The genus Lomandra was chosen for a numerical taxonomic study because of its dioecious condition. This enabled classifications to be made separately on the basis of male and female plants as well as on other character combinations. The inclusion of anatomical data in the analyses supplemented traditional morphological data and revealed several unsuspected taxonomic problems.
Thirty-eight taxa of Lomandra are found in eastern Australia, all of which had been included in the present study. Besides the morphological and anatomical studies undertaken for the numerical work, the investigation has been extended to include pollination experiments and the study of environmental plasticity in the genus to determine whether different genotypes differ in their capacity to react to variations in the environment. The pollination experiments have revealed that some Lomandra species may be agamospermic. The plasticity-experiments have shown that the Lomandra plants tested are quite stable (both in vegetative and reproductive characters) and do not respond markedly to slight variations in environmental conditions. Because of the difficulties encountered in germinating Lomandra seeds breeding experiments have played only a minor role in this investigation.
Although Lomandra taxa are mostly strictly dioecious, irregularities in sex expression have been met with for a few species.
For each of 38 taxa seventy eight attributes were scored using both morphological and anatomical structures. These comparative data were subjected to computer analyses using three classificatory programs, namely. Information Gain Statistic (MULTBET), Incremental Sum of Squares (I.S.S.) and Mean Squared Distance (M.S.D.). All three of these numerical analyses indicated the existence of three homogeneous groups within the eastern Australian members of the genus Lomandra (the leucocephala-, glauca- and multiflora-groups). The remaining taxa are less homogeneous. Both the I.S.S. and MULTBET programs suggested they comprised two groups only whereas the M.S.D. program suggested them to constitute three or four groups.
Of the three programs employed, it was recognized that the classification generated by the I.S.S. program using the total data is almost identical to that produced by the MULTBET program. Hence this clustering technique was ignored for further comparisons based upon subsets of the total data. It has been observed that the form of clustering produced by the remaining two programs (MULTBET and M.S.D.) gives a somewhat different form of classification, though there are considerable areas of agreement in that three of the groups recognized are common to both classifications; a fact which indicates that the computed hierarchy genuinely reflects correlations amongst certain sections of the data, although the M.S.D. program gave a much less intense form of clustering than MULTBET. A set of 6 analyses were further made on the basis of different subsets of attributes using the MULTBET and M.S.D. programs separately. When the MULTBET program was used the agreement between the classification produced on the basis of total data and that produced by the total reproductive attributes, or the male reproductive attributes only is excellent. This is in contrast to the results generated by the M.S.D. program where in no instance were the results based on pairs of attribute-sets identical and the resulting classification showed varying degrees of difference.
The MULTBET classification was considered to be better than the M.S.D. classification (p. 96). It also arranges the species into useful taxonomic groups with members possessing similar geographical distributions. The breeding results are found to be in accordance with the groups produced by the MULTBET program. The MULTBET analysis has also arranged the species in groups which accord with the similarities in the distribution of chromosome numbers. Furthermore, the MULTBET classification using all attributes was found to agree well with those classifications produced by previous students of the genus.
The considerable differences observed between classifications based upon both different sets of attributes and different clustering strategies indicates that although numerical taxonomic techniques are fully objective the answers obtained by their use are very dependent upon the data employed and the manner in which it is treated. In all such studies it is essential that the data and methodology be fully displayed to be of use to other workers.