I have a group of 144 people. I have 3 categorical observations and each of them is described by three variables. After performing a k-means on 3 clusters in r I see that one of the groups, "overt" is clustered together and the other two groups overlap a lot. Moreover when clustering with 2 clusters, the "overt" group again has their own cluster while the other two groups overlap.

My question: Is this enough to conclude that the groups "chemical" and "normal" are statistically the same and the third group is different?

I feel like I didn't really perform any statistical experiment in here. Could anyone give me some hints on how to analyse this correctly? Thanks.

DIABETES.new = diabetes 
DIABETES.new$Class <- NULL 
DIABETES.standard <- scale(DIABETES.new)  
results1 <-kmeans(DIABETES.standard, 3) 
table(diabetes$Class, results1$cluster)

#           1  2  3
#Chemical  0 18 18
#Normal    0 11 64
#Overt    20  3  9

results2 <-kmeans(DIABETES.standard, 2) 
table(diabetes$Class, results2$cluster)

#         1  2
#Chemical 36  0
#Normal   75  0
#Overt    12 20
  • $\begingroup$ K-means is sensitive to data preprocessing. So you can only say that this applies to your preprocessing. $\endgroup$ – Anony-Mousse May 1 '18 at 6:45
  • $\begingroup$ Of course I understand that and with this assumption is what I am saying correct? $\endgroup$ – Patrick May 1 '18 at 12:19
  • $\begingroup$ No. KMeans is not a statistical test. It is a randomized algorithm, it may fail to find patterns and it can produce wrong patterns. $\endgroup$ – Anony-Mousse May 1 '18 at 22:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.