Info

set of data, in a sample of fewer dimensions (or axes) that is the best summary of the information contained in the data (Greenacre 1984; Everitt and Dunn 2001). Our data set contains both categorical and continuous variables that were used in the MCA as active (Table 5) and illustrative variables (PL, ESD, CI and Ab defined below), respectively. The factorial axes of the MCA were computed using active (categorical) variables whereas the continuous illustrative (or supplementary) variables were simply projected into this factorial plane without participating in its computation. The location of these continuous illustrative variables, is shown as arrows along the factorial axes in Fig. 1A, and expresses their linkage to the pattern of the categorical active variables displayed by the factorial axes (Lebart et al. 2000). Observations with missing data for carbon-specific ingestion and experimental characteristics were excluded from the MCA. Only four observations with Phaeocystis antartica were available, and were therefore also excluded. A data matrix of the remaining 296 observations with eight active and four illustrative variables were used for the analyses (Table 5). The size of predators (pro-some length; PL) and prey (equivalent spherical diameter; ESD) were treated as illustrative variables.

These continuous variables were used to calculate a predator-to-prey size ratio (P:p) with five categories (or modalities) used in the statistical treatment. Similarly, C-specific ingestion (CI) and Phaeocystis abundance (Ab) were transformed into categorical variables of three and five categories, respectively, to be tested both as illustrative and active variables in the MCA. The metric used in the MCA is based on the x2. This was also the metric used in the following cluster analysis. The type of HCA used here is an agglomerative clustering, i.e., a procedure that successively groups the closest objects into clusters, which then are grouped into larger clusters of higher rank (Legendre and Legendre 1998). The programme identifies: (i) the cluster (group of observations) which has the smallest within-group variance and the greater variance between groups, and (ii) the descriptors (variables) that are highly representative of each cluster. The descriptors of the observations used in the clustering were their factorial coordinates on the first six axes obtained in the MCA. These first six axes explained about 58% of the total variability (inertia) in the data, and the additional amount of variability explained by the axes decreased markedly after the first six axes.

Fig. 1 First factorial plane of MCA of data on crustacean grazing experiment on Phaeocystis. (A) Projections of continuous illustrative variables in the correlation circle (radius 1) and ordination of active variables: Phaeocystis species (A), growth (<) and abundance (O), crustacean species (V), predator-to-prey

Fig. 1 First factorial plane of MCA of data on crustacean grazing experiment on Phaeocystis. (A) Projections of continuous illustrative variables in the correlation circle (radius 1) and ordination of active variables: Phaeocystis species (A), growth (<) and abundance (O), crustacean species (V), predator-to-prey size ratio ([>), experiment location (□), selection (O). (B) Ordination of data (O) labelled according to their C-specific ingestion characteristic and delimitation of the four groups designed by the hierarchical clustering (y2 distance). See Table 5 for label identification

Results from statistical analysis

The result of the MCA indicated that the Wrst and second axes accumulate about 19% and 12% of the total variability, respectively (Fig. 1A). Two variables basically contribute to the Wrst axis: the selection

("Against" and "Single", cumulated 18% of the contribution to axis 1) and the experiment location ("Lab" and "Field", cumulated 17.5% of the contribution). Thus, this axis differentiates laboratory experiments where Phaeocystis was the only prey ("Single"), as opposed to field experiments where

Table 7 Characterization of the most typical properties of the four groups designed by the hierachical cluster analysis defined in Fig. 1 panel B

Group n 1

0 0

Post a comment