2
$\begingroup$

Suppose we have independently distributed $X_i \sim \text{Uniform}(0,a_i)$ where the $a_i>0$ are fixed numbers. I want to obtain the probability that $X_j=X_{n-i+1,n}$ where $X_{n-i+1,n}$ is the $n-i+1$-th order statistic of the sample $X_1,...,X_n$.

Obviously, $\{X_i\}_{i=1}^n\overset{d}{=}\{a_i U_i\}_{i=1}^{n}$ where $U_i$ are i.i.d standard uniform random variables. Is there a smarter way to calculate this probability other than integrating (and summing) over sets of the form $\{a_iU_i> a_jU_j>... \}$?

$\endgroup$
1
$\begingroup$

Denote $X_{(1)}$ as any first observation from the original sample, $X_{(2)}$ as any second observation etc. We can calculate the probability as follows: \begin{align}P(X_j&=X_{n-i+1,n})\\ &=\mathbb{E}(P(X_{(1)}\leq X_j,...,X_{(n-i)} \leq X_j, X_{(n-i+1)}>X_j,...,X_{(n-1)}>X_j \text{ for all possible combinations() }|X_j))\\ &=\int_{0}^{a_j} \frac{1}{a_j} \sum_{\text{any combination}}^{n-1 \choose n-i}\prod_{i=1}^{n-i} \frac{x}{a_i} \prod_{i=1}^{i-1} \frac{1-x}{a_i}dx\\ &=\left(\sum_{\text{any combination}}^{n-1 \choose n-i}\prod_{i=1}^{n-i} \frac{1}{a_i} \prod_{i=1}^{i-1} \frac{1}{a_i} \right)\int_{0}^{a_j} \frac{1}{a_j}x^{n-i}(1-x)^{i-1} dx \\ &=\left({n-1 \choose n-i}\prod_{i=1}^{n} \frac{1}{a_i} \right)\int_{0}^{a_j} x^{n-i}(1-x)^{i-1} dx \end{align}

$\endgroup$
-1
$\begingroup$

I believe the $i-th$ order statistic of a uniform sample follows a $Beta(i, n-i+1)$ distribution.

Proof is likely integrating those sets you mention once so to show what the density ought to be.

$\endgroup$
2
  • 1
    $\begingroup$ You are answering a different question: your assertion holds only for identically distributed uniform$(0,1)$ variables. $\endgroup$ – whuber Nov 30 '20 at 18:53
  • $\begingroup$ Indeed, you are right. Woops $\endgroup$ – Three Diag Dec 1 '20 at 3:34

Your Answer

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

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