Measures

The algorithm converges to a solution after 154 iterations, with a penalized stress (marked final function value) of 0.0000990. Since the penalty term has been turned off, penalized stress is equal to Kruskal's Stress-I (the stress part of the function value is equivalent to Kruskal's badness-of-fit measure). Low stress values generally indicate that the solution fits the data well, but there are several warning signs of a degenerate solution:

• The coefficient of variation for the transformed proximities is very small relative to the coefficient of variation for the original proximities. This suggests that the transformed proximities for each row are near-constant, and thus the solution will not provide any discrimination between objects.
• The sum-of-squares of DeSarbo's intermixedness indices are a measure of how well the points of the different sets are intermixed. If they are not intermixed, this is a warning sign that the solution may be degenerate. The closer to 0, the more intermixed the solution. The reported value is very large, indicating that the solution is not intermixed.
• Shepard's rough nondegeneracy index, which is reported as a percentage of different distances, is equal to 0. This is a clear numerical indication that there are insufficiently different distances and that the solution is probably degenerate.

Next