficients.. Is this normal?
Thank you!!
Hello Victoria,
VIP values represent the potential import/utility of individual variables in accounting for scores on some outcome variable or vector. The key word here is, "individual." Regression coefficients show the "adjusted" weight/emphasis given to one variable, in light of the weights applied to the other variables in the model. As these have to account for overlap/redundancy/collinearity, they can differ quite a bit from the relative ordering of VIP values (as is happening in your own data set). If all IVs were uncorrelated with each other, then the relative ordering would/should be comparable.
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression