8
$\begingroup$

I have a problem with the following exercise from Geoffrey G. Grimmett, David R. Stirzaker, Probability and Random Processes, Oxford University Press 2001 (page 155, ex. 6):

Let $X$ have the binomial distribution bin($n$, $U$), where $U$ is uniform on $(0,1)$. Show that $X$ is uniformly distributed on $\{0,1,\ldots,n\}$.

So far, what I have is this: $$P(X=k | U) = {n \choose k} U^k (1-U)^{n-k}$$ $$P(X=k) = \int_0^1 {n \choose k} u^k (1-u)^{n-k} f_U(u) \text{ d}u$$

where $f_U(u)$ is density function of random variable $X$. Of course, $f_U(u) = 1$ on $u\in (0,1)$.

$$P(X=k) = {n \choose k} \int_0^1 u^k (1-u)^{n-k} \text{d}u$$ And I don't know what to do next... How to calculate this integral? Am I solving it right so far?

$\endgroup$
1
  • $\begingroup$ For the record, and to make this easier to find for future generations: this is a so-called compound distribution, which is confusingly also known as a mixture distribution. $\endgroup$ Jun 17, 2020 at 12:40

3 Answers 3

9
$\begingroup$

A way to avoid using pre-knowledge about Beta integrals (for a more conceptual explanation, see the second part of this post) is to compute the generating function of $X$, that is, $$ \mathbb E(s^X)=\sum_{k=0}^ns^k\mathbb P(X=k)=\int_0^1\sum_{k=0}^n\binom{n}ku^k(1-u)^{n-k}s^k\mathrm du. $$ By the binomial theorem, $$ \sum_{k=0}^n\binom{n}k(su)^k(1-u)^{n-k}=(1-(1-s)u)^n, $$ hence $$ \mathbb E(s^X)=\int_0^1(1-(1-s)u)^n\mathrm du\stackrel{[v=1-(1-s)u]}{=}\frac1{1-s}\int_s^1v^n\mathrm dv=\frac{1-s^{n+1}}{(n+1)(1-s)}, $$ that is, $$ \mathbb E(s^X)=\frac1{n+1}\sum_{k=0}^ns^k. $$ This formula should make apparent the fact that $X$ is uniform on $\{0,1,2,\ldots,n\}$...


...But the "real" reason why $X$ is uniform might be the following.

First, the distribution of a sum of i.i.d. Bernoulli random variables is binomial. Second, if $V$ is uniform on $[0,1]$, the random variable $\mathbf 1_{V\leqslant u}$ is Bernoulli with parameter $u$. Hence, if $(U_i)_{1\leqslant i\leqslant n}$ is i.i.d. uniform on $[0,1]$, the random variable $\sum\limits_{i=1}^n\mathbf 1_{U_i\leqslant u}$ is binomial with parameter $(n,u)$.

Thus, $X$ may be realized as $X=\sum\limits_{i=1}^n\mathbf 1_{U_i\leqslant U_{n+1}}$ where $(U_i)_{1\leqslant i\leqslant n+1}$ is i.i.d. uniform on $[0,1]$. The event $[X=k]$ occurs when $U_{n+1}$ is the $(k+1)$th value in the ordered sample $(U_{(i)})_{1\leqslant i\leqslant n+1}$. By exchangeability of the distribution of $(U_i)_{1\leqslant i\leqslant n+1}$, $U_{n+1}$ has as much chances to be at each rank from $1$ to $n+1$. This fact means exactly that $X$ is indeed uniform on $\{0,1,2,\ldots,n\}$.

$\endgroup$
3
  • 1
    $\begingroup$ That's a cool solution! $\endgroup$
    – Alex
    Jan 19, 2013 at 23:53
  • $\begingroup$ The limits for your integration after the variable substitution don't totally make sense to me. if $v=1-(1-s)u$, then since the original lower limit was $u=0$, then plugging that in gives $v=1$ for the lower limit and since the original upper limit was $u=1$, then plugging that in gives $v=s$ for the upper limit, right? $\endgroup$
    – makansij
    Nov 5, 2018 at 1:01
  • 1
    $\begingroup$ @Hunle Except that $s<1$ hence $1$ is the upper limit, not the lower one, and $s$ is the lower limit, not the upper one. $\endgroup$
    – Did
    Nov 5, 2018 at 7:19
2
$\begingroup$

The integral $$\int_0^1 x^\alpha (1 - x)^\beta dx$$ has a well-know representation in terms of $\Gamma$ functions. In the integer case, it boils down to factorials.

$\endgroup$
2
$\begingroup$

Hint: Integrating by parts:

$$\int_0^1 u^k (1-u)^{n-k} \text{d}u = \frac{n-k}{k+1}\int_0^1 u^{k+1} (1-u)^{n-k-1} du$$
so repeat until the exponent of $(1-u)$ reduces to $0$ and you have $$\frac{1}{n \choose k} \int_0^1 u^{n} \text{d}u$$

$\endgroup$

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.