-------------------------------------------------------------------------------
help for hcaccprox
-------------------------------------------------------------------------------

Hierarchical Clusters Analysis with conditional proximity measures

        hcaccprox varlist [, prox(keyword) method(keyword) partition(numlist)
              measures details detect(#) ]

        partition numlist

Description

    hcaccprox realize a Hierarchical Clusters Analysis on dichotomoux items
        based on specific measures of proximity as conditional proximity
        measures. The program permit to obtain indexes to test the obtained
        partition (the detect program is necessary in this case).

    partition permit, after a hcaccprox step to obtain the composition of
        some specific partitions of the items.

Options

    prox:(keyword) define the method to compute the proximity between the
        items.  Six measures are possible. The three first ones are
        unconditional measures named a, ad and cor.  The three last ones are
        conditional measures named ccov, ccor and mh. See Roussos, Stout and
        Marden (1998) for details of these six measures. By default, the ccov
        option is used.

    method define the method to aggregate two clusters, single for a single
        linkage, complete for a complete linkage, and UPGMA for the
        Unweighted Pair-Group Method of Average. By default, the UPGMA option
        is used.

    partition(numlist) lists the partitions to detail by the program. List
        like (2 4 6) or (2(2)6) are authorized.

    measures display the used proximity measures between the items.

    details display the results of the algorithm of aggregation.

    detect(#) specifies for all the partitions with a number of clusters
        inferior or equal to # to compute the DETECT, Iss and R indexes.

    numlist, for the partition program, define the partitions with the number
        of clusters indicated in the numlist to detail.

Examples

    . hcaccprox q1-q10

    . partition 3 5 6

    . hcaccprox item1-item9 dotest1-dotest6, detect(6) measures

    . hcaccprox c1 c2 c3 c4 c5 c6 c7, prox(a) method(single)

Outputs

    . r(varlist) is a macro who contain varlist

    . r(nbitems) is a macro who contain the number of items

    . r(nodes) is a matrix who contain all the informations about all the
        possible clusters of items. Each column represent a node (the first
        ones represent each item of varlist, and the following columns
        represent each aggregation of clusters), the first line represent the
        number of items in each cluster, the third and the fourth lines
        represent the two cluster who are aggregated to form the new cluster,
        and the following lines represent the list of items composing each
        cluster

    . r(mempart) list the number of cluster composing each possible partition
        : the last column is the partition in only one cluster, the
        preceeding column represent the partition in two cluster, and so on

    . r(affect#) is obtained with the partition option. In this vector, the
        number of the cluster (of the partition in # clusters) is associated
        to each item

    . r(indexes) is obtained with the detect option. This matrix contain the
        DETECT, Iss and R indexes associated to each partition with a number
        of clusters inferior to the number defined in the detect option


Reference

    Roussos L. A, Stout W. F. and Marden J. I., Using new proximity measures
        with hierarchical cluster analysis to detect multidimensionality.
        Journal of Educational Measurement, 35(1), pp 1-30, 1998.

    Zhang J. and Stout W. F., The theorical DETECT index of dimensionality
        and its application to approximate simple structure. Psychometrika,
        64(2), pp 213-249, 1999.

Also see

    help for detect

Author

    Jean-Benoit Hardouin, Regional Health Observatory (ORS) - 1, rue Porte
        Madeleine - BP 2439 - 45032 Orleans Cedex 1 - France.  You can
        contact the author at jean-benoit.hardouin@neuf.fr and visit the