Contents

  1. Program for transformation of frequency data

  2. Program for biplot species scores

Program for transformation of frequency data
User's notes

Five transformations for frequency data, including species abundances, have been described by Legendre and Gallagher (2001). The program (distribution: see below) converts a matrix of species abundances in such a way that the Euclidean distance among rows of the transformed matrix is equal to one of the following distances among rows of the original data matrix:
  1. Chord distance
  2. Chi-square metric
  3. Chi-square distance
  4. Distance between species profiles
  5. Hellinger distance
Input file: Rectangular table of frequencies where the rows correspond to objects (e.g., sites) and the columns to variables (e.g., species). There are no row (= site) or column (= species) identifiers. It is recommended to add a carriage return at the end of the last row of data. Output file: Rectangular table of transformed frequencies where the rows correspond to the same objects and the columns to the same variables. This table can now be used as input to programs of data analysis that, normally, preserve the Euclidean distance among rows, i.e., principal component analysis (PCA), redundancy analysis (RDA), or K-means partitioning of data sets; they will now preserve the selected distance among objects.
Example Input file: Frequency data for 3 species (columns) at 3 sites (rows), from Legendre and Legendre (1998, p. 457): 
        10      10      20
        10      15      10
        15      5       5
Output file: Transformation such that Euclidean distances computed among rows of transformed data are equal to Chord distances among the original sites: 
        0.40825         0.40825         0.81650
        0.48507         0.72761         0.48507
        0.90453         0.30151         0.30151
Output file: Transformation such that Euclidean distances computed among rows of transformed data are equal to Chi-square distances among the original sites: 
        0.42258         0.45644         0.84515
        0.48295         0.78246         0.48295
        1.01419         0.36515         0.33806
Output file: Transformation such that Euclidean distances computed among rows of transformed data are equal to Hellinger distances among the original sites: 
        0.50000         0.50000         0.70711
        0.53452         0.65465         0.53452
        0.77460         0.44721         0.44721

Program Distribution

Updated March 22, 2001
A pre-print version of the article is Acrobat document. Computer programs to carry out these transformations are available from the following WWWeb sites:

