[R] logistic regression or discriminant analysis ?

[Daniel Amorèse]

Le 2002.05.24 16:55, Marc R. Feldesman a écrit :
> At 01:49 AM 5/24/2002, Daniel Amorèse took a glob of electronic fairy
> dust, 
> mushed it together in odd and bizarre ways, and ruminated:
>  >Le 2002.05.23 19:38, Marc Feldesman a écrit :
>  >> At 06:25 PM 5/23/2002 +0200, Daniel Amorèse wrote:
>  >> >Hello,
>  >What I have done: the correlation matrix tells me that many
>  >variables are correlated. Thus, I performed a lda using only 5
>  >variables (this selection is arbitrary performed among uncorrelated
>  >variables). The graphical output shows points clouds that are not
>  >circular: this result may suggest difference in covariance
>  >matrices, hence lda seems not to be the more suitable method for
>  >separating groups.
> 
> Elliptical point clouds do not, in themselves, suggest unequal covariance
> 
> structures.  If the major axes of the ellipses are all oriented in the
> same 
> direction and are more-or-less parallel, you don't have much evidence of 
> unequal covariance structure.  Unless the covariance structures are
> really 
> different, I've never seen much improvement using quadratic discriminant 
> analysis.  The problem with quadratic discriminant analysis, in my 
> experience, is that it is very difficult to interpret the results in the 
> context of the variables producing the group separations.

Ok, the elliptical point clouds I obtain are not oriented in the same
direction,
so I should give up with lda or qda

> 
> I'd strongly recommend Professor Ripley's "Pattern Recognition and Neural
> 
> Networks" (Cambridge University Press, 1996), as well as Hastie, 
> Tibshirani, and Friedman's "The Elements of Statistical Learning" (2001, 
> Cambridge University Press).  Both will help you considerably with your 
> problem.
>

Thanks, for these references. What a pity the library of my university
seems to 
be allergic to english-written books
 
> I also agree with Jon Baron that clustering techniques may be appropriate
> 
> here.

Ok, my ideas about this kind of approach are confused (I do not need
clustering
because my groups are already defined).
Some people told me about solving my problem using multinom() and
stepAIC().....  

> 
>  >Perhaps, qda should be used ?
>  >or logistic regression ? (this last method seems to be the more
>  >robust, independent to data properties).
>  >I know qda(), lda() or multinom() do not perform stepwise analysis,
>  >but, what I hope, is that some outputs from these functions can
>  >help in the selection of the most discriminatory variable subset.
>  >Thanks again for your help.
> D. Amorese
> 
> 
> Dr. Marc R. Feldesman
> Professor and Chairman
> Anthropology Department - Portland State University
> email:  feldesmanm at pdx.edu
> email:  feldesman at attglobal.net
> fax:    503-725-3905
> 
> 
> "Sometimes the lights are all shining on me, other times I can barely
> see,
> lately it's occurred to me, what a long strange trip it's been..."  Jerry
> &
> the boys
> 
> 
> 
> Powered by Tyrannochoerus - the 2.2 GHz WinXPP Box
> 
> 
> 
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._