Dear all
I am using ridge regression (RR) for modelling a set of data, and I
would like to calculate the adjusted R-square (adjusted multiple
correlation coefficient) that in contrast to the simple R-square takes
into account the number of parameters (degrees of freedom) used for the
model.
However I guess that in ridge regression this number is not defined as
in Ordinary Least Squares (OLS) where we simply use the number of
variables, since RR still uses all the variables but not in the way OLS
does.
Therefore I think that using the same degrees of freedom for RR and OLS
will be unfair for the former, in case I want to compare the predictive
abilities of the two methods.
Is anybody aware of how to deal with this problem and of any relevant
references on how to calculate Rsq-adjusted for RR?
Thanks in advance for your help
Best Regards
----------------------------------------------------------------------------------
Eleftherios Kaskavelis
Centre for Process Analytics and Control Technology
University of Newcastle Upon Tyne
Merz Court, NE1 7RU, UK
Tel: ++44-191-2225331,++44-191-2452849
E-mail: [log in to unmask]
-----------------------------------------------------------------------------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|