Angela
Elite member
- Messages
- 21,823
- Reaction score
- 12,329
- Points
- 113
- Ethnic group
- Italian
I am sure Angela knows how to defend herself, but it is obvious she is not accusing of fraud to anybody. Have you felt called in?
The issue of supervised or non-supervised is not the relevant point, if samples have been "massaged" ex-ante. For sure, misclassification is a serious issue, but if some samples that should be in, are not there, or viceversa, the unsupervised algorithm may fail to find the right structure. In fact, unsupervised algorithms tend to be "worse" than supervised ones (of course, lots of caveats here), so I am surprised that an unsupervised algorithm is seen as an evolution of supervised ones.
About Iberian regions: if there is so much overlap, maybe it would be interesting to have not only a measure of the distance, but also a measure of the error. For example, if all the first 5 estimated regions were within the error bands, one could not say that the first region is more important than the fifth. Now, most people believe that the first one is the most important one, and anything above the 2nd or 3rd is discounted. But the ranking could be purely due to noise, if the distance among regions is so overlapping.
Couldn't have put it better.