problem with sas code

[atulsingh]

Hi all,

I have analysed the correlation between two variables: hp and cyl and  i'm writing a code for cluster analysis for the same varibles but ending up with error, need help. The code is:

proc corr data=rank_mtcars out=cluster_mtcars; /*Here rank_mtcars is a predefined dataset*/
var hp cyl;run;
proc cluster data= cluster_mtcars method=centroid rmsstd
outtree=cluster_mtcars;
var cyl hp;run;

ERROR: TYPE=ACE, CORR, COV, EST, FACTOR, SSCP, UCORR, or UCOV data sets must be converted to TYPE=DISTANCE. Note that CLUSTER

ignores the _NAME_ and _TYPE_ variables, if any.

[PaigeMiller]

Don't you want to do the clustering on the original data set rank_mtcars? It doesn't make sense to me to cluster the correlation matrix.

[atulsingh]

Thanks Sir.

One  more question: After clustering i got the output as dendogram. 

How to obtain the optimum number of clusters from the dendogram?

[PaigeMiller · Accepted solution]

PROC CLUSTER documentation explains the common ways to do this:

http://documentation.sas.com/?cdcId=statcdc&cdcVersion=14.2&docsetId=statug&docsetTarget=statug_clus...

[atulsingh]

Thanks Sir..!!

[Ksharp]

BTW, Maybe you need 1-rho to make CORR matrix looks like Distance matrix .

[vijaylaxmi]

Hi,

Can you please provide me dataset rank_mtcars. I am not able to find it anywhere.

Thanks!!

Vijay