80
J . M . LAMBERT AND M. B . D.ALE
the relationships between species, however, a number of measures of
“importance” of each species, in either the whole population or in a
particular part of the population, can be calculated. This information
can then be imported into the sites as weighting coefficients, thereby
increasing the informativeness of the individual sites. I n subdivisive
methods, the weights can be recalculated for each subgroup as it is
obtained; agglomerative methods can only use weights calculated over
the whole population throughout the analysis, and this provides a
further reason for preferring subdivisive method as a general rule.
Again, there is the choice between using “self-structuring” and
“transposed-structuring” techniques. The former examine site/site
relationships directly in terms of their species composition ; the latter
employ species/species relationships to generate the groups of sites.
I n addition to the decisions to be made as to the form of the method,
there are also choices within the techniques themselves as to the actual
parameter to be used in calculation. For instance, in self-structuring
methods, where the parameter commonly chosen is some measure of
“similarity”, some parameters involve correction for “double-zero”
matches between species while others accept such matches as informative. Again, in transposed-structuring, which usually employs a “correlation” measure, the relative weight to be given to small values can
be altered by appropriate transformation.
With so many choices available, it is not surprising to find that a
variety of different combinations have been used in practice. Thus
Goodall (1953) devised a basically valuable subdivisive, monothetic,
transposed method but used an inefficient parameter for subdivision ;
the underlying idea was later adopted by Williams and Lambert (1959,
1960), who recast and refined the technique and used it with success
on several test-communities. Examples of the use of other methods in
ecology can be seen in the work of Bellamy (1962), who used a simple
self-structuring agglomerative method for analysing the vegetation of a
number of Polish bogs; and of Harberd (1962), who again used a selfstructuring technique but employed a distance measure derived by a
needlessly circuitous route.
The ultimate decision as to the specific method to be used in a given
situation must depend partly on the relative power of the methods
themselves, and partly on practical considerations. The most efficient
method is that which uses the greatest amount of relevant information
to maximize the intergroup differences (or minimize the intragroup
differences) subject to computational practicability. Thus, although a
polythetic subdivisive method with internal weighting might be
theoretically most desirable, the time involved in actual computation
could easily put the method out of reach for any but the simplest
J . M . LAMBERT AND M. B . D.ALE
the relationships between species, however, a number of measures of
“importance” of each species, in either the whole population or in a
particular part of the population, can be calculated. This information
can then be imported into the sites as weighting coefficients, thereby
increasing the informativeness of the individual sites. I n subdivisive
methods, the weights can be recalculated for each subgroup as it is
obtained; agglomerative methods can only use weights calculated over
the whole population throughout the analysis, and this provides a
further reason for preferring subdivisive method as a general rule.
Again, there is the choice between using “self-structuring” and
“transposed-structuring” techniques. The former examine site/site
relationships directly in terms of their species composition ; the latter
employ species/species relationships to generate the groups of sites.
I n addition to the decisions to be made as to the form of the method,
there are also choices within the techniques themselves as to the actual
parameter to be used in calculation. For instance, in self-structuring
methods, where the parameter commonly chosen is some measure of
“similarity”, some parameters involve correction for “double-zero”
matches between species while others accept such matches as informative. Again, in transposed-structuring, which usually employs a “correlation” measure, the relative weight to be given to small values can
be altered by appropriate transformation.
With so many choices available, it is not surprising to find that a
variety of different combinations have been used in practice. Thus
Goodall (1953) devised a basically valuable subdivisive, monothetic,
transposed method but used an inefficient parameter for subdivision ;
the underlying idea was later adopted by Williams and Lambert (1959,
1960), who recast and refined the technique and used it with success
on several test-communities. Examples of the use of other methods in
ecology can be seen in the work of Bellamy (1962), who used a simple
self-structuring agglomerative method for analysing the vegetation of a
number of Polish bogs; and of Harberd (1962), who again used a selfstructuring technique but employed a distance measure derived by a
needlessly circuitous route.
The ultimate decision as to the specific method to be used in a given
situation must depend partly on the relative power of the methods
themselves, and partly on practical considerations. The most efficient
method is that which uses the greatest amount of relevant information
to maximize the intergroup differences (or minimize the intragroup
differences) subject to computational practicability. Thus, although a
polythetic subdivisive method with internal weighting might be
theoretically most desirable, the time involved in actual computation
could easily put the method out of reach for any but the simplest