I just need a very short summary of what the standard way to deal with compositional data is.

I've skimmed pages in a 500-page long book on the topic, and I didn't really gather much. I would like to know the general idea before I delve into the details.

By compositional data, I mean response variables $p_1, ..., p_k$ with $\sum p_i = \mathcal{S}$ for some fixed value $\mathcal{S}$.

So far, I gather that we consider the "simplex" of ALL such response values. This is a $k-1$ dimensional space. On this, it seems a geometry can be defined with some nice properties. We also consider transformations that take a value in the simplex, and returns some coefficients.

And then what? What's the point of those coefficients? Is that choice arbitrary? What will we do with those coefficients to help us analyze the data?

  • $\begingroup$ Do you mean something like the Wikipedia entry? en.wikipedia.org/wiki/Compositional_data $\endgroup$
    – mdewey
    Commented Mar 12, 2017 at 17:11
  • $\begingroup$ This is a broad subject. What is your specific question? $\endgroup$
    – whuber
    Commented Mar 12, 2017 at 17:18

1 Answer 1


Lecture Notes on Compositional Data Analysis These lecture notes by Pawlowsky-Glahn, Egozcue, and Tolosana-Delgado are excellent. There are a few errors in them but it is a much shorter (≈80 pages) read than the full textbook that came out of these notes.

Even shorter: Compositional Data Analysis in a Nutshell (2 pages).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.