Can I perform Z-score values on percentages? It is correct? Or should I perform Z-values starting from frequencies?

I want to perform Z-score on percentages to 'declosed' my data, as, If I understood right, percentage are compositional data, so they are closed.

My raw data is a cross-tab, with 10 cases on the rows (X), and 8 variables in the columns (Y). Percentages are calculated for each row, and each one sum to 100%. It is right to perform Z-score on them, before running a cluster analysis?

  • 2
    $\begingroup$ Of course you can carry out these calculations. Whether they are "correct" depends on what you will do with the results! What purpose do you have in mind? $\endgroup$ – whuber Jan 22 '15 at 20:48
  • $\begingroup$ What do "vertical" and "horizontal" mean in this context? When you refer to rows and columns, could you show us how you are holding data? $\endgroup$ – Nick Cox Jan 22 '15 at 21:08
  • $\begingroup$ Not clarified. Voting to close as unclear. $\endgroup$ – Nick Cox Jan 25 '15 at 12:46
  • $\begingroup$ You should really state the real problem you want to solve! cluster-analysis on ten cases doesnt look like a good idea. $\endgroup$ – kjetil b halvorsen Jan 29 '15 at 10:16
  • $\begingroup$ Igorning any other possible issues, what you have is a "compositional data" problem. A search for that term will lead you to various methods used to deal with such data. Often some kind of log-contrast transform is made to move the data onto real space where "traditional" statistical methods can be employed as normal. en.wikipedia.org/wiki/Compositional_data There is a very nice group in Girona Spain who focus on this kind of analysis and they run summer courses every year. $\endgroup$ – D L Dahly Jan 29 '15 at 10:27

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.