DECREASE OF EIGENVALUES IN PRINCIPAL COMPONENT ANALYSIS - COMPARISON WITH BROKEN STICK MODEL
DECREASE OF EIGENVALUES IN PRINCIPAL COMPONENT ANALYSIS - COMPARISON WITH BROKEN STICK MODEL
复制标题
DOI：
10.1016/0022-0981(76)90076-9
复制
发表时间：
1976-01-01
影响因子：
2
通讯作者：
FRONTIER, S
中科院分区：
文献类型：
--
作者：
FRONTIER, S
本文比较了20个变量44个个体的主成分分析中连续特征值的减小与将固定长度随机折断成若干段的断棒模型的减小。在实际分析中，第21个特征值明显高于模型值，表明存在作用于变量的因素。根据接近断棒的模型，残差方差似乎在其他轴之间划分。调整并不完美，因为随机方差的未知部分也有助于第一个特征值。该模型允许经验确定显著向量的数量。
The decrease of successive eigenvalues in some principal component analysis of 20 variates and 44 individuals, is compared with that of the broken stick model, where a fixed length is broken at random into a number of segments. In the real analysis, the 2 1st eigenvalues stay distinctly above the model values, indicating the occurrence of factors acting on the variables. Residual variance seems partitioned between other axes, according to a model which approaches the broken stick. The adjustment is not perfect, since an unknown part of the random variance also contributes to the 1st eigenvalues. The model allows an empirical determination of the number of significant vectors.