Cluster Analysis

Hi,

I have just completed the data gathering for my final thesis. It was a conjoint analysis to evaluate different e-service attributes. I have two types of variables: first, the utility scores for the various e-service attributes; secondly, classification (moderating) variables such as Internet experience, self-efficacy, need for interaction... Goal is to classify the utilities according to these latter variables to see if there are any pos. or neg. moderating effects.
For that purpose I am now supposed to run a cluster analysis to classify the utilities. Is that the right thing to do (k-means or hierarchial)? I rather had the impression that the cluster analysis reveals previously unknown clusters, while I have my classification sheme already. Is there a better method for this purpose?
Thanks for the help.

Christian Knigge

[sdobney]

The most effective use of cluster analysis is to confirm the existance of groupings that you've identified in other research. So I'd be carrying out the cluster analysis and checking how it fits with your a priori groups.

Some might disagree with this because they see cluster analysis as the tool to build these cluster groups, but then you almost always run into the implementation problem - that is that cluster analysis creates groups (and it always will regardless of whether the groups are 'real'), but to make these groups operational you need to be able to apply the clusters in real life. You can't go out and conduct the same attitude or conjoint battery with a customer every time you want to follow up the cluster analysis either with other research (for instance to check the marketing is appropriate to the group), or in terms of practical marketing implementation - just how do you target the clusters accurately. What's the point of a sophisticated cluster analysis only to find that you have to approximate to a crude age or income classification when you actually implement it.

So we advise that you try to identify the groups first, then use the cluster analysis as the confirmation of your hypothesis.


Saul Dobney