I am just now reading a paper on LASSO and it begins by stating that it assumes all predictors are centered. Is this also true if, say, 28/30 predictors are binary? Should the binary predictors be centered? Should the two continuous predictors be centered? Are the results void if they are not?
Asked
Active
Viewed 22 times
0
-
@EdM Thanks for the link. It seems that there is no generally accepted response? I wonder if the answer changes when there are both binary and continuous predictors in the model? Also, I wonder if centering is always necessary even if scaling is not? – Dave Aug 28 '20 at 17:10
-
There's no "generally accepted" method; I'm afraid that many people just blindly accept the software defaults. You have to apply your knowledge of the subject matter and make the decision yourself for your categorical predictors, making sure to report what you did when you write it up. The linked page outlines the issues to consider. That's independent of handling continuous predictors, which almost always should always be on the same relative scales. Centering isn't necessary for continuous predictors, but it's easiest to scale (by standard deviation) after centering (to a mean of 0). – EdM Aug 28 '20 at 18:13