For some time ago I asked for help on Multivariate models for
continuous AND categorical variables. Thanks for all answers. Here is
a summary of the answers:
Locaton models and generalisations of it:
-----------------------------------------
The categorical variables are treated as being multinomial and the
continuous variables are assumed to have a multivariate normal or
elliptic distribution GIVEN the categorical variables. The mean of the
multivariate continious variables typically vary between cells of the
multinomial, whereas the covariance matrix may have similar structure
over cells, for instance equal covariance or proportional covariance.
Olkin and Tate (1961) is the original reference. The latest reference
I have found is
Liu, C. and Rubin, D.B., 1998, "Ellipsoidally
symmetric extensions of the general location model for mixed
categorical and continuos data", Biometrika 85, 673-688.
Graphical models and conditional gaussian distribution:
------------------------------------------------------------_
In the context of graphical models and hierarchical interaction
models, a model similar to the location model are the conditional
gaussian (CG-) distribution.
References:
S.L. Lauritzen, "Graphical models", Oxford University Press, 1996 (good
theoretical background),
D. Edwards, "Introduction to Graphical Modelling", Springer, 1995.
including for free the statistical package MIM ?
Whittaker, J. (1990) Graphical Models in Applied Multivariate
Statistics, Wiley
Multilevel models:
------------------
Goldstein, H, 1995, "Multilevel Statistical Models", Arnold (Chapter 7).
MLwiN is a multilevel modelling package from the institute of
education:
http://www.ioe.ac.uk/multilevel/
Other suggestions were:
_______________________
Lance, G.N. and Williams, W.T. (1967). Mixed data classifiicatory
programs. I. Agglomerative systems. Austal. Comput. J. 1: 15-20.
The stats package GENSTAT (ex Rothamsted) will handle these problems.
Discriminant analysis:
A. Faraj, "Interpretation Tools for Generalized Discriminent Analysis"
in
New approaches in Classification and Data Analysis, Springer Verlag,
1994
Multiple correspondence analysis
--
Magne Aldrin email: [log in to unmask]
Norwegian Computing Center tel +47 22 85 25 00
P.O. Box 114 Blindern fax +47 22 69 76 60
N-0314 Oslo WWW: http://www.nr.no/~aldrin/
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|