I have data that represent some aspect of driver behavior. I want to cluster the set of data into behavioral profiles of some sort. Some of my variables are categorical and some are continuous.
My questions relate to how one might go about categorising this data. What are some (common?) approaches to this analysis? Is it possible or even sensible? Can this be achieved in SPSS? Are there some references or directions that will help me further understand how to deal with this challenge?
I believe strongly that you can achieve this sort of analysis using SPSS. It might be more useful if you can provide your data for more help. Thanks!
I would be inclined to use R rather than SPSS. For R it is pretty straightforward:
http://stats.stackexchange.com/questions/24540/categorical-variables-clustering-with-r
As Frederic says, more info will be helpful, but if you are comparing sets of variables, one approach is canonical correlations (which are in SPSS), but this makes many assumptions which looks like are inappropriate for your data. SPSS used to (may still have, I don't use it) CANALS and OVERALS which may work for what you need.
Van der Geer (1993) Multivariate analysis of categorical data: Applications. Sage. goes through examples with numeric and categorical data.
Thanks, Daniel. I'll have a look at the Van der Geer book.
Thanks, John! Bobbie mentioned that you and Linda have done a webinar on R for Data Analysis and Visualization - I've found reference to it on the HFES website, but I can't find the actual webinar. If there is one, could you direct me to an online link, please?
In SPSS you have the possibility to use PCA with continuous and Cathegorical Variables.
Some references:
Meulman J.J., van der Kooij A.J., and Babinec A. (2000). New Features of Categorical Principal Components Analysis for Complicated Data Sets, Including Data Mining. In W. Gaul and G. Ritter (Eds.) Classification, Automation and New Media (pp. 207-217). Berlin,Germay: Springer.
Van der Kooij, A.J. and Meulman, J.J. (1999). Categorical Principal Components Analysis. In J.J. Meulman, W.J. Heiser, and SPSS Inc. (Eds.), SPPS Categories 10.0. (pp. 1-9, 103-126, 221-237). Chicago: SPSS Inc.
Also notice "A number of different measures have been proposed to measure ’distance’ for binary and
categorical data. For details see the book by Everitt, Landau and Leese. Readers are also
referred to this text for details of what to do if you have a mixture of different data types. " Everitt, B.S., Landau, S. and Leese, M. (2001), Cluster Analysis, Fourth edition,
Arnold.
Tyron, without knowing more about your data it's hard to make *the* recommendation but if you find your way over to the R side (and please do!), you might find your answer in the following packages:
cluster - Contains both hierarchical and partitioning-around-medoids algorithms. This is Rousseeuw's package....and I recommend his book Finding Groups in Data.
dpmixsim - Dirichlet Process mixture modeling. Really neat stuff...fits Gaussians and lines to your data.
flexmix - Flexible mixture modeling. Flexible mixture modeling to fit Gaussians to your data.
These approaches are great for visualizing your data.
Use Twostep Cluster algorithm. its useful for handling both continuous and categorical variables also available in SPSS.
K-prototype clustering can be used for categorical and continuous variables.
- Badges
- Science topic
- Similar topics
- Psychology