1
$\begingroup$

I have two following problems about the hypothesis testing about the normal distribution:

I. A random sample of size $n$ from a normal population with unknown mean and variance is to be used to test the null hypothesis $\mu=\mu_0$ against the alternative $\mu\ne\mu_0$. Using the simultaneous maximum likelihood estimates of $\mu$ and $\sigma^2$, show that the values of the likelihood ratio statistic can be written in the form:
$$\lambda=\Big(1+{{t^2}\over{n-1}}\Big)^{-{n\over2}},$$where $t={{\bar x-\mu_0}\over{s/\sqrt n}}$.

II. Given a random sample of size $n$ from a normal population with unknown mean and variance, find an expression for the likelihood ratio statistic for testing the null hypothesis $\sigma=\sigma_0$ against the alternative hypothesis $\sigma\ne\sigma_0$

$\endgroup$
  • $\begingroup$ Would you please give me a hint about where to start? I just have no idea where to start. $\endgroup$ – Scorpio19891119 Mar 1 '13 at 15:48
  • 1
    $\begingroup$ Have you found the MLEs? $\endgroup$ – Michael Hardy Mar 1 '13 at 15:50
  • $\begingroup$ @MichaelHardy yes. $\endgroup$ – Scorpio19891119 Mar 1 '13 at 15:52
1
$\begingroup$

The likelihood function is $$ \begin{align} L(\mu,\sigma) & = \text{constant} \cdot\prod_{i=1}^n \left(\sigma^{-1} \exp\left(\frac{-1}{2}\left(\frac{x_i-\mu}{\sigma}\right)^2\right)\right) \\[8pt] & = c\sigma^{-n}\exp\left(\frac{-1}{2\sigma^2}\sum_{i=1}^n(x_i-\mu)^2\right) \\[8pt] & = c\sigma^{-n}\exp\left(\frac{-1}{2\sigma^2}\left(n(\bar x - \mu)^2+\sum_{i=1}^n(x_i-\bar x)^2 \right)\right) \\[8pt] & = c\sigma^{-n}\exp\left(\frac{-1}{2\sigma^2}\left(n(\bar x - \mu)^2+ns^2 \right)\right). \end{align} $$ Evaluating this at the MLEs $\hat\mu=\bar x$ and $\hat\sigma = s$, we have $$ L(\bar x, s) = cs^{-n}\exp\left(\frac{-1}{2s^2}ns^2\right) = cs^{-n}\exp\left(\frac{-n}{2}\right) . $$ Evaluating it at the hypothesized value $\mu_0$, we have $$ L(\mu_0,s) = cs^{-n}\exp\left(\frac{-1}{2s^2} \left(n(\bar x - \mu_0)^2+ ns^2 \right)^2\right) = cs^{-n}\exp\left(\frac{-n}{2s^2} \left((\bar x - \mu_0)^2+ s^2 \right)\right) $$ So we get a likelihood ratio: $$ \frac{L(\bar x, s)}{L(\mu_0,s)} = \exp\left(n\left(\frac{\bar x - \mu_0}{s}\right)^2\right) = \exp(t^2). $$ This is a monotone function of $|t|$, so one rejects the null hypothesis if $|t|$ is too big.

I'll let you work on further details.

$\endgroup$

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.