The Canonical Correlation Analysis (CCA) estimates the correlation between two vector variables by maximizing the correlation of linear combinations of their respective components. Here, the CCA is used to find correlation patterns in the last five successive, per pairs, earthquakes ($M\ge 4.0$) preceding 271 main shocks ($M\ge 5.5$) that occurred in the Greek territory during 1964–2018. The vector variables have two components, the earthquake magnitude and interevent time. The statistical significance of CCA is determined by the standard parametric test along with two proposed randomization tests, one using random shuffling of each paired dataset and one using randomly selected pairs of successive earthquakes. Simulations were designed on synthetic data from vector variables having the statistical characteristics of the real observations. The results on uncorrelated variables showed the correct size for the two randomization tests but larger type I error for the parametric significance test for small sample size. For correlated variables, the test power was equally high for both test types. The application of CCA and the significance tests to the Greek seismicity evidence the significant correlation among the last five successive preshocks, proving to be a promising tool in an a posteriori short-term earthquake forecasting.

典型相关分析 (CCA) 通过最大化它们各自分量的线性组合的相关性来估计两个向量变量之间的相关性。在这里，CCA 用于查找最近五次连续每对地震中的相关模式（$米\ge 4.0$) 之前的 271 次主要震荡 ($米\ge 5.5$) 发生在 1964-2018 年期间在希腊境内。矢量变量有两个分量，地震震级和间隔时间。CCA 的统计显着性由标准参数检验以及两个建议的随机化检验确定，一个使用每个配对数据集的随机洗牌，另一个使用随机选择的连续地震对。对来自具有真实观察统计特征的矢量变量的合成数据设计了模拟。不相关变量的结果显示两个随机化检验的大小正确，但小样本量的参数显着性检验的 I 型误差较大。对于相关变量，两种测试类型的测试功效同样高。