 | Hi!
I want to do a cluster analysis with dicotomized data and want to use a
measure of distance which handles this type of data, preferably the
Jaccard. From what I have experienced there is a limit of how many cases
you can cluster on using the Jaccard measure (in SPSS). I tried to cluster
on 1000 subjects but I didn't receive cluster membership for any subject.
However, if doing the analysis with the procedure two-step cluster under
the menu Analyze- Classify in SPSS, specifying the variables as
categorical and the distance measure as log-likelihood I receive
clustermembership.
From what I understand two-step clusteranalysis both perform the
clusteranalysis and chooses the correct amount of clusters. On the other
hand, you cannot choose distance measure or the method to be used (Ward
etc). In one way, it feels like " a black box".
Is it OK to do use Two-step cluster analysis in this situation ( in SPSS
the Help-menu states that categorical in this situation means ordinal
variables and not necessarily dicotomized variables)? I also have access to
SAS and does anyone know if the cluster analysis is carried out for about
1000 subjects using the Jaccard option?
Thankful for any help!
Magnus Alderling
|
|
