Key points are not available for this paper at this time.
Types are modes in a multidimensional distribution, and can be found either by operating on a Q ‐matrix of distances between individuals or by calculating densities in Cartesian sub‐spaces. Pursuing the former approach, reasons are given for preferring the family of r p ‘coefficients of likeness’ to Mahalanobis' D or other distance measures. A general formula is presented for r p , applicable to factors of any obliqueness and any assigned weight, permitting classification by what an individual docs as well as by what he is. Classification procedures require recognition of the differences between types as homostats and as segregates. A homostat is a set of unusually similar people: a segregate is a series with above average similarities which thus has continuity, although there may be negligible resemblance between first and last members. A Boolean algorithm will objectively find homostats and sort them into phenomenal and nuclear clusters, and a further algorithm will operate on these to locate segregates. These steps have been combined in a computer program called Taxonome, two examples of the application of which are given. Search for types is a topographical descriptive procedure in which finding the ‘texture’ of a whole domain is as important as the location of individual types. Type analysis contributes to non‐linear prediction, to theories of type origin, and to higher order systematization and conceptualization.
Cattell et al. (Tue,) studied this question.