Hello. I want to use cluster analysis in SPSS because I want to discover the structure of my data. So I do not know how many class I want to see on my output. That's why I can not apply k-means clustering algorithm. But I also can't apply hierarchical clustering algorithm because my sample size is really wide. So I started to search about two-step cluster analysis but I couldn't reach the assumptions of two-step cluster analysis. And I wonder can I apply two-step cluster when n>18000?
And If I can apply it, on what base? Can you offer me an academic text about it?
Thank you in advance.
Dear friend.
It is difficult to give a clear answer as we do not know a number of circumstances related to your case.
If possible, it would be appropriate to create segments and then solve them through cluster analysis (e.g. if it is a dataset in which there are 20 national groups, it is advisable to solve them in terms of national affinity - data aggregation). In the case of a large number of variables, it is appropriate to perform EFA.
More information about Two Step CA:
https://www.ibm.com/docs/en/spss-statistics/25.0.0?topic=features-twostep-cluster-analysis
https://arifkamarbafadal.files.wordpress.com/2011/09/ebook-038-tutorial-spss-two-step-cluster-analysis.pdf
Two suggestions
1) find and remove any duplicates (they will always belong to the same cluster) thereby reducing overheads
2 use k mean clustering and go for 5 10 20 40 clusters and plot some goodness of fit statistic against the number of clusters - that should give you some indication of a suitable number of cluster.
On two step clustering there is this
https://www.youtube.com/watch?v=BrmfYtT98W0
I'm not sure that spss will handle that big of a problem. They have the smallest size of any of the major packages. Best wishes David Booth
- Badges
- Science topic
- Similar topics
- Indexes