Some Statistical Methods for Bivariate Circular Data
提出用两个圆形单位向量叉积矩阵的最小奇异值函数作为相关系数度量，并引入聚类依赖概念，发现一个单位向量的预测取决于依赖类型，最后刻画了包裹双变量正态分布。
Summary A function of the smallest singular value of the cross product matrix between two circular unit vectors is suggested as a measure of correlation. It possesses most of the desirable properties for a correlation coefficient given by Jupp and Mardia (1980). The concept of cluster dependence is introduced. A good predictor of one unit vector given the other is shown to depend on the type of dependence observed. The wrapped bivariate normal is characterized.