Re: CateqCovar or equivalent, used in Thema24?

From: Sylvain Doledec (
Date: Wed Apr 08 1998 - 07:13:41 MET DST

Dear ADE user

As usual, David Stodolsky points out some critical problem on our product I
mean the update of documentation. You refer to a program that was formerly
available in the ADE 3.7 version. This module was used there as a numerical
help for the interpretation of multiple correspondence analysis.

MCA is some kind of equivalent to PCA for qualitative variable and maybe
the best references for this method are Tenenhaus, M. & Young, F.W. (1985)
An analysis and synthesis of multiple correspondence analysis, optimal
scaling, dual scaling, homogeneity analysis and other methods for
quantifying categorical multivariate data. Psychometrika : 50, 1, 91-119.
Pialot, D., Chessel, D. & Auda, Y. (1984) Description de milieu et analyse
factorielle des correspondances multiples. Compte rendu hebdomadaire des
séances de l'Académie des sciences. Paris, D : 298, Série III, 11, 309-314.

This method investigates the relationships among qualitative variables
which amounts to transform the qualitative data into quantitative scores.
The link between the initial qualitative variables and the new quantitative
scores is investigated by way of correlation ratios which are the ratio of
the variance of scores of individuals that share a given modality to the
total variance. The more the correlation ratio the best modality are
separated using the scores. The property of these scores is to maximize the
average correlation ratios of variables (eigenvalues of the MCA). So with
this method you can see the prominence of the variables of your data set
and the relationships between variables.

If you are kean on Matrix algebra you can easily compute the average of row
scores for each modality for each axis using the Mat Alg: Diagonal Inner
Product option using xx.cmta as X file xx.cmli as Y file and xx.cmpl if
necessary (this is necessary in the "Chat" example). If you divide the
resulting value for one axis by the corresponding eigenvalue then you get
the scores of modalities that are computed in xx.cmco. If you use the
DDUtil: Add norm scores before and use the xx.cml1 file in place of the
xx.cmli file then you directly get the xx.cmco file
If you want to have update information about MCA by ADE-4 then use the MCA
module documentation that incorporates much more now than the Thema
documentation. Especially new graphical helps (module TabCat for example)
are available

>I would also like to read in an Excel format file that includes variable
>names, if there is a way to do this.

You cannot read directly Excel format with ADE-4. You can only read text
files. Variable names should be incorporated in a separate text file to be
used in graphical modules.

>When trying to work the example in Thema24 on the Mac, I kept crashing when I
>tried to use text2bin. However, I discovered that the Text2Bin in MacMul

This is always the narrow path to ADE-4 graal. I have no advise there just
verify you file (carriage return), characters in you file, decimals,
seprator (tab or something else). We formerly had a text about the various
way of entering data may be we should reincorporate this information.
So be persevering!

>Then I had trouble with "Read Categ File". Restarting without inits did't
>solve the problem. When I tried to extract the Meteq data from the "Data"
>hypercard stack to test "Read Categ File", this also caused a crash.

Well no crash for me. I must say here it works very well. Please tell us more.
>Now I crash whenever I try to convert a large quantitative file to a
>qualitative file.

Again please tell us more.

>Getting documentation has also has had its problems. Some of the English
>docs are in ./ADE4/DocThem, not in ./ADE4/DocThemUS. What is the point of
>the separate directories?

Again you point of some critical problem of our documentation. The six
first topic documentation of the volume starting ordination are available
only in English because they are the obly available file they are
incorporated to ./ADE4/DocThem

>My impression is that ADE-4, which has obviously had a lot of work invested
>in it, could be much more useful with a bit more support and documentation.

goush!n Thanks anyway

 Sylvain Dolédec
 Universite Claude Bernard Lyon 1
 Bat 401C
 69622 Villeurbanne CEDEX - France
 Tel : (33) 4 72 43 13 63
 Fax : (33) 4 72 43 11 41
ADE-4 package is available on the Internet

This archive was generated by hypermail 2b30 : Sat Feb 10 2001 - 10:21:55 MET