I would to understand more on the multivariate and integral representations. I am fascinated by the possibility of a multivariate distribution like this one:

$$\mathbb{P}(X_1 \le Z + a, X_2 \le Z + a, ..., X_n \le Z + a),$$

to be transformed into:

$$\int \limits_{- \infty} ^{+ \infty} {\Phi(z+a)^n \phi(z)dz}.$$

and vice-versa. Here $\Phi(z)$ and $\phi(z)$ are, respectively, the cdf and the pdf. Everything started reading about this: Integrating pdf times cdf squared.

How is that a random vector is transformed into a scalar? is this some sort of projection? is there an analog transformation between a euclidean vector and a scalar?


1 Answer 1


Lets assume $Z$ is a random variable defined over a support of $\mathbb{R}$ and $X_1,...,X_n$ are i.i.d and have cdf $\Phi$, in which case

\begin{align} \text{Prob}(a) &\equiv \mathbb{P}(X_1 \le Z + a, X_2 \le Z + a, ..., X_n \le Z + a) \\[18pt] &= \mathbb{E}_Z[\mathbb{P}(X_1 \le Z + a, X_2 \le Z+ a, ..., X_n \le Z + a | Z = z)] \\[12pt] &= \int \limits_{-\infty}^{\infty} \mathbb{P}(X_1 \le Z + a, X_2 \le Z+ a, ..., X_n \le Z + a | Z = z) \phi(z) \,dz \\[6pt] &= \int \limits_{-\infty}^{\infty}\mathbb{P}(X_1 \le z + a, X_2 \le z + a, ..., X_n \le z + a) \phi(z) \,dz \\[6pt] &= \int \limits_{-\infty}^{\infty}\mathbb{P}(X_1 \le z + a) \cdot \mathbb{P}(X_2 \le z + a) \cdots \mathbb{P}(X_n \le z + a) \phi(z) \,dz \\[6pt] &= \int \limits_{-\infty}^{\infty}\Phi(z + a)^n \phi(z) \,dz. \\[6pt] \end{align}

Intuitively, $z$ (the variable of integration) is a scalar as opposed to a vector because it is the same variable that all $X_1,...,X_n$ are being compared with, and it allows us to compute the total probability utilizing the conditional probability given $Z=z$, which is much easier probability to compute. This alternative, easier formulation is allowed as per the law of total probability.

Observe that the integration variable would instead be the vector $(z_1,...,z_n$) if we were instead computing $\mathbb{P}(X_1 \le Z_1 + a, X_2 \le Z_2 + a, ..., X_n \le Z_n + a.)$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.