跳转至

相关系数置信区间

假设总体相关系数用 \(r\) 表示,经 Fisher's z 转换后用 \(z_r\) 表示:

\[ z_r = \operatorname{arctanh}r = \frac{1}{2} \ln{\frac{1+r}{1-r}} \]

则样本相关系数 \(\hat{r}\) 经 Fisher's z 转换后,近似服从正态分布:

\[ z_\hat{r} \sim N\left(z_r, \frac{1}{n-3}\right) \]

通过计算 \(z_\hat{r}\) 的置信区间,并利用以下公式将置信限反转,可得到样本相关系数 \(\hat{r}\) 的置信区间:

\[ r = \operatorname{tanh}r = \frac{e^{2 z_r} - 1}{e^{2 z_r} + 1} \]

以下仅给出 \(z_{\hat{r}}\) 的置信限。

未校正偏倚

\[ \begin{align} L & = z_{\hat{r}} - z_{1-\alpha/2} \sqrt{\frac{1}{n-3}} \\ U & = z_{\hat{r}} + z_{1-\alpha/2} \sqrt{\frac{1}{n-3}} \end{align} \]

置信区间宽度:

\[ d = \operatorname{tanh}U - \operatorname{tanh}L \]
\[ \begin{align} L & = z_{\hat{r}} - z_{1-\alpha} \sqrt{\frac{1}{n-3}} \\ U & = 1 \end{align} \]

从相关系数到置信下限的距离:

\[ d = r - \operatorname{tanh}L \]
\[ \begin{align} L & = -1 \\ U & = z_{\hat{r}} + z_{1-\alpha} \sqrt{\frac{1}{n-3}} \end{align} \]

从相关系数到置信上限的距离:

\[ d = \operatorname{tanh}U - r \]

校正偏倚

未校正偏倚 的基础上加入校正项 \(\frac{r}{2(n-1)}\),即:

\[ z_\hat{r} + \frac{\hat{r}}{2(n-1)} \sim N\left(z_r, \frac{1}{n-3}\right) \]
\[ \begin{align} L & = z_{\hat{r}} - \frac{\hat{r}}{2(n-1)} - z_{1-\alpha/2} \sqrt{\frac{1}{n-3}} \\ U & = z_{\hat{r}} - \frac{\hat{r}}{2(n-1)} + z_{1-\alpha/2} \sqrt{\frac{1}{n-3}} \end{align} \]

置信区间宽度:

\[ d = \operatorname{tanh}U - \operatorname{tanh}L \]
\[ \begin{align} L & = z_{\hat{r}} - \frac{\hat{r}}{2(n-1)} - z_{1-\alpha} \sqrt{\frac{1}{n-3}} \\ U & = 1 \end{align} \]

从相关系数到置信下限的距离:

\[ d = r - \operatorname{tanh}L \]
\[ \begin{align} L & = -1 \\ U & = z_{\hat{r}} - \frac{\hat{r}}{2(n-1)} + z_{1-\alpha} \sqrt{\frac{1}{n-3}} \end{align} \]

从相关系数到置信上限的距离:

\[ d = \operatorname{tanh}U - r \]