A unifying theory for methods of systematic analysis

A unifying theory for systematic analysis states that a number of methods should be used jointly to cope with various kinds of data; also that groups should be as consistent as possible, be made with least information loss, and where needed, be polythetic. A test of relationship, homogeneity, can use various kinds of data. It can take account of the internal variation of aggregate items such as genera. It can give due emphasis to smaller clusters that have likely important contexts of external items. It helps in analysing trends, cores and hazes in dendrograms. A proposed detector for formal groups can be based on measures of isolation, identifiability and inclusiveness. Non-mathematical, inter-item reaction tests such as hybridization and serology can also be used in grouping. All relationship data are used polythetically to reveal natural groups. This leads to a unified informational concept for taxa. This is more useful than the biological species concept that is restricted to inter-breeding data. All the methods appear to be analogues of the powerful human grouping instinct. The resulting compatibility is important as precise methods are needed mainly when the data are too complex for the mind to use reliably. Cladograms can be made by self-graded deweighting of homogeneity and agglomerative clustering. Unlike classical cladistics this can reveal any polythetic group. Finding the derived states for making cladograms is often much too hypothetical for a fully cladistic approach to be properly precise. Instead, where the evidence is weak, a milder strength of graded deweighting is used for the cladistic properties, which help to show relationships along with the others. Axiomatic failures of other classes of grouping methods are discussed. Unavoidable remnants of instinctive processing lower the precision of all the methods. The Uniter computer program, based on the theory, is tested with finely graded values of artificially ‘evolved’ items and with coarsely coded cladistic data. The results show that with natural data, the program should act as a fairly sensitive probe of past evolutionary branching. Another test shows how specimens from species complexes can be grouped and how distinctions between groups are analysed.