Relative impact

noetsi

Fortran must die
#1
Say for OLS regression. Say you get asked which of these variables have the most impact on the dependent variable. So you can rank the predictors from most to least influence. Say most of the predictors are dummy variables but some are not. Is there a good way to do this? In a decade plus I have not found one. :p Standardized predictors won't work with dummy variables as best I can tell.
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
How is the DV formatted? Bayes network or partial R^2, but you need to control for precision around it, which may be sample error dependent. If you have holdout data, perhaps you can test in the new data using partial R^2 or whatever.
 

noetsi

Fortran must die
#3
In this case as an interval variable although I have dummy DV as well. I don't have a hold out data set.

I am not familiar with bayes or partial R squared. I will look that up. I almost never use Baysian approaches.
 

hlsmith

Less is more. Stay pure. Stay poor.
#4
I mean to say naive bayes, but it is usually used more with binary DVs.

Partial R**2 with confidence intervals is my go to for linear regression.