Matt,
There is in principle no ‘meaning’ in the number assigned to each cluster. Your finding about the last two clusters is completely accidental.
In the k-means procedure, given initial centroids (given by you or assigned by the procedure itself), the first round assigns points to the nearest centroid (with distance measured in Euclidean fashion), then the centroids are recalculated, and then probably some points change from one cluster to another because they are now closer to the new centroid of a different cluster, and so on until no further changes occur. Cluster No.1 is (I think) the cluster corresponding to the first initial centroid specified in your input matrix of centers.
Clusters, once created by whatever method, may be ordered by an external criterion, i.e. by the intra-cluster means of some variable of interest.
Hector
De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de Matthew Pirritano
Enviado el: Saturday, June 18, 2011 14:18
Para: [hidden email]
Asunto: Is there meaning to the order of clusters in k-means?
All,
I have created 5 clusters on a coping scale. I used ward’s method to create my starting cluster centers and then ran in K-means to create my clusters.
Is there any meaning to the order that clusters are created? In other words there any meaning to why my cluster 1 is cluster one, cluster 2 is cluster 2 etc. I’ve been asked to reorder them, but it seems to me that the order might have some meaning. Based on subsequent analyses comparing clusters on outcome variables (my validity procedure for the clusters) it appeared that the last two clusters were more different from the first 3 clusters. I’ve included some line graphs of the means of the clusters for a presentation, and I’m afraid if I reorder them the picture will become clouded.
Thanks
Matt
No virus found in this message.
Checked by AVG - www.avg.com
Version: 10.0.1382 / Virus Database: 1513/3711 - Release Date: 06/18/11
Locked
