Out in the woods we have discovered species A-F and have had different ideas about how to classify them. A is a circumboreal species, while B-E are known only from the Himalayas and adjacent mountain ranges in southeastern China and Vietnam. E is known only from the type collection, although it has been described under two different names. F is only discovered when you investigate the genus to which these species belong in the mountains of Taiwan. This is how this section of the genus appears to us. Run your mouse over the image to see how it might look in the eyes of an omniscient deity.
This graphic is meant to suggest the problem that taxonomists encounter when they set out to do the taxonomy of a group. The organisms they seek to study appear as if they could be classified in a number of different ways, even though they have only a single true set of historical relationships. How then does a taxonomist proceed in order to arrive at a classification that reflects as much as possible that single truth?
of the study group
Note the importance of casting one's net wider than the immediate group of know group members, so as to ensure that one includes all of the "real" group (cf. species F in the example above).
Choice of representatives for the study group
A single individual is used to represent each of the entities in the group (e.g. to represent A-F in the example above).
The entities in the group are represented by a vector of character averages or modes. In this case the sampling from which these summaries are derived can be very important. Where specimens are lacking, and very difficult to come by, it may be necessary to make recourse to original descriptions or illustrations.
The entities in the group are represented by samples of specimens and other material linked to those specimens. Samples may represent local populations from part or all of the range of the entity in question. Sampling intensity will vary according to the cost of obtaining specimens and other samples, and the cost of collecting and analyzing data (below).
characters and their states, if appropriate Covariation with other variables Relevance
Accessibility, commonness Logical correlation Redundant states
To be useful, characters should vary most between entities in the study group, and least within those entities. The following issues are also worth considering:
This is often what data analyses attempt to reveal, i.e. whether the characters covary with binary (multistate) descriptors of membership in groups of interest, e.g. taxa, vegetation units, or experimental blocks, treatments, etc.
Where prior knowledge is available, it may be possible to exclude some potential variables if they likely have little to do with the phenomena under investigation. In a taxonomic study of variation in leaf morphology it might be of limited relevance to measure leaf thickness if this varies appreciably between sun and shade leaves and it is impossible to tell which kind of leaves are present in your sample.
Variables should be easy to measure unambiguously. Measuring above-ground ramets of a rhizomatous (or otherwise clonal) plant is unlikely to be a good way of assessing variation between individual genotypes.
Variables should be measurable on all study objects at a given hierarchical level (i.e. at the most inclusive level, only the (non-) occurrence of a rare character state can be scored; the rare state itself can be studied only within the subset of study objects where it occurs).
Avoid measuring variables that are logically correlated. In a crude description of leaf shape, leaf length above the widest point, leaf length below the widest point, and total leaf length are logically correlated. Only two of these measurements are needed. However, the leaf lengths above and below the widest point, and the leaf maximum width, are likely to covary in characteristic ways that may be of taxonomic or other interest; this covariation (cf. above) is not the same thing as logical correlation.
If variation in one variable resembles that in another, analysis of both may be unnecessary since either one can predict the other (redundancy is symmetrical). For example, if narrow leaves are found only in individuals whose flowers have 10, rather than 20, stamens then measuring leaf width as well as stamen number is likely to be redundant.
Covariation with other variables
of character variation
This will be discussed in subsequent lectures.
Assessment of relationships
Construct hierarchy of relationships
Put together the practical apparatus (monograph; descriptions, keys)
innovation versus acceptance of existing results
splitting versus lumping
analytical intensity and thoroughness versus superficiality
synchronic versus diachronic
Davis, P. H. & V. H. Heywood (1965). Principles of Angiosperm Taxonomy. Edinburgh, Oliver & Boyd.
A lecture by Prof. J. E. Eckenwalder in March 1992, and an ongoing collaboration with Prof. S. P. Vander Kloet (Acadia University).
|BOT300S Home Page | U of T Botany | University of Toronto |
© 2003 Botany Department, University of Toronto.Please send your comments to firstname.lastname@example.org; last updated 23-Mar-2003