Difficult confidence interval problem

1.1k Views Asked by At

just needed some help on the following question. I've already attempted it twice but still can't get the answer out...

Let $X_i,$ $i=1,...,n$ denote a random sample of size n from a population with a uniform distribution on the interval $(0,\theta)$. Let $X_{(n)}=max{(X_1,...,X_n)}$ and define $U=(1/\theta)X_{(n)}$.

Part one required us to show $F_U(u)=u^n$ if $0\le u\le1$ which I managed to do.

Part two: Because the distribution of U does not depend on $\theta$, U is a pivotal quantity. Find the 95% lower confidence bound for $\theta$.

My attempt: I did a bivariate transformation with $U_1=\frac{X_{(n)}}{U}$ and $U_2=U$ and obtained a joint density function $$f_{U_1,U_2}(u_1,u_2)=n^2(\frac{u_1u_2}{\theta})^{n-1}\frac{1}{\theta}u_2^n$$

Then integrated over the range of $U_2$ to get the marginal density of $U_1$ $$f_{U_1}(u_1)=\frac{nu_1^{n-1}}{\theta^n}$$

I then integrated from 0 to a for the density function of $U_1=\frac{X_{(n)}}{U}=\theta$, equated this to 0.95 and came up with $$a=(1.90)^{1/n}\theta$$

However I'm pretty sure this is wrong. Any help would be much appreciated! Thanks!

1

There are 1 best solutions below

0
On BEST ANSWER

You are making this question much more difficult than it actually is. If we wish to be $100(1-\alpha)\%$ confident that $\theta \in [L, \infty)$, then $L$ must satisfy $$\Pr[\theta \ge L] = 1-\alpha;$$ equivalently, $$1-\alpha = \Pr\biggl[\frac{X_{(n)}}{L} \ge \frac{X_{(n)}}{\theta}\biggr] = \Pr\biggl[U \le \frac{X_{(n)}}{L}\biggr] = \biggl(\frac{X_{(n)}}{L}\biggr)^{\!n}.$$ Note this immediately ensures that $L \ge X_{(n)}$. Solving for $L$ then yields $$L = \frac{X_{(n)}}{(1-\alpha)^{1/n}}.$$