Hi,
I am working with a logistic regression model with a binary dependent
variable. My question is how to assign the relative importance (%) of each
independent variable in explaining the variation in the dependent variable.
I found one approach to do this in Marketing Science, Vol 17, No.1 (1998)
pp45-65: "Importance weights were calculated for each independent variable
as |B|S, where B is the estimatated coefficient and S is the STD.DEV. of the
predictor variable. The weights were rescaled to sum to 100" Does this make
sense? Is there another approach to do this?
Any comments are much appreciated and welcome.
Thanks!
Leo Guelman
|