Hi, I would be very grateful for some help on a problem I have. I would like
to find out the latent roots and the first element of each vector of the A'A
Matrix which is equivalent to the calculation of the two augmented matrix
[Y'X]'[Y'X]. The reason I would like this information is because according
to Subhash Sharma and William L. James paper 'Latent root Regression: an
alternative approach for estimating parameters in the presence of
multicollinearity' (Journal of marketing research, may 1981) you can
calculate if a singularity (multicollinearity) is non-predictive or
predictive. If it is predictive then you want to retain it in the model and
use a unbiased estimation procedure (OLS), however if it is non-predictive
(Absolute value of all the first elements are less than 0.3 and their
corresponding latent roots are <0.1) then you need to use a biased
estimation precedure such as Ridge Regression. The type of data I would be
looking at will include a large number of variable (20+) and approx. 300
cases (respondents).
I currently use SPSS but I am not aware of any procedure in SPSS to do these
calculations.
Thanks
JB
_________________________________________________________________
Express yourself with cool new emoticons http://www.msn.co.uk/specials/myemo
|