THE USE OF STATISTICS I N PHYTOSOCIOLOQY
65
pertaining to each site and, secondly, all other properties of the sites
which could conceivably affect the growth of plants. For the general
ecological picture, where plantlhabitat relationships are ultimately
sought, the choice before us now is either to examine both sets of
properties simultaneously in a single operation, or first to analyse one
set alone and use it as a reference system to which the other can then
be related. From mere considerations of economy in the number of
analyses involved, the first would seem the more efficient operation;
but, apart from the fact that computing-time often increases disproportionately with any increase in the number of attributes handled
together, there are both theoretical and practical objections to this
course.
In the f i s t place, the features of the habitat which could be recorded
are almost infinite; to select only those features regarded on a priori
grounds as relevant immediately introduces a subjective bias based on
preconceptions which may not be applicable to the particular system
under study. Secondly, whereas the plants are visible entities which
can usually be recorded directly in the field with reasonable ease,
environmental features tend to be far more difficult and time-consuming
to assess with any degree of confidence beyond a very superficial level.
Thirdly, it is statistically inconvenient - and therefore inefficient - to
mix two sets of attributes of very diverse nature in a single analysis if
this can be avoided; although this may be necessary in certain statistical
fields (as in a numerical taxonomy, where only diverse information
may be available), a vegetational system offers the opportunity of
dealing separately with a set of attributes measurable in comparable
units for the primary analysis. Lastly - and most important - by
definition our main concern in plant ecology is with plants in relation
to habitats; and this in itself suggests that the plant component of the
plantlsitelhabitat system should be considered first.
But even if habitat features are excluded from the primary analysis,
they still must be eventually considered in a full ecological investigation.
The usual practice in ecological survey is to attempt to record a number
of environmental features at the same time as the plants. However,
apart from the question of bias considered above, this practice is not
necessarily the- most efficient. It has already been indicated earlier
(p. 62) that an analysis of plantlplant relationships by hypothesisgenerating methods can be used to give a lead as to the particular
environmental features most likely to be involved. In certain circumstances, therefore, a preliminary phytosociological survey may be more
rewarding, in terms of total effort, than a more elaborate programme of
recording plants and habitat together.
Investigations into the most appropriate statistical techniques for
65
pertaining to each site and, secondly, all other properties of the sites
which could conceivably affect the growth of plants. For the general
ecological picture, where plantlhabitat relationships are ultimately
sought, the choice before us now is either to examine both sets of
properties simultaneously in a single operation, or first to analyse one
set alone and use it as a reference system to which the other can then
be related. From mere considerations of economy in the number of
analyses involved, the first would seem the more efficient operation;
but, apart from the fact that computing-time often increases disproportionately with any increase in the number of attributes handled
together, there are both theoretical and practical objections to this
course.
In the f i s t place, the features of the habitat which could be recorded
are almost infinite; to select only those features regarded on a priori
grounds as relevant immediately introduces a subjective bias based on
preconceptions which may not be applicable to the particular system
under study. Secondly, whereas the plants are visible entities which
can usually be recorded directly in the field with reasonable ease,
environmental features tend to be far more difficult and time-consuming
to assess with any degree of confidence beyond a very superficial level.
Thirdly, it is statistically inconvenient - and therefore inefficient - to
mix two sets of attributes of very diverse nature in a single analysis if
this can be avoided; although this may be necessary in certain statistical
fields (as in a numerical taxonomy, where only diverse information
may be available), a vegetational system offers the opportunity of
dealing separately with a set of attributes measurable in comparable
units for the primary analysis. Lastly - and most important - by
definition our main concern in plant ecology is with plants in relation
to habitats; and this in itself suggests that the plant component of the
plantlsitelhabitat system should be considered first.
But even if habitat features are excluded from the primary analysis,
they still must be eventually considered in a full ecological investigation.
The usual practice in ecological survey is to attempt to record a number
of environmental features at the same time as the plants. However,
apart from the question of bias considered above, this practice is not
necessarily the- most efficient. It has already been indicated earlier
(p. 62) that an analysis of plantlplant relationships by hypothesisgenerating methods can be used to give a lead as to the particular
environmental features most likely to be involved. In certain circumstances, therefore, a preliminary phytosociological survey may be more
rewarding, in terms of total effort, than a more elaborate programme of
recording plants and habitat together.
Investigations into the most appropriate statistical techniques for