Hi everybody. After applying a two-step cluster in SPSS, involving both continuous and nominal variables, how can I validate if the results are satisfactory? In other clustering methods there are some procedures that support the results of the cluster analysis, namely the Anova analysis to evaluate the significance of variables to classify the clusters. How do you validade a two-step clustering?
After debating myself with this question. Here is what appears to be the best method to validate a two step cluster solution:
You can validate two step cluster solution following the methodology suggested by Norusis (2008). First, model fit must be evaluated by the silhouette coefficient, which is a measure of cohesion and separation of clusters. The obtained value must be above 0.0, suggesting validity of the within- and between-cluster distances. Second, Pearson’s chi-squared tests or independent samples T-tests for the categorical and continuous variables respectively, can be applied to confirm that the clusters varie significantly across the segmentation variables. Finally, the cluster solution can be assessed by splitting the dataset into random halves and running the same statistical procedures (two step clusters for each halve). The obtained results for each halved sample must be similar in terms of the number and characteristics of the clusters, silhouette measures of cohesion and sepration and predictors importance.
Norusis, M. J. (2008). SPSS 16.0 guide to data analysis (2nd ed.). Upper Saddlr River: Prentice Hall.
Dear Nuno Batista
Maybe, after you finished two-step cluster analysis via SPSS, the result table will be created and some indexes will be known. Your result is good. SO you just need to prove that you are right. ANOVA analysis may work well. Moreover, in my opinion (one of my phd idea), if you have many kinds of good data, you can also do some regression analysis, canonical correlation analysis, factor analysis, path analysis etc. Anyway it is all based on your purpose.
Hope this can help you.
Best wishes.
Than you for your answer. The results table in SPSS refers to an index named "silhuete measure of coehesion and separation" that is cited in some studies.
Dear Batista this link will help you:
https://www.spss.ch/upload/1122644952_The%20SPSS%20TwoStep%20Cluster%20Component.pdf
After debating myself with this question. Here is what appears to be the best method to validate a two step cluster solution:
You can validate two step cluster solution following the methodology suggested by Norusis (2008). First, model fit must be evaluated by the silhouette coefficient, which is a measure of cohesion and separation of clusters. The obtained value must be above 0.0, suggesting validity of the within- and between-cluster distances. Second, Pearson’s chi-squared tests or independent samples T-tests for the categorical and continuous variables respectively, can be applied to confirm that the clusters varie significantly across the segmentation variables. Finally, the cluster solution can be assessed by splitting the dataset into random halves and running the same statistical procedures (two step clusters for each halve). The obtained results for each halved sample must be similar in terms of the number and characteristics of the clusters, silhouette measures of cohesion and sepration and predictors importance.
Norusis, M. J. (2008). SPSS 16.0 guide to data analysis (2nd ed.). Upper Saddlr River: Prentice Hall.
Hi, Nuno.
You mentioned that silhouette coefficient is a measure of cohesion and separation of two-steps clusters. Do you know what should I use for hierarchical clusters?
The cross-validation procedure with the discriminant analysis is a best solution to overcome this problem. It is like what you wrote your last answer. But there are also steps to be run in addition to what you write
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA