howto_sindscal.txt

《多元数据分析》是我正在看的一本书

  PC VERSION OF SINDSCAL AT RUTGERS BUSINESS SCHOOL

     When Doug Carroll and I were teaching a course in
1993 on MDS at what was then locally known as "GSM"
(Rutgers Graduate School of Management, now renamed the
"Rutgers Business School of Newark and New Brunswick"
and still headquartered at the Newark campus), it
became apparent that our students needed IBM PC-
compatible versions of some of the classic Bell Labs
MDS programs, originally written for certain
mainframes.

     Using the Microsoft MS-DOS FORTRAN (v. 5.1 of
1991) compiler, I minimally modified SINDSCAL
(Pruzansky, 1975).  Only this memorandum has been
updated; the .exe file remains as in 1993.  The full-
length 17 pages of paper documentation for the original
program can be obtained at the site:

http://www.duxbury.com/  (Select "Book Companions")

That documentation is not easy to scan optically and
include here as a file because column alignment in
sample input listings gets lost.  On a related topic,
the user of SINDSCAL should note the when it and its
precursors were written during the 1960's and 1970's,
the input FORMAT options in FORTRAN were much less
flexible and forgiving than those today.  If you are
not willing to study the older formats from a reference
manual, then you should probably consider using one of
the alternative programs mentioned below.  I have seen
too many would-be users of SINDSCAL fail simply because
they did not understand these formats, or how their
PC's text processors were arbitrarily breaking rows of
an input data matrix, etc.

     This program will not run under releases of the
Windows 2000 operating system or subsequent versions of
that Microsoft operating system.  The .exe file was
created under a DOS-based compiler and assumes that the
user's PC has a math co-processor (not automatically
included in PC's until the Intel 586 chip became
available; you will need to check to see that your PC
has the co-processor).

     This CD-ROM contains SINDSCAL.EXE.  Either copy it
to the hard disk and then type

SINDSCAL

or run it (more slowly) from the CD-ROM by typing, for
example,

e: SINDSCAL

or using a different device letter for your CD-ROM,
specific to your PC.

     The program will prompt you for the filenames 
(paths are optionally allowed but should be VERY brief)
of all input and output files.  Note that all input
files must be ascii (.txt format), with no left margin. 
These should be prepared using a text editor and saved
in ascii format.  SINDSCAL has no internal editor to
prompt you for input data.

     The output file will also be an ascii file, with
the usual FORTRAN carriage control conventions.  When
you read the output file with an editor, remember to:
(a) declare that the file is to be read as an ascii
file, (b) set the horizontal margins to be as narrow as
possible, and (c) take care of carriage control.  

     Finally, it is to your advantage to print the
output file in "landscape" format and to use a small
font size if one is available to you.

     The CD-ROM contains a sample data set (Paul Green's
"letter R" artificial data) given in the original Bell 
Labs documentation (Pruzansky, 1975) for SINDSCAL.  The
input file is called pegrdat; the output file is pegrout.

     You will be asked for a filename for the machine-
readable output (obtained from specifying a value other
than -1 for the control parameter IPUNCH; see
Pruzansky, 1975, p. 6) regardless of whether you have
asked for such output.  In general, you should always
ask for it.  Note that if the user provides an initial
configuration for the stimulus space (as is done in the
file pegrdat, by setting IRN = 0), that configuration
must follow the input data on the same file; see
Pruzansky (1975, p. 14).

     We also wish to endorse some alternative software
for MDS.  The recently implemented version of SMACOF,
now available as PROXSCAL in the CATEGORIES module for
SPSS, includes a three-way nonmetric option (fitting
the INDSCAL weighted Euclidean model to ordinal -- or
nonmetric -- data) as well as two-way nonmetric MDS,
with various options for missing data, unfolding
analysis, etc.

     Well-written software for fitting both two- and
three-way MDS models is included in SYSTAT, originally
founded by Leland Wilkinson (SYSTAT, 2002).  

     Numerous theoretical and methodological problems
make ALSCAL considerably less appropriate for most MDS
analyses (see, e.g., Weinberg and Menil, 1993, or
MacCallum, 1977).

                                        Phipps Arabie
                                        October 2002

                     References


Arabie, P., Carroll, J. D., & DeSarbo, W. S. (1987).
Three-way scaling and clustering. Newbury Park, CA:
Sage.  See Appendices A and B.

Carroll, J. D., & Chang, J. J. (1970). Analysis of
individual differences in multidimensional scaling via
an N-way generalization of "Eckart-Young"
decomposition. Psychometrika, 35, 283-319. (Reprinted
in P. Davies & A. P. M. Coxon (Eds.), (1984). Key texts
in multidimensional scaling. Portsmouth, NH:
Heinemann.)


MacCallum, R. C. (1977). Effects of conditionality on
INDSCAL and ALSCAL weights.  Psychometrika, 42,
297