I have categorical (nominal) variables (22 variables), which are responses to the survey questions. I would like to reduce number of these variables to get different response patterns. After applying the multiple correspondence analysis (MCA) in SPSS I get many dimensions, each of them explaining in average 10% and less variance. So I was thinking to apply cluster analysis on the scores from the MCA in order to get some reasonable number of response patterns. Is it ok from a statistical point of view?
Yes, this is usually done for example by the SPAD software. After quantification (i.e. a Multiple Correspondence Analysis) a clustering step on the scores is performed. So, this is not an heresy!!! However it is much interesting the description of clusters, that is made through the analysis of the proportions of units with a specific category in the cluster with respect the proportion of unit with the same category in all sample. Something is present in the R package FactomineR and is explained here https://groups.google.com/forum/#!topic/factominer-users/55w8gvADWJo
the pvalue and the v.test are based on the comparison of the category proportion in th cluster and the category proportion in the sample.
Yes they can. Actually, MCA is often combined with Clusters Analysis (CA). If you use SPSS, just check" save object scores to the active dataset" . After that, a hierarchical clustering on scores can be carried out. The Ward’s method is normally used to aggregate cases. With this method, the sum squared distances of points from their cluster centroids are minimized. You can confirme the "optimized" number of clusters through a K-means clustering procedure. Finally, the found clusters can be projected on the multidimensional plan (biplot). I hope it helps. Cheers.
Dear Antonio and Pedro
Thank you very much for your answers.
As I know SPAD do the clustering somehow simultaneously with MCA and I wasn't sure whether in SPSS is it ok to do these processes consecutively.
But you helped me :)
Antonio thanks also for the link to FactormineR.
Since you are at University of Évora, there is a portuguese book explaining all the procedures through SPSS! The book's author is Helena Carvalho and all the content is about MCA combined with CA. Check the link :
http://www.wook.pt/ficha/analise-multivariada-de-dados-qualitativos/a/id/204640
- Badges
- Science method
- Similar topics
- Methodology
- Research Methodology
- Qualitative Research