General Recommendations for Ordination:

A dichotomous key

---

1. Direct Gradient Analysis................................2
  2. Few species...........................................4
     4. Monotonic responses to gradients (low beta).......
              .............................Linear regression
     4. Nonmonotonic responses to gradients.(high beta)......
              .....................Generalized linear models
  2. Many species..........................................5
     5. Monotonic responses  ............................RDA*
     5. Nonmonotonic responses.............................6
        6. concerned about arch effect..................DCCA
        6. not concerned about arch effect...............CCA
1. Indirect Gradient Analysis..............................3
  3. Only distance values are available....................7
     7. Monotonic responses ............................PCoA
     7. Nonmonotonic responses..........................NMDS
  3. Raw data available....................................8
     8. Monotonic responses ...............................9*
        9. Variables noncommensurate......PCA - corr. matrix
        9. Variables commensurate..........PCA - cov. matrix
     8. Nonmonotonic responses............................10
        10. Feel OK about prespecifying number of dimensions,
 not worried about local optima, not interested in 
 species scoes..............NMDS
        10. Not as above, but willing to accept either arch
 effect or detrending/rescaling................11
            11. Don't like arch, detrending OK ..........DCA
            11. Arch OK, or only interested in axis 1.....CA

The above is just meant as a starting point for discussion. Clearly, the issues are too complex to reduce the decision about which method to use to a dichotomous key. Suggestions for revision are welcome!

*Recently, ter Braak and Smilauer (1998) have pointed out that Correspondence Analysis (and by implications, its derivatives DCA, CCA and DCCA) has two 'faces': a unimodal face and a linear face.  This implies that dichotomies #5 and  #8 may be outdated.  It seems that the major difference between PCA/RDA and the 'linear face' of CA/CCA is that the latter is better for species composition, and not for overall trends in abundance of all species combined.

---

This page was created and is maintained by Michael Palmer.
To the ordination web page