Key points are not available for this paper at this time.
Summary A standard method of comparing classifications is the “cophenetic correlation” technique of Sokal and Rohlf (1962). The method is re‐examined, and is shown to possess two shortcomings which make it unsuitable for the truncated, intensely‐clustering strategies now in common use. These shortcomings are (i) it is unable to distinguish between changes due to group composition and changes due to configuration of groups within the hierarchy, and (ii) it is sensitive to the convention used for representing the hierarchy in two dimensions. A revised form of the strategy is proposed which appears to overcome these difficulties.
Williams et al. (1971) studied this question.