Hi,
the "naive" algorithm (NOINITIAL option) seems to work slightly
better. I hope the option to hand-supply initial cluster centers will be
implemented soon, so that one may start with a hierarchical cluster on a
small sample (maybe in R or hand-crafted: see
https://www.morewithdata.net/blog/20230424-slow-cluster-hierarchical-clustering-with-pspp.html).
Apart from that, I believe DESCRIPTIVES' output is wrong: listwise valid
N cannot exceed a variable valid N. This is not caused by weighting for
it also happens with WEIGHT OFF.
On 8/26/26 12:24, ftr wrote:
My previous email lacked the data which produced the curious result.
Hi,
I come back to my error message in Quickcluster some time ago. Now I
can reproduce the NaN error in the results.
Valid N 4186
N of cases in clusters : 2194;
This case remains a puzzle to me.
DESCRIPTIVES VARIABLES=community2 ego2 order2/save
Mapping of Variables to Z-scores Source Target
community2 Zcommunity2
ego2 Zego2
order2 Zorder2
Descriptive Statistics
N Mean Std Dev Minimum Maximum
community2 4681,53 -,01 ,87 -3,96 2,40
ego2 4681,53 -,02 ,91 -3,14 2,91
order2 4681,53 -,02 ,85 -3,65 2,31
Valid N (listwise) 5056,29
Missing N (listwise) 374,77
QUICK CLUSTER zcommunity2 zego2 zorder2
/CRITERIA=CLUSTERS(3)
/MISSING=EXCLUDE
/PRINT=INITIAL CLUSTER
/SAVE=CLUSTER(clus3) DISTANCE(distclus3)
Initial Cluster Centers
Cluster
1 2 3
Zcommunity2 Z-score of community2 2,02 -4,44 ,58
Zego2 Z-score of ego2 ,61 -2,30 -2,99
Zorder2 Z-score of order2 -3,33 -2,62 2,59
Final Cluster Centers
Cluster
1 2 3
Zcommunity2 Z-score of community2 -,01 NaN 1,50
Zego2 Z-score of ego2 ,03 NaN -6,33
Zorder2 Z-score of order2 -,02 NaN 3,23
Number of Cases in each Cluster
Count
Cluster 1 2193
2 0
3 1
Valid 2194
