INITIAL Subcommand (PREFSCAL command)

INITIAL defines the initial or starting configuration of the common space for the analysis. When a reduction in dimensionality is specified on the CRITERIA subcommand, a derivation of coordinates in the higher dimensionality is used as starting configuration in the lower dimensionality.

• You can specify one of the five keywords that are listed below.
• You can specify a variable list containing the initial configuration.

CLASSICAL. Classical scaling start. This specification is the default. The rectangular proximity matrix is used to supplement the intra-blocks (values between rows and between columns) of the complete symmetrical MDS matrix by means of the triangular inequality or Spearman distances. When the complete matrix is formed, a classical scaling solution is used as initial configuration.

ROSSCLIFF. Ross-Cliff start. The Ross-Cliff start uses the results of a singular value decomposition on the double centered and squared proximity matrix as the initial values for the row and column objects.

CORRESPONDENCE. Correspondence start. The correspondence start uses the results of a correspondence analysis on the reversed data (similarities instead of dissimilarities) with symmetric normalization of row and column scores. See the topic CORRESPONDENCE for more information.

CENTROIDS(n). Centroids start. PREFSCAL starts by positioning the row objects in the configuration by using an eigenvalue decomposition. Then, the column objects are positioned at the centroid of first choices (or second if n=2 or third if n=3, etc.). The number of choices (n) must be a positive integer between 1 and the number of columns. The default is 1.

RANDOM(n). (Multiple) random start. You can specify the number of random starts (n), where n is any positive integer. The random sequence can be controlled by the SET SEED procedure (thus, not by a subcommand within the PREFSCAL procedure). All n analyses start with a different random configuration. In the output, all n final Stress values are reported, as well as the initial seeds of each analysis (for reproduction purposes), followed by the full output of the analysis with the lowest penalized Stress value. The default number of random starts is 1.

CLASSICAL Keyword

TRIANGLE. Imputation using the triangle inequality. If TRIANGLE is specified, the intra-blocks are filled by using the triangular inequality.

SPEARMAN. Imputation with Spearman distances. If SPEARMAN is specified, the Spearman distances between all objects are used to create a symmetrical MDS matrix.

Instead of these keywords, a filespec in parentheses can be given to specify the IBM® SPSS® Statistics data file containing the coordinates of the initial configuration. The row and column coordinates are stacked, with the column coordinates following the row coordinates. The closing parenthesis of the filespec can be followed by a variable list. If the variable list is omitted, the procedure automatically selects the first MAXDIM variables in the external file, where MAXDIM is the maximum number of dimensions that are requested for the analysis on /CRITERIA = DIMENSIONS(min, max). Missing values are not allowed as initial coordinates. An error is issued whenever this situation occurs.

Example

PREFSCAL var01 TO var15
  /INITIAL = RANDOM(100).

• This example performs 100 analyses (each analysis starting with a different random configuration). The results of the analysis with the lowest final Stress are displayed in the output.