0
$\begingroup$

I have the following question: Find the best size $\alpha$ test of $H_0:\theta=\theta_0$ vs $H_1:\theta =\theta_1$, with $\theta_1>\theta_0$, write down the expression for the power function when $X_1,\dots,X_n$ are IID exponential random variables, with parameter $\theta$.

Use the $\chi^2$ tables to find the least sample size which will allow us to test $H_0:\theta=1$ against $H_1:\theta=3$ with $\alpha=5\%$ and $\beta\leq 10\%$. Describe the appropiate critical region numerically.

I have managed to find the best test size in the following way:

First of all, finding the likelihood ratio:

$\begin{gather}\frac{f_1(x|\theta_1)}{f_0(x|\theta_0)}= \left(\frac{\theta_1}{\theta_0}\right)^n \cdot e^{-(\sum x_i)(\theta_1-\theta_0)} \end{gather}$.

Then, for the sufficient statistic $T = \sum X_i$, we condition that the likelihood ratio be bigger than $k(\alpha)$. Computing the inequality, we get that we need to have $T<c_s$, where $c_s=\frac{n(ln\theta_1 - ln\theta_0)-lnk(\alpha)}{\theta_1-\theta_0}$.

Now, we know further that $T\sim Gamma(n,\theta)$. Hence, by Neyman-Pearson Lemma we have that :

$\begin{gather}\alpha = \int_{c_s}^{\infty}\frac{1}{\Gamma(n)}t^{n-1}e^{-t}dt \end{gather}$

We can compute $c_s$. However, I do not know what to do for the next part.

$\endgroup$
1
  • $\begingroup$ Rejection region is $T<c$ but not sure how you got the $c$. It is found subject to the size restriction $P_{\theta=\theta_0}(T<c)=\alpha$, where you get $c$ in terms of chi-square quantiles. Note that if $X_i$s are iid Exp with mean $1/\theta$ then $2\theta X_i$s are iid $\chi^2_2$, so that $2\theta T\sim \chi^2_{2n}$. The power function is given by $P_{\theta}(T<c)$. $\endgroup$ Jul 1 '20 at 6:48
1
$\begingroup$

As you calculated, we Reject $\mathcal{H}_0$ if $\sum_i X_i <c^*$

where $c^*$ is calculated by

$$\int_0^{c^*}\frac{\theta_0^n}{\Gamma(n)}t^{n-1}e^{-\theta_0 t}dt=\alpha$$

Then the power of the test is, by definition,

$$\mathbb{P}[\text{Reject }\mathcal{H}_0|\mathcal{H}_1]$$ that is

$$\gamma=\int_0^{c^*}\frac{\theta_1^n}{\Gamma(n)}t^{n-1}e^{-\theta_1 t}dt$$

Now, considering the second part, we have the following system to verify

$$ \begin{cases} \mathcal{H}_0: \theta=1 \\ \mathcal{H}_1: \theta=3 \end{cases}$$

We know that $\sum_i X_i\sim Gamma(n;\theta)$ that means $2\theta\sum_i X_i\sim \chi_{(2n)}^2$

Considering the text request, we must search the minimum $n$ such that $\alpha=5\%$ with a power $\gamma=1-\beta \geq 90\%$

In other words we are looking for the following condition to be satisfied:

$$ \begin{cases} \mathbb{P}[2\sum_i X_i<c_1]=0.05 \\ \mathbb{P}[6\sum_i X_i<c_2] > 0.90 \end{cases}$$

By simply scrolling the $\chi^2$ table, we have to find the first row where the 90° percentile is more than 3 times the 5° percentile. This row is the one corresponding to the 16 degrees of freedom:

$$F_{(16)}^{-1}(0.05)=7.96$$

$$F_{(16)}^{-1}(0.90)=23.50$$

and $7.96\times3=23.88$

So the minimum sample size is 8 which gets a test with $\alpha=5\%$ and a power $\gamma \approx 91\%$

$\endgroup$

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.