I finally understand cluster analysis
After a few days of reading boring articles about statistics, looking at dendrograms, pulling data apart and seeing how things work, I finally understand the basics of cluster analysis. I'm not sure whether my description of it and how to read dendrograms makes sense, but at least I basically get it.
Some conclusions about cluster analysis and card sorting tools:
- Cluster analysis is a particularly poor way to visualise outcomes from a card sort
- It would be possible for card sorting tools to allow sub-grouping and still use cluster analysis in the output (I always figured this is why tools didn't include sub-groups)
- Current card sorting tools do nothing to help people understand the results they display. There is no information about the method and little transparency
- To do better statistical analysis, I'd need a much deeper understanding of statistical methods than I currently have (or want to have) and some very expensive statistical analysis tools
- Until someone adds good analysis capabilities to card sorting tools, practitioners will continue to be ignorant of better statistical analysis methods
No wonder everyone tells me they find it hard to analyse the outcomes of a card sort!
Comments
Hi!
I would like to comment some points about this entry, in what relates to the existing card sorting tools, specifically xSort.
First, I would like to clarify that the latest version of the software, xSort 1.5, supports cluster analysis over a card sorting exercise with sub-groups. The most important decision we had to make to support both features was how to measure the distances between the cards on different situations. We believe we achieved a nice result with out algorithm. Besides, xSort users may choose to include the sub-group data in the cluster analysis, or to flatten the sub-groups into simple groups.
xSort was also designed with the purpose of simplifying the data analysis and results view as much as possible. This does not mean our users shouldn't know about cluster analysis and the concepts behind it. That's why we explain them, and include examples, in the xSort documentation.
We want to state very clearly that xSort is a tool to be used by professionals. It's goal is not to produce a straight-forward solution for one's specific problem. That's simply an impossible talk for a machine. Those decisions must be made by humans, based on several types of data and information.
That's where xSort may be useful - we provide a very simple, yet powerful cluster analysis tool. It's usage is very simple - for example, you don't even need to trigger the calculations, when you switch to the result tab, they are already done for you. Also, when you drag a mark over the dendogram with the results, color shades will appear, suggesting the resulting clusters.
Naturally, xSort is not comparable to a complex and expensive statistical tool, as it's not it's purpose. We believe card sorting is a simple method, so we decided that xSort shouldn't also be complex. Again, that doesn't mean people shouldn't know the concepts behind, and rely only on the software. We just provide a very easy way to work with those concepts.
Good luck with your book, hope to read it soon! :)
Yours
Miguel Arroz
http://www.ipragma.com
Posted by: Miguel Arroz | July 25, 2006 02:23 AM
Thanks so much for outlining that. I posted this before I had a good look at xSort - I had to get a friend to run me through it as I don't have a Mac...
Posted by: Donna Maurer | July 31, 2006 08:59 PM