3.6 Conclusion
Although association matrices are in most cases intermediate entities in a data
analysis, this chapter has shown that their computation deserves close attention.
Many choices are available, and crucial decisions must be made at this step of the
analytical procedure. The graphical tools presented in this chapter are precious helps
to make these decisions, but great care must be taken to remain on solid theoretical
ground. The success of the analytical steps following the computation of an association matrix largely depends on the choice of an appropriate association measure.
Commonly used distance functions in the Q mode, available in R packages, are
presented in Table 3.1.
The similarity coefficients for presence-absence data and the Gower similarity
may be transformed into dissimilarities using
ffiffiffiffiffiffiffiffiffiffiffi
1 À S
p
to avoid the production of
negative eigenvalues and complex eigenvectors in principal coordinate analysis.
This transformation is made automatically by ade4, but not by vegan. dist.
ldc() of adespatial does this transformation for Jaccard, Sørensen and
Ochiai.
In Table 3.1, functions decostand() and vegdist() belong to package
vegan; function dist.binary() belongs to package ade4; function daisy
() belongs to package cluster. Other R functions propose dissimilarity coefficients. Some are mentioned in the course of this chapter. The Euclidean distance can
be computed either by vegdist(.,"euc") as shown in the table, by daisy
(.,"euc"), by dist.ldc(.,"euc") or by dist(.).
3.6 Conclusion
57
Although association matrices are in most cases intermediate entities in a data
analysis, this chapter has shown that their computation deserves close attention.
Many choices are available, and crucial decisions must be made at this step of the
analytical procedure. The graphical tools presented in this chapter are precious helps
to make these decisions, but great care must be taken to remain on solid theoretical
ground. The success of the analytical steps following the computation of an association matrix largely depends on the choice of an appropriate association measure.
Commonly used distance functions in the Q mode, available in R packages, are
presented in Table 3.1.
The similarity coefficients for presence-absence data and the Gower similarity
may be transformed into dissimilarities using
ffiffiffiffiffiffiffiffiffiffiffi
1 À S
p
to avoid the production of
negative eigenvalues and complex eigenvectors in principal coordinate analysis.
This transformation is made automatically by ade4, but not by vegan. dist.
ldc() of adespatial does this transformation for Jaccard, Sørensen and
Ochiai.
In Table 3.1, functions decostand() and vegdist() belong to package
vegan; function dist.binary() belongs to package ade4; function daisy
() belongs to package cluster. Other R functions propose dissimilarity coefficients. Some are mentioned in the course of this chapter. The Euclidean distance can
be computed either by vegdist(.,"euc") as shown in the table, by daisy
(.,"euc"), by dist.ldc(.,"euc") or by dist(.).
3.6 Conclusion
57