Vörös A. szerk.: Fragmenta Mineralogica Et Palaentologica 14. 1989. (Budapest, 1989)
lationship between the variables and the new axes are illustrated by the loadings. High loading value of a variable indicates its significant role in the dispersion, along the direction defined by the given eigenvector. The original data are then projected onto the eigenvectors, producing principal component scores. A principal component variation diagram can be obtained by the plot of the first principal scores against the second and/or the third one. Table 2 Eigenvectors and eigenvalues of Mecsek volcanics (n= 75 samples) Eigenvectors extracted from var-cov matrix: Eigenvalues : Eigenvalues : 1 2 3 4 Si0 2 -0. 649 -0. 436 -0 . 234 0. 415 Ti0 2 0. 151 -0. 024 0 . 055 0. 062 A1 2 0 3 -0. 1 9 3 0. 732 0 . 066 -0.037 FeOtot 0. 376 -0. 176 -0 . 424 -0.359 MgO 0. 316 -0. 102 0 . 678 0. 461 CaO 0. 444 -0. 018 -0 . 336 0. 376 Na 2 0 -0. 177 0. 430 -0 . 181 D. 194 K 2 0 -0. 221 -0. 206 0 . 386 -0.502 P 2°5 0. 009 -0. 057 0 . 0 58 -0.224 58 . 71 14 .03 3. 11 2. 45 % 71 . 83 16 i. 87 3. 74 2. 94 Acc% 7 1 . 83 . 70 92. 44 95. 38 extracted from corr. matrix : 1 2 . 3 4 Si0 2 -0. 374 0. 249 -0 . 311 -0. 125 Ti0 2 0. 375 -0. 0 58 0 . 001 0.47 5 A1 2 0 3 -0. 261 -0. 516 0 . 357 0. 282 FeOtot 0. 392 0. 027 -0 . 002 -0.526 MgO 0. 371 -0. 019 -0 . 091 0. 486 CaO 0. 40 5 -0. 118 0 . 042 -0.209 Xa 2 0 -0. 30 3 -0. 462 0 . 245 -0.182 K 2 0 -0. 319 0. 444 -0 . 075 0. 296 P 20 5 0. 053 0. 492 0 . 836 -0.026 5 . 34 1 . 72 0. 79 0. 37 % . 59 19 . 19 8. 81 4. 12 Acc% 59 . 59 78 . 78 87. 59 91.71 Eigenvectors and eigenvalues can be extracted from the variance-covariance or the correlation matrix. An effective guide to the choice of the method are given by LE MAITRE (1982). Our test confirms his suggestion that the variance-covariance matrix is recommended to use for major element composition data. The correlation matrix may be seriously influenced by various effects, e. g. wild data values, very common sample types overwhelming the characteristics of less usual samples, scaling of the data etc. (HOWARTH 1983). Due to standardization, the weight of the variables having a great influence on the distortion decrease (all the variables will have unit variance) and the result can hardly be interpreted. As we noted above, it seems preferable to weight the variables according to their magnitude.
