回归诊断的可视化分析

Visual Analysis of Regression Diagnosis

传统意义上,我们进行影响分析都是利用探测统计量来识别离群点和强影响点,而面对样本容量很大时,计算探测统计量要花费大量的时间,并且分析结果也不一定准确。因此,有必要提出可视化影响分析方法来补充回归诊断中探测离群点和强影响点方法的单一性和繁琐性。而且子集变化分析更是让影响分析的回归诊断过程变得更加严密精确。

In doing influence analysis, traditionally, we use detection statistics to identify outliers and influential points. For a large sample, however, it takes lots of time to calculate the detection statistics, and their resuits are not right, sometimes. So it is necessary to visualize influence analysis in order to supplement the single and complex method of detecting the outliers and influential points. And change of subsets will make the influence analysis of regression diagnosis more precise and tighter.