Kaszab Zoltán (szerk.): A Magyar Természettudományi Múzeum évkönyve 76. (Budapest 1984)

Demeter, A. ; Lázár, P.: Morphometric analysis of field mice Apodemus: character selection for routine identification (Mammalia)

Table 4. Twenty-two characters, accounting for 95% of the total variance, rank-ordered by the sum of squares method (ORLÓCI 1973) Rank Character Specific variance Relative importance (%) 1 Dim3 46.58 71.66 2 Wipa 1.43 2.20 3 Venx 1.29 1.98 4 Dory 1.27 1.95 5 Dorv 1.15 1.77 6 Wich 1.11 1.71 7 Venv 1.02 1.67 8 Lefi 0.94 1.46 9 Dorx 0.84 1.30 10 Veny 0.79 1.22 11 Hefm 0.72 1.11 12 Dcna 0.63 0.98 13 Wpml 0.55 0.85 14 Dcoc 0.48 0.74 15 Wzyg 0.46 0.71 16 Ghes 0.43 0.67 17 Wifm 0.42 0.65 18 Lmra 0.39 0.60 19 Spsp 0.37 0.57 20 Habl 0.34 0.52 21 Lioc 0.26 0.40 22 Hnac 0.25 0.38 Total 62.60 95.00 exceptions. Most of these loadings are high, therefore it is impossible to identify any variable, or a selected group of variables, with this new axis. Only a few variables had significant loadings on P.C. II and III, therefore they are more readily identified. Broadly speaking, these two components are best described as ones containing the variance of characters which are not related to the major axis of the body /skull of the mice. P.C. I is therefore regarded as a "size" component, which accounts for most of the variation in the sample. This is a wellknown outcome of PCA in morphometries. The principal factor analysis was even less successful. Highly significant loadings with the first two factors were found for most of the variables, and only factor III stands out clearly as the one responsible for covariation in measurements along the minor axis of the skull. Fig. 8 depicts the original variables in the 3-dimensional factor space. There are no definite clusters of variables with which the factors could be identified; there is a compact group containing most of the variables, and some others, not forming any other recognizable cluster, are farther dispersed. It may be seen that even factor III cannot really be considered as common to the 3 variables indicated. In general, principal component and factor analyses have turned out to be rather inefficent methods for selecting a subset of the original variables.