I'm reading the paper by Lourme A. et al (2016) and try to plot the Fig.2 from this paper. The question regarding to programming of confidence areas and my example in R is here.

4.1.2. Confidence areas

Let $u = (u_1,…,u_d)^\top$ be a random vector with uniform margins on $(0, 1)$ and let $y = (y_1,…, y_d)^\top \in \mathbb{R}^d$ be defined by: $y_j = G(u_j)$, $j \in \{1,…,d\}$ where $G$ is a continuous increasing map defined between 0 and 1.

Let us assume that the multivariate c.d.f of $u$ is a Gaussian copula with parameter $R_g$. Taking the quantile function of $N_1(0, 1)$ as $G$, then $y$ is distributed as $N_d(0, R_g)$ and the random variable $y^\top R_g^{-1} y$ as $\chi^2_d$. Given $\alpha \in (0, 1)$, the latter variable $y^\top R_g^{-1} y$ is less than $\chi^2_{d,\alpha}$ – the $\alpha$ order quantile of $\chi^2_d$ – with probability $\alpha$. Hence, $$\Gamma_g(\alpha) = \{ v= (v_1 ,…,v_d )^\top \in [0, 1]^d :\\ (G(v_1),…, G(v_d)) R_g^{-1}(G(v_1),…, G(v_d))^\top \leq\chi^2_{d,\alpha}\}$$ is a $d$-dimensional compact confidence area inside (resp. outside) of which $u$ falls with probability $\alpha$ (resp. $1-\alpha$).

$R_g$ is a $d × d$ correlation matrix.

Question. I have statistical questions about distribution.

How to statiscally check whether variable $y^\top R_g^{-1} y$ is distributed as $\chi^2_d$?

It's known that mean $M(\chi^2_d)=d$, and variance $D(\chi^2_d)=2d$, where d is degreee of freedom.

Edit. Is it possible to use the Kolmogorov-Smirnov test here?


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.