CRITERIA Subcommand (PROXSCAL command)

Use CRITERIA to set the dimensionality and criteria for terminating the algorithm, or minimization process. You can specify one or more of the following keywords:

DIMENSIONS. Minimum and maximum number of dimensions. By default, PROXSCAL computes a solution in two dimensions (min=2 and max=2). The minimum and maximum number of dimensions can be any integers inclusively between 1 and the number of objects minus 1, as long as the minimum is less than or equal to the maximum. PROXSCAL starts computing a solution in the largest dimensionality and reduces the dimensionality in steps, until the lowest dimensionality is reached. Specifying a single value represents both minimum and maximum number of dimensions, thus DIMENSIONS(4) is equivalent to DIMENSIONS(4,4).

MAXITER. Maximum number of iterations. By default, n=100, specifying the maximum number of iterations that is performed while one of the convergence criterion below (CONVERGENCE and STRESSMIN) is not yet reached. Decreasing this number might give less accurate results but will take less time. N must have a positive integer value.

DIFFSTRESS. Convergence criterion. PROXSCAL minimizes the goodness-of-fit index normalized raw stress. By default, PROXSCAL stops iterating when the difference in consecutive stress values is less than 0.0001 (n=0.0001). To obtain a more precise solution, you can specify a smaller value. The value specified must lie between 0.0 and 1.0, inclusively.

MINSTRESS. Minimum stress value. By default, PROXSCAL stops iterating when the stress value itself is small, that is, less than 0.0001 (n=0.0001). To obtain an even more precise solution, you can specify a smaller value. The value specified must lie between 0.0 and 1.0, inclusively.

Example 

PROXSCAL VARIABLES=object01 TO object24 
  /CRITERIA=DIMENSIONS(2,4) MAXITER(200) DIFFSTRESS(0.00001).
  • The maximum number of dimensions equals 4 and the minimum number of dimensions equals 2. PROXSCAL computes a four-, three-, and two-dimensional solution, respectively.
  • The maximum number of iteration is raised to 200.
  • The convergence criterion is sharpened to 0.00001.