Professional Documents
Culture Documents
University of New York at Stony Brook for updates and corrections to the contents
of the CD, thanks to the generosity of Professor F. James Rohlf. In two recent
multivariate texts of RAR, the accompanying programs are written in copyrighted
code, which means that the user is required to acquire the means of accessing this
code. We have desisted from this practice, since it would defeat the main practical
purpose of the book. We supply our own compiled F O R T R A N and C programs
with instructions for entering data; each method is illustrated by one or more sets
of observations typical for the class of problem treated with an emphasis on the
peculiarities of geological data. The programs have been constructed outgoing from
our own research commitments. Note, however, it is not our intention to provide a
self-contained, hierarchical system such as offered by F. James Rohlf's NTSYSpc.
The compiled programs have mainly a didactic p u r p o s e - a simple means of illus-
trating the ideas expressed in the text. The important aspect of graphical presen-
tation has been the province of Enrico Savazzi, whose new language Graph
Server (GS) for displaying plots forms an integral part of the enterprise.
We have not provided more than a few introductory notes on the elements of
matrix algebra, deeming that the basic manipulations required for being able to
use the primer can be most effectively introduced at the appropriate points in
the text. References to introductory manuals presenting the elements of linear
algebra are provided wherever we have thought it necessary.
The idea for writing this introduction for geologists stems from more than 35 years
of experience of R A R in teaching statistics to geologists and biologists in many parts
of the world. This accumulated experience has convinced us that no matter how well
people seem to be grasping a course in multivariate applications, and despite a maxi-
mum of teaching effort, the number of the participants who will really stay with the
subject is small indeed. This is no reflection on the value of the discipline, but rather
an indication of the difficulty experienced by the tyro in understanding what can
be asked of statistical methods and how powerful multivariate methods are when
applied in an appropriate manner. Frustrations occasioned by the incorrect use
of techniques, including the "loyalty syndrome" with respect to a particular one (e.g.
Correspondence Analysis among francophones) is a major source of disaffection.
The main reason for this unfortunate situation is that when the average student
has been cast out to swim on his own he will drown unless he h a s a lifebuoy with
him. Only time can tell if this modest text is that lifesaver.
We wish to make it clear that the primer is not concerned with multivariate
modelling of geological processes, regionalized variables, etc. Cluster-analysis as
a specially delimited topic is likewise not taken up, notwithstanding that some
of the techniques that have come to be associated with the concept of "clustering"
are made use o f - f o r example, the minimum spanning tree, similarity coefficients
and Q-mode latent roots and vectors. (An easy introduction to clustering analysis
is available in the book by Everitt (1974).) It is solely concerned with the simple
application of standard methods of multivariate statistical analysis to geological
data in the form of arrays of measurements and compositional data (e.g. chemical
Preface ix