• Recent writings of J.-P. Benzécri: Revue MODULAD, "Si j'avais un laboratoire ..., No. 38; and "L'Analyse des données : histoire, bilan, projets, ..., perspective", No. 35. (English translation of part of page 1 of "Si j'avais un laboratoire ..." in text - Latex format.)

• A recent issue of Electronic Journal for History of Probability and Statistics, Vol.4/2 December 2008: "About the History of Multivariate Exploratory Data Analysis", with many articles dealing with the history and development of Benzécri's approach to analysis of data and science.

  This includes a version in English of the article L'Âme au bout d'un rasoir, The Soul at the Razor's Edge, originally publised in the journal Les Cahiers de l'Analyse des Données, vol. V, no. 2, 1980, pp. 229-242.

  Updated version (10 May 2009) of L'Âme au bout d'un rasoir, The Soul at the Razor's Edge.

• Table of Contents (in PDF).
• Chapters:
  1. Avant-Propos, Foreword (in French and English) by J.P. Benzécri
  2. Introduction
  3. Theory of Correspondence Analysis
  4. Input Data Coding
  5. Examples and Case Studies
  6. Content Analysis of Text
  7. Concluding Remarks, References, Index
• Link to online description - Chapman & Hall/CRC Computer Science & Data Analysis series
• Link to Amazon
• Link to Amazon.co.uk

Some of the programs, especially the R and C ones, are in ascii text. Some others are binary (e.g. the clustering DLL program, and the Java class files). The Java code and the data sets are collected together in tar files, to be extracted using WinZIP or tar or some similar utility.

1. Software in R

• Correspondence analysis
  • Correspondence analysis, determining rates of inertia, projections on factors, contributions, and correlations.
  • Supplementary rows.
  • Supplementary columns.
• Hierarchical clustering
  • Background details, and examples of use on Intel Windows and Unix systems.
  • All functions in one (standalone) R/S-Plus file (call in R as function hierclust).
  • C source code for the clustering.
    Next a DLL library for use on Intel/Windows systems.
    Alternatively compile on a Unix system to get the object code (cf. "Background details..." above); and
    R/S-Plus script that calls the C code (call in R as function HierClust).
• Interpretation aids
  • FACOR - factors and clusters.
  • Print and format FACOR results.
  • VACOR - variables and clusters.
  • VACOR - variables and clusters, supporting C code for clustering.
  • Print and format VACOR results.
• Utilities and data
  • Fuzzy or piecewise linear data coding.
  • Plot axes.
  • Data sets to be used with R programs. Currently 3 data sets (but see also the data sets available below that are ready for use with the Java program: these can also be easily prepared for use with the R programs).

• Analysis of multiple text files.
  • aviation-reports-data.tar, 47 aviation accident reports, the list of these files, programs txtanalysis.c and xtabulate.c, and output files words.txt, words0.txt, and xtabulate.txt, used as examples in the following.
  • txtanalysis.c program, that is run as follows: txtanalysis filelist.txt [words.txt]
  • xtabulate.c program, that is run as follows: xtabulate words.txt filelist.txt [xtabulate.txt]
  • word_analysis.c, program to check for sufficient number of occurrences of words in all texts. Hence, this program yields a common word-list. This common word-list can be used by xtabulate.
• Analysis of a single (large) text file.
  • arist10.txt, Aristotle's Categories. Note: we removed the legal information (so as not to influence the analysis) to yield the file arist10x.txt.
  • docanalysis.c program, to produce a word list from a single text file. Example of use: docanalysis arist10x.txt words.txt. for the Categories 1260 words are found. It is best to filter or cull these (or else the cross-tabulations, to follow will be very large).
  • doctabulate.c program, to produce a cross-tabulation for each of the chapter and section levels in the Categories. Use: doctabulate arist10x.txt words.txt out. This produces the cross-tabulations, or contingency tables, out1.txt, out2.txt, out3.txt, out4.txt, corresponding to the different section levels in this book.
• Notes: programs txtanalysis and xtabulate should handle acceptably (i) accented characters, and (ii) use in a Mac OS X environment. (The latter issue is that memory allocation is already catered for; so the line "#include <malloc.h>" at the start of the file should not be present.)

• Tar file containing all *.java source files. Compile these using the JDK (Java Development Kit) command:
  javac *.java -d java-classes
  where java-classes is the target directory.
• Tar file containing all *.class class files. Run these using the JRE (Java Runtime Environment) command:
  java DataAnalysis
  It is best to do this as a command-line instruction, in the classes directory. A pop-up window will prompt for the input file name.
• Tar file containing a set of *.dat data files for use with the Java software. Currently containing 28 data files. By removing certain lines and labels these data sets can be easily read into R.

Errata

• P. 35, 2 lines above expressions (2.4), change j \in J to: i \in I.
• P. 37, line -14, change f_K to f_I. And in paragraphs on lines -11, -10, delete opening words and terms to begin sentence with: We can right-multiply the eigen-equation above ....
• P. 113, line 6 of text, 40% should be 48%.
• See above for changes to the C program for hierarchical clustering (minimum variance/Wards with weighting of rows/cases), and associated R calling script.

Updates

• A new version of facor, with an example of use at the start of the program. Input data set, casa2.prn (text file), a characterization with 13 person and place attributes of the 77 scenes of the film, Casablanca.

5. Other R Software for Correspondence Analysis

• FactoMineR
• CARME-N.ORG

6. Book Reviews and Survey Papers