THE USE O F STATISTICS I N PHYTOSOCIOLOQY
69
The choice between using simpler quantitative measures and purely
qualitative records rests on more subtle considerations. It seems rarely
recognized in phytosociological work that, whereas presence-orabsence records form a self-contained logical system, all quantitative
data are truncated in that no record is possible except zero for the
absence of a species : although the amount by which a species is present
can be recorded, there is no corresponding measure of the extent to
which it is absent. The quantitative pattern is thus superimposed on an
underlying qualitative pattern and, even with sites from roughly similar
vegetation, the qualitative differences may easily override the quantitative element in information-content. We would therefore dissent from
the views of Greig-Smith (1964, p. 160) when he states: “Even if the
arbitrary simplification is made of considering presence and absence of
species only, the important difference between stands lies in the amount
of different species. . . . With only slight exaggeration we may, within
the limits of a set of broadly similar stands, regard absence as simply
the extreme value of a continuous variable.” In fact, a method now
exists for assessing the relative importance of qualitative and quantitative information in a given set of data; this is the partition correlation
analysis of Williams and Dale (1962). So far, only one set of quantitative data, based on a crude frequency measure, has been tested by this
method (Dale, unpublished), but the results are very suggestive and
worth quoting briefly here. For twenty-eight species in fifteen rather
similar heathland sites, the information-content (as measured by the
sums of squares of the partitioned correlations) of the purely qualitative
element was more than double that of the purely quantitative element,
and also substantially exceeded that of both qualitative/quantitative
and purely quantitative elements added together.
Although quantitative differences may be important in detailed
studies of pattern involving relatively few species, in large-scale phytosociological work any increase in information gained by using quantitative instead of qualitative methods is thus likely to be extremely small.
It is common experience that the recording of quantitative measures
is vastly more time-consuming than mere presence-or-absence recording ; moreover, the use of quantitative instead of qualitative data in any
subsequent statistical analysis may well increase computing-time by as
much as a hundredfold. In general, therefore, there is much to recommend the adoption of simple floristic recording as the “standard”
method of data-collecting on grounds of general efficiency.
69
The choice between using simpler quantitative measures and purely
qualitative records rests on more subtle considerations. It seems rarely
recognized in phytosociological work that, whereas presence-orabsence records form a self-contained logical system, all quantitative
data are truncated in that no record is possible except zero for the
absence of a species : although the amount by which a species is present
can be recorded, there is no corresponding measure of the extent to
which it is absent. The quantitative pattern is thus superimposed on an
underlying qualitative pattern and, even with sites from roughly similar
vegetation, the qualitative differences may easily override the quantitative element in information-content. We would therefore dissent from
the views of Greig-Smith (1964, p. 160) when he states: “Even if the
arbitrary simplification is made of considering presence and absence of
species only, the important difference between stands lies in the amount
of different species. . . . With only slight exaggeration we may, within
the limits of a set of broadly similar stands, regard absence as simply
the extreme value of a continuous variable.” In fact, a method now
exists for assessing the relative importance of qualitative and quantitative information in a given set of data; this is the partition correlation
analysis of Williams and Dale (1962). So far, only one set of quantitative data, based on a crude frequency measure, has been tested by this
method (Dale, unpublished), but the results are very suggestive and
worth quoting briefly here. For twenty-eight species in fifteen rather
similar heathland sites, the information-content (as measured by the
sums of squares of the partitioned correlations) of the purely qualitative
element was more than double that of the purely quantitative element,
and also substantially exceeded that of both qualitative/quantitative
and purely quantitative elements added together.
Although quantitative differences may be important in detailed
studies of pattern involving relatively few species, in large-scale phytosociological work any increase in information gained by using quantitative instead of qualitative methods is thus likely to be extremely small.
It is common experience that the recording of quantitative measures
is vastly more time-consuming than mere presence-or-absence recording ; moreover, the use of quantitative instead of qualitative data in any
subsequent statistical analysis may well increase computing-time by as
much as a hundredfold. In general, therefore, there is much to recommend the adoption of simple floristic recording as the “standard”
method of data-collecting on grounds of general efficiency.