Hi Voltolini,
I have a couple of suggestions.
1) When dealing with a large number of variables which may or may not be
interrelated, I
recommend the use of Principal Component Analysis. PCA allows you to find a
smaller
set of linear combinations of the original variable set that takes into
consideration most of the
variation in the that original set. It also decorrelates already correlated
variables.
2) Discriminant Analysis is most useful in determining to what class a
particular data set was
derived. For example, if there are several classes of dispersion syndrome
(say A, B, ...), then DA
can be used to interrogate a data set and determine to which class that data
set was taken. This
requires that you have data sets that are known a priori to have come from a
particular class of
dispersion syndrome.
3) If my assumption in (2) of a priori sets of data belonging to
well-defined classes of dispersion
syndrome is wrong, then try Cluster Analysis. CA takes raw data and
determines a set of classes to
which the data belongs if you already don't know the classes before hand.
In any event, you are embarking on the great adventure of Multivariate
Statistics for which I
whole heartedly recommend the book "Multivariate Analysis", by K. V. Mardia,
J. T. Kent, and
J. M. Bibby, . London: Academic Press, 1979.
Good luck.
Rgds
mjg ([log in to unmask] ; 203-353-8100 x277)
PS - I hope that this isn't an introductory course. I suspect that this is
at least advanced undergraduate
or graduate level.
=======================================================================
Dear friends, some of my students are organizing a data bank with
several qualitative and quantitative variables like fruit color,
dispersion syndrome, fruit length, geographical distribution, type of
leaf morphology, presence of spines, etc., and we would like to test
what variable is more associated with the type of dispersion syndrome.
Probably this is the case for a log linear and/or logistic models
but.... I have in the same analysis qualitative (e.g., fruit color) and
quantitative variables (e. g., fruit size) and I would like to know if
it is possible to analyze the complete data set to test interactions,
etc.!
Any suggestions (please!!!) ?
Thanks..... Voltolini
Disclaimer: Opinions expressed in the above message or in any attachment are
not those of Telekurs (USA) Inc. or its parent and related companies
("Telekurs"). Any advice contained herein is not to be used as legal advice.
Telekurs accepts no responsibility for errors in the above message or for
losses or damages resulting from its use, including losses or damages caused
by the running of any attached program.
|