Let $X_1, ... X_n \stackrel{i.i.d} \sim Unif(0,2).$ Find $P(Y_1 < \frac12 < Y_n),$ where $Y_1 = \min \{X_1, ..., X_n\} $ and $ \ Y_n = \max\{X_1, ..., X_n\}$

When I attempted this problem, I decided that $n>2 \implies Y_1 $ and $ Y_n$ are independent, which leads me to the answer

$$P(Y_1 < \frac12 < Y_n) = \left(1-\left(\frac14\right)^n\right) \cdot \left(1-\left(\frac34\right)^n\right)$$

I have since decided is untrue. Clearly, if I know $Y_n = 1,$ then $Y_1 \neq 1.5.$ I have also discovered that in my notes there is a joint pdf of two order statistics of a given sample:

$$f_{Y_i,Y_j}(u,v) = \cfrac{n!} {(i-1)!(j-i-1)!(n-j)!} [F_X(u)]^{i-1} [F_X(v)-F_X(u)]^{j-i-1} [1-F_X(v)]^{n-j} f_X(u)f_X(v), \forall \ u<v$$

I am looking first of all for the derivation and perhaps some intuition for this equation. For context, I am familiar with the derivation of the cdf/pdf of a single order statistic, but this one is throwing me for a loop. Second of all, I would like to know if I am using this equation correctly:

$$P(Y_1<\frac12<Y_n) = P(Y_1 < \frac12 \ \bigcap \ Y_n > \frac12) \\ = \int_0^\frac12 \int_\frac12^2 \cfrac{n!}{0!(n-2)!0!} \left[\frac u2\right]^{0} \left[\frac v2 - \frac u2 \right]^{n-2} \left[1- \frac v2 \right]^{0} \left(\frac12 \right) \left(\frac12 \right)\mathrm{d}v\mathrm{d}u \\ = \frac14 \int_0^\frac12 \int_\frac12^2 n(n-1) \left[\cfrac{v-u}2 \right]^{n-2} \mathrm{d}v\mathrm{d}u$$

which I don't know how to solve explicitly, but is still useful as it can be solved given a value of $n$.

  • 2
    $\begingroup$ Clearly $Y_1$ and $Y_n$ are not independent, since $Y_1$ cannot exceed $Y_n$. $\endgroup$ – Glen_b Apr 22 '20 at 5:40
  • 1
    $\begingroup$ The integral you have set up in the end misses the $u<v$ restriction. If this is some sort of homework, please add self-study tag. $\endgroup$ – StubbornAtom Apr 22 '20 at 6:31
  • $\begingroup$ See stats.stackexchange.com/questions/225867 for a more complicated version of this problem, because its solution easily applies here. Concerning the issue of independence, see stats.stackexchange.com/questions/208658. A formula for the joint pdf is given at stats.stackexchange.com/a/69582/919 and, in the special case of a uniform distribution, at stats.stackexchange.com/a/207596/919. $\endgroup$ – whuber Apr 22 '20 at 13:46
  • $\begingroup$ @StubbornAtom How does my integral miss the restriction? No $u$ value is above any $v$ value in the integration. Also, this is a former exam question not assigned to me in any capacity - let me know if that still qualifies it as self-study $\endgroup$ – jeremy909 Apr 22 '20 at 15:42
  • 1
    $\begingroup$ Yes, all exam questions etc. qualify as self-study: link through to the tag wiki for details. As far as computing your integral goes, it's far easier than it might look: you're just integrating polynomials, after all. $\endgroup$ – whuber Apr 22 '20 at 16:25

I don't care to evaluate the integral mechanically (although Wolfram Alpha will). And nor should anyone else, because the pedagogical usefulness of that exercise is doubtful.

Instead, considering that the event $Y_{1} \le 1/2 \le Y_{n}$ is the complement of the union of events

$$\left(Y_{1}\gt 1/2\right)\ \cup\ \left(Y_{n}\lt 1/2\right)$$

and, because $Y_1$ cannot exceed $Y_n$ those events are disjoint, it follows that

$$\Pr(Y_{1}\le 1/2\le Y_{n}) = 1 - \Pr(Y_{1} \gt 1/2) - \Pr(Y_{n} \lt 1/2).$$

These probabilities are simple to evaluate because the first is the chance that all the $X_i$ exceed $1/2,$ the second is the chance that all the $X_i$ are less than $1/2,$ and the $X_i$ are independent.

Even if your interest is only on whether you are doing the integration correctly, this approach provides a way to check your answer.


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.