I have the following statistic of which I want to compute confidence intervals: \begin{equation} Z = \frac{(X-Y)^{-1}}{\mathbb{PVE}[(X-Y)^{-1}]}, \end{equation} where \begin{align} X &\sim\mathcal{N}\left(\sigma_{1}^{2},\frac{2\sigma_{1}^{4}}{n-1}\right),\\ Y &\sim\mathcal{N}\left(\sigma_{2}^{2},\frac{2\sigma_{2}^{4}}{m-1}\right), \end{align} $n$ and $m$ are the sample sizes used to compute $X$ and $Y$ and $\mathbb{PVE}[(X-Y)^{-1}]$ is the Cauchy principal value for the expected value of $(X-Y)^{-1}$ (it's just a constant) defined as \begin{equation} \mathbb{PVE}[(X-Y)^{-1}] = \frac{\sqrt{2}}{\sigma_{X-Y}}\,\mathcal{D}_{+}\left(\frac{\mu_{X-Y}}{\sqrt{2}\sigma_{X-Y}}\right), \end{equation} where $\mathcal{D}_{+}(z)$ is the dawson function. I have parametric equations for the pdf/cdf of $Z$. Computing exact confidence intervals seems completely out of the question.

For my application, very large sample sizes are the norm (i.e. $n$ and $m$ on the order of several thousands). It is also typical for $\mathrm{Pr}[(X-Y)\leq0]\approx 0$. How can I go about computing confidence intervals of $Z$? What options do I have in this specific application?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.