From: Daniel Chessel (
Date: Tue Nov 13 2001 - 14:18:17 MET

At 10:17 13/11/2001 +0100, Kent Löfgren wrote:
>I have used ADE (superb).

Merci beaucoup

>Now I have SPSS, "Categories". Difficult. I am NOT statician, not expert.
>I only do Correspondence analysis (binary data, 0 & 1)
>Question: What are major differences between correspondence analysis in ADE
>and correspondence analysis (HOMALS) in SPSS?

L'équivalent de HOMALS dans SPSS est le module MCA dans ADE-4

L'analyse des correspondances multiples est assez diversifiée comme on peut
le voir dans
Tenenhaus, M., and F. W. Young. 1985. An analysis and synthesis of multiple
correspondence analysis, optimal scaling, dual scaling, homogeneity
analysis ans other methods for quantifying categorical multivariate data.
Psychometrika 50:91-119. Il y a de nombreux points de vue.

La MCA de ADE-4 est comparée avec celle de R (mca dans la librairie MASS de
B. Ripley) dans
Elles sont assez différentes.

La MCA est proche de la COA du tableau disjonctif complet, mais il vaut
mieux dans ADE-4 utiliser MCA: Multiple Correspondence Analysis que COA:
COrrespondence Analysis après CategVar: Categ->Disj. Elle est proche aussi
de la COA sur le tableau de Burt (COA: COrrespondence Analysis après
CategVar: Categ->Burt)

HOMALS de SPSS est de la famille. Mais, c'est une grande différence, elle
accepte les données manquantes et utilise les moindres carrés alternés.
Exactement comme NIPALS a comme cas particulier la PCA quand il n'y a pas
de données manquantes (voir Tenenhaus, M. 1999. L'approche PLS. Revue de
Statistique Appliquée 47:5-40) MCA est un cas particulier de HOMALS quand
il n'y a pas de données manquantes.

Voir bibliographie dans :

Most commercial software packages contain a procedure that performs
Homogeneity Analysis (also called Multiple Correspondence Analysis) -PROC
CORRESP in SAS, program CA in BMDP, program Homals in SPSS [9, 10, 11].

The first two procedures are much closer to the "French school" and treat
homogeneity analysis as an extension to the simple correspondence analysis
[7, 8]; therefore, they transform homogeneity analysis into an eigenvalue
problem of the Burt matrix.

Whenever there are missing values for one or more variables in a given
observation, the entire observation is excluded from any subsequent
analysis. Their graphical output consists (in our terminology) of the
object scores plot and the combined category plot. The main advantage of
these two programs is that they come with all of the data manipulation
functions that are part of a general statistical package.

The third program (in SPSS) is a translation of the Homals code written in
the University of Leiden in the mid-80's. Hence, it reflects the philosophy
on homogeneity analysis of the "Dutch school" (as our program does).

It uses the ALS algorithm in and contains most of the two dimensional plots
of our program. The main advantage of our program is that it is menu
driven, utilizes the information from partly incomplete data (as opposed to
the SAS and BMDP programs), offers a large variety of plots as well as
three dimensional plots (as opposed to the SPSS program), and allows the
user to "zoom in" into clouds of points of particular interest. Since,
homogeneity analysis is a data exploratory technique, our program remains
very close to this goal (and the dictates of the "Dutch school"), by
utilizing the advanced graphic capabilities of LISP-STAT.

Daniel Chessel
Universite Lyon 1 - Biométrie et Biologie Evolutive - Bât 741
69622 Villeurbanne CEDEX
Tel : 04 72 44 82 77 - (33) 4 72 44 82 77

This archive was generated by hypermail 2b30 : Thu Jul 04 2002 - 09:39:14 MEST