Hi, I answer me ;-)

Lasse Ruokolainen & Kauko Salo (2006). Differences in performance of four ordination methods on a complex vegetation dataset. Ann. Bot. Fennici 43: 269–275

The question is complex by:

- Each species can be a different type of gradient (modal vs linear or others)

- Properties of the analysis had been derived of artificial data sets, simulations.

- Each analysis recover a part of the total information of real data sets

- Certain analysis are good recover the main gradient (CA, DCA) and others the general information (PCO, NMDS).

Consequently, we have not the best analysis. However is this paper the authors only test the analysis in a reduced context (two main axis and Bray-Curtis distance for Principal Coordinate Analysis and Non-metric Multidimensional Scaling).

According the authors, the research probably need test different methods, compare between them and finally choose. However, my question remain as relevant. What criteria we must choose?. Because science is objectivity and apparently the multivariate analysis have certain degree of subjectivity. I believe that I have the solution but I need another opinions.

Regards