Dear colleagues,
I am analysing a data set on the morphology of 128 plant specimens of
Flaveria (Asteracea) from 13 localities in Peru.
Morphology has been characterized by means of 47 variables, the
particular aspect being that they are dimensionally heterogeneous: 18
are quantitative, 7 are semiquantitative, 15 are binary and 7 are
qualitative.
The overall working hypothesis is that there are may be different
species, so I am interested in exploring morphological gradients and
discontinuities, and identifying variables that best differentiate
morphotypes.
Initially Principal Component Analysis and Discriminant analyses will be
performed. Prior to multivariate analyses, I have turned the qualitative
variables (5 have 3 states, 1 has 4 states and 1 has 5 states) to
dummies (25 in total), and for each qualitative variable I have
eliminated one of the (correlated) dummy states resulting in a total of
17 dummies. My final data set then consists of 18 quantitative
variables, 7 semiquantitative variables and 32 binary variables.
My main doubt is on how to handle all the variables to address my
working hypothesis. In the literature, you may find examples where:
1.- Multivariate analysis are performed separately for each subset of
quantitative, ranked and binary variables. This may make difficult to
bring results from each analysis together.
2.- Correlation-based PCA and z-score standadization prior to DA are
employed. Some authors report aberrant (i.e. ecologically meaningless)
correlations when such standardization is used.
3.- Using similarity measures with standardization power like Gower's or
Estabrook's. PCA could not be used, still generalized DA could be applied.
I would appreciate if you could give me some feedback on the latter
issue. How would you go about such an analysis?
Thanks for your time.
Salva
Salvador Herrando-Pérez, Biólogo acuático, BSc. MPhil.
FUNDACIÓN OMACHA, Associated Researcher (www.omacha.org)
DIRECCIONES POSTALES / POSTAL ADDRESSES:
C/ Padre Jofre 19, 12006 Castellón de la Plana, ESPAÑA
Fundación Omacha, Diagonal 86ª # 30-38. Barrio El Polo, Bogotá, COLOMBIA
PARA CONTACTAR / CONTACT INFORMATION
[EMAIL PROTECTED] / [EMAIL PROTECTED] / [EMAIL PROTECTED]
(mensajes/messages>
5 Mb)
Teléfonos, España: +34 964036814 / +34 657787340
Teléfono, Colombia: +571 2362686
Ahora estoy en España... / I am currently in Spain...
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
