76
J. M. LAMBERT AND M . B . DALE
that the number of axes to be used is also determined by the
investigator.
One of the more revealing features of much ordination work is that
many of the proponents of this approach use data selected on the basis
of an earlier subjective classification before ordination is attempted (cf.,
for instance, the use of stands of the same vegetation type by Ssrensen
(1948) and Curtis (1959)). We have already been at some pains to
emphasize that the more powerful ordination methods (factor analysis)
are only reasonably successful when they are applied to data which
have a number of species in common between the sites: although
component analysis can be used in any situation to explore the degree
of heterogeneity of the data, the usefulness of factor analysis is limited
when applied to qualitatively different sites. In investigations concerned essentially with the overall picture of the vegetation of an area,
therefore, the relative efficiency of ordination and classification must
now be looked at in this light.
It is unfortunate that much of the discussion concerning the applicability of the one or the other method has frequently been confounded
by the erection of false antitheses. For instance, Goodall (1954a) discusses the “inadequacies of classification” and considers that “ordination can hardly fail to be more precise than a classificatory system”.
I n fact, both ordination and classification are equally precise as methods.
In either case, as many constructs - axes or classes - are needed to
accommodate the data completely as there are non-identical sites; in
either case, the primary aim is to simplify a complex situation by
reduction in the number of axes or classes to be assessed; and, in either
case, the degree of simplification eventually adopted depends on
criteria imposed by the user and not on the method itself.?
A general assessment of the two approaches has recently been made
by Greig-Smith (1964, Chap. 7), who comes to the conclusion (p. 161)
that “unless . . . there are strong reasons for preferring classification in
a particular case, ordination is the sounder initial procedure even though
the results may conceivably indicate that a classification is the best
means of summarizing the information finally”. Bearing in mind the
requirement for overall efficiency, we shall here take the opposite point
of view, at least as far as primary survey is concerned: if most vegetation is inherently heterogeneous (as distinct from discontinuous) in
ff oristic composition, if ordination techniques are computationally
t In his most recent paper (cf. footnote, p. 72), Goodall (1963), p. 303, further claims
that ordination is “more informative” than classification, and contrasts a classification
of four towns i n t e r n of their land-mass relationships with an ordination of the towns
on latitudinal and longitudinal axes. However, the example is not valid, since different
criteria are used i n the two caaes, so that the two processes give different information
and are not comparable.
J. M. LAMBERT AND M . B . DALE
that the number of axes to be used is also determined by the
investigator.
One of the more revealing features of much ordination work is that
many of the proponents of this approach use data selected on the basis
of an earlier subjective classification before ordination is attempted (cf.,
for instance, the use of stands of the same vegetation type by Ssrensen
(1948) and Curtis (1959)). We have already been at some pains to
emphasize that the more powerful ordination methods (factor analysis)
are only reasonably successful when they are applied to data which
have a number of species in common between the sites: although
component analysis can be used in any situation to explore the degree
of heterogeneity of the data, the usefulness of factor analysis is limited
when applied to qualitatively different sites. In investigations concerned essentially with the overall picture of the vegetation of an area,
therefore, the relative efficiency of ordination and classification must
now be looked at in this light.
It is unfortunate that much of the discussion concerning the applicability of the one or the other method has frequently been confounded
by the erection of false antitheses. For instance, Goodall (1954a) discusses the “inadequacies of classification” and considers that “ordination can hardly fail to be more precise than a classificatory system”.
I n fact, both ordination and classification are equally precise as methods.
In either case, as many constructs - axes or classes - are needed to
accommodate the data completely as there are non-identical sites; in
either case, the primary aim is to simplify a complex situation by
reduction in the number of axes or classes to be assessed; and, in either
case, the degree of simplification eventually adopted depends on
criteria imposed by the user and not on the method itself.?
A general assessment of the two approaches has recently been made
by Greig-Smith (1964, Chap. 7), who comes to the conclusion (p. 161)
that “unless . . . there are strong reasons for preferring classification in
a particular case, ordination is the sounder initial procedure even though
the results may conceivably indicate that a classification is the best
means of summarizing the information finally”. Bearing in mind the
requirement for overall efficiency, we shall here take the opposite point
of view, at least as far as primary survey is concerned: if most vegetation is inherently heterogeneous (as distinct from discontinuous) in
ff oristic composition, if ordination techniques are computationally
t In his most recent paper (cf. footnote, p. 72), Goodall (1963), p. 303, further claims
that ordination is “more informative” than classification, and contrasts a classification
of four towns i n t e r n of their land-mass relationships with an ordination of the towns
on latitudinal and longitudinal axes. However, the example is not valid, since different
criteria are used i n the two caaes, so that the two processes give different information
and are not comparable.