0
$\begingroup$

enter image description here

You may ignore the questions $a$, $b$ and $c$ as they are relatively easy. I know how to test hypotheses for a sample data but question $d$ is confusing me. How do I know what type of test statistic I should be using for this and what quantile distribution I should be comparing this with? We've only really dealt with testing hypotheses for sample distributions. Thanks in advance for any help.

$\endgroup$
5
  • $\begingroup$ It seems to me that this can be rephrased as: give a rule such that if $\theta=2$ then you reject the null hypothesis $\theta=2$ with probability 0.05. So this would amount to finding a symmetric interval centered around 2 whose complement has probability 0.05 under a Exp (2) distribution. This is easy enough to do... $\endgroup$
    – Ian
    Sep 18 '16 at 4:16
  • $\begingroup$ I found this interval to be -2log(0.05) to infinity. Is this correct? Although I'm not entirely sure what this interval means. $\endgroup$ Sep 22 '16 at 12:12
  • $\begingroup$ I said a symmetric one: so you're looking for $a$ such that $\int_{2-a}^{2+a} \frac{1}{2} e^{-t/2} dt = 0.95$. So $e^{-(2-a)/2}-e^{-(2+a)/2}=0.95$ so $\sinh(a/2)=\frac{0.95e}{2}$, $a=2\sinh^{-1}(0.95e/2)$. This breaks down because actually this $a$ is greater than $2$; so you're really looking for $1-e^{-x/2}=0.95$ so $x=-2\log(0.05)$. Then the relevant interval is $[0,-2\log(0.05)]$. $\endgroup$
    – Ian
    Sep 22 '16 at 12:20
  • $\begingroup$ To clarify what this means: if you have the null hypothesis $H_0 : \theta=2$, you will accept the null hypothesis if your result is in $[0,-2\log(0.05)]$ and you will reject it otherwise, with the overall level of significance of the test being 0.95 (i.e. the probability of rejecting the null when it holds is at most $0.05$). $\endgroup$
    – Ian
    Sep 22 '16 at 12:58
  • $\begingroup$ Thank you a lot, I understand it now $\endgroup$ Sep 22 '16 at 14:04
1
$\begingroup$

If you draw a single observation attempting to test $H_0 : \theta=2$, then the standard test is the same as it usually is for the mean: you reject this null hypothesis with level of significance $1-\alpha$ if your result falls outside of an interval $[2-\delta,2+\delta]$ which has probability $1-\alpha$ when $\theta=2$. The probability of such an interval is $\int_{2-\delta}^{2+\delta} \frac{1}{2} e^{-t/2} dt=e^{-(2+\delta)/2}-e^{-(2-\delta)/2}=2e^{-1}\sinh(\delta/2)$ if $0 \leq \delta<2$ and $\int_0^{2+\delta} \frac{1}{2} e^{-t/2} dt = 1-e^{-(2+\delta)/2}$ if $\delta \geq 2$. If $\alpha=0.05$, you note that $2e^{-1}\sinh(1)<0.95$, so that you look at the second case. This gives the interval as $[0,-2\log(0.05)] \approx [0,6]$. You accept this null hypothesis given a result in this interval and reject it otherwise.

$\endgroup$

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.