We have a dataset of subgroups of the bacteria E. coli. We have frequency data for four subgroups in 38 locations and we want to cluster these locations by using the frequencies of subgroups which occur in each.

Initially we used two clustering methods:

  1. Euclidian distance using the dist{stats} function in R; and

  2. clustering after FactoMineR, using metric="euclidean", and method="ward".

Following this we received feedback that "median clustering would be better due to non-parametric aspects of data". Does this make sense?

  • $\begingroup$ No. Chi-sq or phi distance would be preferrable fot count data, and a non-geometric linkage in clustering, such as average or complete methods. To note also, "median" method actually is not about median in cluster, it is about how the centroid is defined, it has nothing to do with "nonparametrical analysis". See stats.stackexchange.com/a/217742/3277. $\endgroup$ – ttnphns Sep 21 '17 at 2:17

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.