Hello,

You can first apply hierarchical cluster analysis and consult the produced dendrogram, about the correct cluster number, depending on how central clusters are created. You can also include the dendrogram in your analysis, prooving the number of cluster you used in the K-means method that you will use afterwards.

Hello Mayanka,

The use of k-means implies that you had in mind a specific number of clusters at the outset, so it doesn't really seem like the best path if you are unsure about the number of clusters. Stamatis Ntanos' suggestion of trying a hierarchical approach makes more sense if identification of the "right" number of clusters is a key question in your investigation.

One very simple way to determine whether the derived clusters have any utility is to compare the clusters on variables external to the clustering process. For example, if I cluster students applying to law school based on their academic credentials (at the time of application), do the resultant clusters differ at all on variables such as: (a) grades actually earned in law school subsequent to admission; (b) completion/noncompletion status for graduation; (c) instructor ratings of student competence; etc? If not, then the clusters are not very useful.

Good luck with your work.

Hello Professor,

The very first stage I have used hierarchical clustering only, after knowing the number of cluster from there which came out as 2 in numbers, i then again used k-means cluster by using 2 and 3 clusters. My confusion went ahead when i found that two clusters and three clusters shall be finalised on the basis of some comparative value.

I have 6 factors extracted from factor analysis with desired discriminant validity and reliability.

All the factor in cluster in ANOVA table of cluster are differentiating with each factor.

The reference I have taken for my study has used Latent Class clustering software, in which one can finalise the number of cluster on the basis of BIC value, which SPSS fails to give.

How can I now further decide, that two clusters are logical or three?

Or simply changing the software will give me better results, will it be worth.

Or shall i submit my paper on the basis of results from SPSS and let reviewers give their inputs for my paper.

My main objective behind this study is to make Typology of mobile application users based on their risks. Which I have identified from EFA, now to make the cluster on the basis of these factors, and to explain the cluster traits/features, I shall have some rational decision making to validate the finalised clusters in hand.