[Traminer-users] Antw: selecting the number of clusters

Gerhard Wührer Gerhard.Wuehrer at jku.at
Wed Jun 10 12:55:06 CEST 2015

** Proprietary **
** Reply Requested When Convenient **


please try the R - package 'nbclust' do decide how many clusters are
feasable. In addition to that statistical measures, inspect the
different cluster solutions by content and how meaningful
interpretations are. You can also do some kind of x-square test where
you align the clusters with variables not used in the cluster analysis. 
At least you have some kind of face validity. 

Best regards - Gerhard A. Wührer

o. Univ.-Prof. Dkfm. Dr. Gerhard A. Wührer
Institut für Handel, Absatz und Marketing
Johannes Kepler Universität Linz
Altenberger Str. 69
4040 Linz/Austria
tel.: 004373224689401
mail: gerhard.wuehrer at jku.at
URL: www.marketing.jku.at

>>> Rimantas Vosylis <rvosylis at live.com> 10.06.2015 12:30 >>>
Dear Traminer users,
I am trying to build a typology of sequences by using cluster analysis
with OM and Ward algorith.
I have a problem of choosing the number of clusters. I use several
empirical indexes, but they don‘t help me a lot. I use Calinski and
harabasz (CH) index, but it has a peak at two cluster solution and the
goes down. I also use average shilloute width but it gives me the
similar results as CH index. I also run pseudo ANOVA to see which
cluster solution explains most variance, but it tells me the opposite –
the more the clusters the higher the pseudo R2 gets. When I look at the
various plots (e.g. seqdplot) I see that the most meaningful solutions
(I have several types of sequences) lie somewhere between 4-6 clusters.
Could You perhaps suggest which indexes worked best for You and matched
Your expectations / theoretical knowledge and that I could use in my
Thank You in advance!!
Rimantas Vosylis
PhD student, lecturer
Insitute of Psychology
Faculty of Social Technologies
Mykolas Romeris University
e-mail: rimantasv at mruni.eu
e-mail2: rvosylis at live.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/traminer-users/attachments/20150610/49e29173/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: W?hrer, Gerhard.vcf
Type: application/octet-stream
Size: 348 bytes
Desc: not available
URL: <http://lists.r-forge.r-project.org/pipermail/traminer-users/attachments/20150610/49e29173/attachment.obj>

More information about the Traminer-users mailing list