Hi Dr. Franggakis;
This can be explained because of collinearity and suppressor variable in
multiple regression models. In the first scenario, you have both correlated
variables and suppressor variables in the second scenario you do not have this
problem. I do wonder why to do not use the scale elastic net for this
particular problem.
Good luck,Oslo
On Saturday, November 5, 2016 4:29 PM, Constantine Frangakis
<cfran...@jhu.edu> wrote:
I would appreciate any comments to the following question.
I am trying to build a model for survival based on 155 patients and 70
covariates using lasso. Lasso picks, three variables only, say X1,X2,X3, and
omits the others. I wanted to check why a particular (clinically important)
variable, say X4, is omitted by lasso. One of the things I did was I ran lasso
on X1,X2,X3 and X4 only. The results (coefs) I get are different from running
all 70 variables, and in fact now X4 is not omitted.
Why is that ? should it not be that the global (among all 70 variables)
optimum, which is X1,X2,X3 and not X4, be also the local (among the four only)
optimum ?
Thank you for your consideration
Constantine Frangakis, PhD
Professor
Departments of Biostatistics
Psychiatry, and Radiology
Johns Hopkins University
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
