Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Starting Points k-means CCEA

Dear Sawtooth-Community,

I am currently working on a cluster analysis in your CCEA software. On page 53 in your technical document [CCEA v3, June 9 ,2008] you describe the methods behind the alternative starting points. In my project, I run the clustering on individual part-worth profiles (zero-centered diffs; obtained by CBC/HB) and would like to ask whether the notation "points or object points" refers to "respondents"?

For example: "Choose a random sample of objects from the entire set to be clustered [...] If there are k to be chosen, select the first k object points as a temporary set."

What would the objects and object points, respectively, represent when clustering on let's say 500 respondents characterized by 10 part-worths each?

Furthermore, are there any academic papers regarding the three presented approaches (distance-based, density-based, hierarchical-based) besides your CCEA manual that could serve as further references?

Thank you for your input.

All the best

asked Apr 10, 2014 by Arnold
retagged Sep 1, 2016 by Walter Williams

2 Answers

0 votes
In this case objects or object points refers to the respondents.

I don't know about academic papers regarding the distance or density methods, but hierarchical clustering has extensive documentation across the industry.
answered Apr 10, 2014 by Walter Williams Gold Sawtooth Software, Inc. (20,005 points)
0 votes
Hello, Arnold.

Objects are indeed your respondents.  We use the more general word "objects" because people use cluster analysis for all sorts of applications and the things clustered aren't always survey respondents (e.g. they might be cities, species of flowers, national economies, companies, anything that can be described in multiple quantitative attributes.  

I think of object points as being the 500 points located in a 10 dimensional Euclidean space defined by your 10 part worth utilities.  

As Walt notes there are several academic review papers and books comparing different sorts of cluster analysis and distance metrics.  A classic review article by Milligan and Cooper in the December 1987 issue of Applied Psychological Measurement might make a nice starting point.
answered Apr 10, 2014 by Keith Chrzan Platinum Sawtooth Software, Inc. (92,075 points)