SPSS Statistics

 View Only

  • 1.  Differences in Silhouette Score Calculation

    Posted Thu December 01, 2022 09:19 AM

    Hello I have a question about the two-step output or the silhouette plot.

    When I run a Two-Step analysis, I get the following output and an average silhouette score of 0.7.

    However, if I try to calculate the score exactly using the SIL Extension, I get a different score. I used exactly the same variables as the Two-Step and then get this output.


    According to my calculation, (14*0.592+18*0.508)/32=0.54475 and not 0.7. Where is the difference coming from?



    ------------------------------
    Lorenz Wagner
    ------------------------------

    #SPSSStatistics


  • 2.  RE: Differences in Silhouette Score Calculation

    IBM Champion
    Posted Thu December 01, 2022 10:30 AM
    The STATS CLUSTER SIL extension command does not necessarily compute the silhouette values the same way as two step does.  If all the variables are continuous, the measure should be the same as for two step (possibly apart from outlier handling), but if there is a mixture of types, Gower would be the closest measure, but it might not be quite the same.

    --



    Original Message


  • 3.  RE: Differences in Silhouette Score Calculation

    IBM Champion
    Posted Thu December 01, 2022 10:37 AM
    p.s., The extension command uses the measurement level declared for the variables, so make sure that matches the designations used in the two-step command.

    On Thu, Dec 1, 2022 at 8:28 AM Jon Peck <jkpeck@gmail.com> wrote:
    The STATS CLUSTER SIL extension command does not necessarily compute the silhouette values the same way as two step does.  If all the variables are continuous, the measure should be the same as for two step (possibly apart from outlier handling), but if there is a mixture of types, Gower would be the closest measure, but it might not be quite the same.

    --


    --



    Original Message