THE USE OF STATISTICS I N PHYTOSOCIOLOGY
77
cumbersome, and if - as Greig-Smith implies - classification is the
most generally useful method of eventually docketing the information,
there is much to be said for proceeding directly to a system of classification. Moreover, it need not be assumed that any advantages of the
more informative methods of ordination will automatically be lost if
classification is used, since similar statistical operations are frequently
involved in the better classificatory techniques; in fact, the choice
between ordination and classification as a primary approach rests
mainly on convenience, both in the performance of the analyses and in
the use of the results.
c. METHODS O F CLASSIFICATION
The process of classification is used in so many fields that it is not
surprising that the study of statistical classificatory techniques -
“numerical taxonomy’’ - is considerably in advance of that of pure
ordination methods. The basic problems associated with the manufacture of such techniques have been currently reviewed by Williams
and Dale (1964) in a complementary paper to this contribution: we shall
here only outline the main points a t issue as far as the user is concerned.
The first distinction to be made is that between techniques which are
aimed at allotting individuals to existing classes and those concerned
with the actual erection of the classes themselves from information in
the data. The former type of method, known as “discriminant analysis”,
is much used in situations where the systematic framework is already
widely recognized and accepted : the sole function of statistics then is to
facilitate decisions as to the best placing of new individuals. At present,
however, we are more concerned with methods which aim to extract the
maximum information from the system actually under study, rather
than with attempts to accommodate the data within classes already
erected on external criteria; the justification for this as a primary
approach to phytosociological situations will be dealt with more fully
in a subsequent section (IVA). Methods of discriminant analysis will
therefore not be considered further in the present discussion, which
deals essentially with possible methods of primary analysis.
Among those ecologists who favour a classificatory approach, the
common concept of vegetation as “a multidimensional network” has
led many to suggest that some form of non-hierarchical classification
provides the most effective means of representing such variation. For
instance, one finds in the literature statements such as: “It became clear
in the course of field study that, although communities could be forced
into hierarchical groups, the true relationships of communities were
multidimensional” (Poore, 1955~). Such views, however, stem largely
77
cumbersome, and if - as Greig-Smith implies - classification is the
most generally useful method of eventually docketing the information,
there is much to be said for proceeding directly to a system of classification. Moreover, it need not be assumed that any advantages of the
more informative methods of ordination will automatically be lost if
classification is used, since similar statistical operations are frequently
involved in the better classificatory techniques; in fact, the choice
between ordination and classification as a primary approach rests
mainly on convenience, both in the performance of the analyses and in
the use of the results.
c. METHODS O F CLASSIFICATION
The process of classification is used in so many fields that it is not
surprising that the study of statistical classificatory techniques -
“numerical taxonomy’’ - is considerably in advance of that of pure
ordination methods. The basic problems associated with the manufacture of such techniques have been currently reviewed by Williams
and Dale (1964) in a complementary paper to this contribution: we shall
here only outline the main points a t issue as far as the user is concerned.
The first distinction to be made is that between techniques which are
aimed at allotting individuals to existing classes and those concerned
with the actual erection of the classes themselves from information in
the data. The former type of method, known as “discriminant analysis”,
is much used in situations where the systematic framework is already
widely recognized and accepted : the sole function of statistics then is to
facilitate decisions as to the best placing of new individuals. At present,
however, we are more concerned with methods which aim to extract the
maximum information from the system actually under study, rather
than with attempts to accommodate the data within classes already
erected on external criteria; the justification for this as a primary
approach to phytosociological situations will be dealt with more fully
in a subsequent section (IVA). Methods of discriminant analysis will
therefore not be considered further in the present discussion, which
deals essentially with possible methods of primary analysis.
Among those ecologists who favour a classificatory approach, the
common concept of vegetation as “a multidimensional network” has
led many to suggest that some form of non-hierarchical classification
provides the most effective means of representing such variation. For
instance, one finds in the literature statements such as: “It became clear
in the course of field study that, although communities could be forced
into hierarchical groups, the true relationships of communities were
multidimensional” (Poore, 1955~). Such views, however, stem largely