0
$\begingroup$

I have a stochastic variable with binomial distribution $T\sim B(n,\frac{1}{4})$

I need to estimate $n : P(0.30 < \frac{T}{n} < 0.70) > 0.95$

I have tried to resolve this problem with

$P(0.30 < \frac{T}{n} < 0.70) = P(T <0.70n) - P(T < 0.30n)$

but I don't know how can I proceed because I don't know the $\operatorname{Img}(T)$

$\endgroup$
1
$\begingroup$

No such $n$ exists. For sufficiently "large" $n$, we can use a normal approximation to the binomial and show that the function $$R(n) = \Phi\left(\frac{9\sqrt{n}}{5\sqrt{3}}\right) - \Phi\left(\frac{\sqrt{n}}{5\sqrt{3}}\right)$$ where $\Phi(z) = \Pr[Z \le z]$ is the standard normal CDF, has derivative $$r(n) = \frac{e^{-27n/50}(9-e^{8n/15})}{10\sqrt{6\pi n}}$$ with critical point at $n = \frac{15}{4} \log 3 \approx 4.1198$. This is a global maximum on $n > 0$, but $$R(4.1198) \approx 0.389891 < 0.95,$$ with very little hope of the exact binomial probability getting that large. But we make a table anyway:

$$\begin{array}{c|c|c|c} n & \lceil 0.3n \rceil & \lfloor 0.7n \rfloor & \Pr[0.3n < T < 0.7n] \\ \hline 1 & 1 & 0 & 0 \\ 2 & 1 & 1 & \frac{3}{8} \\ 3 & 1 & 2 & \frac{9}{16} \\ 4 & 2 & 2 & \frac{27}{128} \\ 5 & 2 & 3 & \frac{45}{128} \\ 6 & 2 & 4 & \frac{945}{2048} \\ 7 & 3 & 4 & \frac{945}{4096} \\ 8 & 3 & 5 & \frac{10395}{32768} \\ \end{array}$$

The maximum probability occurs at $n = 3$, but $9/16 < 0.95$.

$\endgroup$
3
  • $\begingroup$ +1 though I would prefer the final column of that table to be as decimal fractions $\endgroup$ – Henry Feb 24 at 1:10
  • $\begingroup$ Why you use a normal approximation and not the Poisson approximation with parameter $np = \frac{n}{4}$? $\endgroup$ – Simone Feb 24 at 9:11
  • $\begingroup$ @Simone It doesn’t matter. Either approximation will show that the desired probability is not attainable. $\endgroup$ – heropup Feb 24 at 9:51

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.