本文研究了线性同归模型的影响分析中常用的三种重要的诊断统计量,即似然距离LDi(β,σ2),LDi(β,α2)和LDi(σ2|β)之间的关系.利用对数函数In(1+X)的马克劳林展开,在非线性最小二乘回归模型中得到了LDi(β,σ2)≈LDi(β|σ2)+LDi(σ2|β),类似的近似关系式对包括非线性分位点同归在内的一类较为广泛的统计模型都成立一个线性和一个非线性数据实例的分析结果很好地说明了我们的结论.
In this article, the relationship among the three commonly used and important diagnostic statistics,LDi(β,σ2),LDi(β,α2) and LDi(σ2|β), in the influence analysis of linear regression model is studied. According to the Maclaurin expansion of the function ln(1 + x), the approximate equality, LDi(β,σ2)≈LDi(β|σ2)+LDi(σ2|β), is obtained in nonlinear least-square regression. The results are well shown by the analysis of a linear and a nonlinear real examples.