Program for biplot species scores
User's notes

After distance-based redundancy analysis (db-RDA, Legendre and Anderson, 1999) using Bray-Curtis (or other) distance, species scores may be obtained by computing correlations between the species and the fitted site scores, also called "site scores that are linear combinations of environmental variables" or "sample scores that are linear combinations of environmental variables". The correlations have to be weighted as follows before being used to draw the species as arrows in biplots:
SpeciesScore(jk) = r(jk)*s(j)/s(k)
where
r(jk) = correlation between species j and fitted site score vector k,
s(j) is the standard deviation of species j,
and s(k) is the standard deviation of fitted site score vector k.
In standard RDA, the lengths of the species score vectors are 1 in distance biplots and they are equal to the square roots of the corresponding canonical eigenvalues in correlation biplots (Legendre and Legendre 1998), just as in PCA. Different programs may scale the site scores in different ways, however, as can be seen by comparing the three examples below. This may result in species scores that are larger than the original ones by a constant, which does not change the interpretation of biplots. The RDA scalings implemented in different programs are described in Legendre and Legendre (1998, pp. 585-586). This program offers the option of scaling the species scores as in standard RDA, or as in program CANOCO versions 3 or 4. Compared to standard RDA, described in Legendre & Legendre (1998), the standard deviations of the site scores differ by a constant in CANOCO, so that the site scores also differ by the following constants:
n / sqrt(totinert) in CANOCO 3.10
sqrt(n*p) / sqrt(totinert) in CANOCO 4
where n is the sumber of objects (e.g., sites), p is the number of variables (species), and 'totinert' is the total inertia in the species matrix. Input file 1: Species data text file. Rectangular table of species presence-absence or frequency data where the rows correspond to objects (e.g., sites) and the columns to species. There are no row (= site) or column (= species) identifiers. Data are separated by spaces or tabs. It is recommended to add a carriage return at the end of the last row of data. Input file 2: Text file of "fitted site scores", or "sample scores that are linear combinations of environmental variables", where rows correspond to objects (e.g., sites) and columns to canonical eigenvalues. This table is copied from the output file of Canoco or similar program into an Ascii (i.e., text) file. There are no row (= site) or column (= species) identifiers. Data are separated by spaces or tabs. The program checks that the number of rows is the same in input files 1 and 2. Add a carriage return at the end of the last row of site scores. Output file: Table of species scores for biplots. The rows correspond to the species and the columns to the canonical eigenvalues. This file may be added to the Species Scores and Site Scores (or Fitted Site Scores) tables to produce biplots. An automatic biplot drawing procedure is available in The R Package, version 4.0 (Casgrain and Legendre, 1999). The examples that follow were all scaled to obtain distance biplots. Program SpeciesScores can also compute biplot species scores from "fitted site scores" scaled to obtain correlation biplots. In all cases, users of CANOCO should request scalings for covariance-based scores in CANOCO 3.1 (negative-number options), or without post-transformation in CANOCO 4.
Example 1 Coral reef fish data from Table 11.3 of Legendre and Legendre (1998, p. 590). There are 10 sites (rows) and 6 species (columns). Input file 1: Species file (text file; no row or column identifiers). 
       1       0       0       0       0       0
       0       0       0       0       0       0
       0       1       0       0       0       0
       11      4       0       0       8       1
       11      5       17      7       0       0
       9       6       0       0       6       2
       9       7       13      10      0       0
       7       8       0       0       4       3
       7       9       10      13      0       0
       5       10      0       0       2       4
Input file 2: fitted site scores from program RdaCca. Scaling 1 was used here to obtain a distance biplot. The columns correspond to the three canonical eigenvalues. (Text file; no row or column identifiers). 
       -6.79498         5.49498        -2.24897
       -6.96197         5.91719        -0.63774
       -7.12895         6.33941         0.97349
       -3.55205        -6.52301        -4.39356
       12.69996         0.24686        -3.17159
       -3.88603        -5.67858        -1.17109
       12.36599         1.09129         0.05088
       -4.22000        -4.83415         2.05138
       12.03201         1.93572         3.27335
       -4.55398        -3.98972         5.27384
Output file: Species scores for biplots, computed by program SpeciesScores. The rows correspond to the species and the columns to the canonical eigenvalues. This table is identical to the table of species scores computed by program RdaCca and reported in Table 11.4 of Legendre and Legendre (1998, p. 591). 
        0.30127        -0.64624        -0.39939
        0.20038        -0.47265         0.74458
        0.74098         0.16813        -0.25689
        0.55013         0.16841         0.26114
       -0.11588        -0.50594        -0.29319
       -0.06292        -0.21535         0.25679
Example 2 Input file 1: Same species file as above. Input file 2: fitted site scores from program CANOCO 3.1. Scaling -1 was used to obtain a distance biplot with covariance-based scores. The columns correspond to the three canonical eigenvalues. (Text file; no row or column identifiers). 
       -0.6741         -0.5452         -0.2231
       -0.6907         -0.5870         -0.0633
       -0.7073         -0.6289          0.0966
       -0.3524          0.6471         -0.4359
        1.2600         -0.0245         -0.3147
       -0.3855          0.5634         -0.1162
        1.2268         -0.1083          0.0050
       -0.4187          0.4796          0.2035
        1.1937         -0.1920          0.3247
       -0.4518          0.3958          0.5232
Output file: Species scores for biplots. The rows correspond to species and the columns to canonical eigenvalues. 
        0.95271         2.04359        -1.26323
        0.63365         1.49471         2.35440
        2.34316        -0.53170        -0.81282
        1.73965        -0.53257         0.82547
       -0.36643         1.59994        -0.92714
       -0.19897         0.68102         0.81207
Example 3 Input file 1: Same species file as above. Input file 2: fitted site scores from program CANOCO 4. Biplot scores emphasizing inter-sample distances, without post-transformation (scaling -1), were computed to obtain a distance biplot. The columns correspond to the three canonical eigenvalues. (Text file). 
       -0.6741         -0.5452         -0.2231
       -0.6907         -0.5870         -0.0633
       -0.7073         -0.6289          0.0966
       -0.3524          0.6471         -0.4359
        1.2600         -0.0245         -0.3147
       -0.3855          0.5634         -0.1162
        1.2268         -0.1083          0.0050
       -0.4187          0.4796          0.2035
        1.1937         -0.1920          0.3247
       -0.4518          0.3958          0.5232
Output file: Species scores for biplots. The rows correspond to species and the columns to canonical eigenvalues. 
        0.73797         1.58296        -0.97849
        0.49082         1.15779         1.82371
        1.81501        -0.41186        -0.62961
        1.34753        -0.41253         0.63941
       -0.28384         1.23931        -0.71816
       -0.15412         0.52751         0.62902
Distribution A computer program to compute the biplot species scores is available from the following WWWeb site:

Fortran source code and compiled versions, written by P. Legendre:

References
Casgrain, P. and P. Legendre. 2001. The R Package for multivariate and spatial analysis, version 4.0 - User's manual. Departement de sciences biologiques, Universite de Montreal. Available from the Web site http://numericalecology.com/ . Legendre, P. & M. J. Anderson. 1999. Distance-based redundancy analysis: testing multi-species responses in multi-factorial ecological experiments. Ecological Monographs 69 (1): 1-24. Legendre, P. and E. Gallagher. 2001. Ecologically meaningful transformations for ordination of species data. Oecologia 129: 271-280. (Reprint available, © 2001 "Springer-Verlag". The original publication is available on http://link.springer.de/) Legendre, P. & Legendre, L. 1998. Numerical Ecology, 2nd English edition. Elsevier Science BV, Amsterdam. xv + 853 pages.