In general, principal component analysis (PCA) is considered an effective procedure to reduce complex information composed of numerous characters to a few incorrelated synthetic groups. The method of PCA on somatic measurements is, however, not established yet. In this report, height and weight, the most commonly used items in somatometry, were examined to clarify the effect of selection of measurement items on PCA.
Materials were 500 boys of age 7, measured in 1966-1967 for their height, weight, posterior arm length, iliospinal height, tibial height, foot length, waist length, posterior shoulder length, neck base girth, bust girth, waist girth, hip girth, scapular and triceps skinfold thickness. PCA was applied to correlation matrices based on all of these 14 items (complete case), on 13 items excluding body weight, and on 12 items excluding height and weight. The conclusions are :
1) Coefficients of determination in each component following PC2 were larger in weight excluding case than in complete case. Eigenvectors and factor loadings in the first three components were also larger in weight excluding case.
2) Excluding both height and weight, eigenvalues became lower while coefficients of determination in PC1 and PC2 became heigher than those in case excluding weight solely. For eigenvectors and factor loadings, all variables in each PC were more loaded in the same way. Therefore, the interpretation of each component became much more easy.
3) From these results, we conclude that height and weight are preferable to be excluded from the item selection in PCA. For the reason that both of them represent total body mass measurement having common information with other variables, they reduce the effect of interrelationship between others.
抄録全体を表示