I have the following question about the Neyman Pearson Lemma. It is an example in my lecture notes of testing $H_0:\theta=\theta_0$ against $H_1:\theta=\theta_1$ for the exponential distribution.

The likelihoods are given by \begin{equation*} L(\theta_i) = \theta_i^{-n}\exp(-\sum x_i/\theta_i); \quad i=0,1 \end{equation*}

Therefore, from the NP lemma, reject $H_0$ if and only if \begin{equation*} \left(\frac{\theta_1}{\theta_0}\right)^n \exp \left[-\left(\frac{1}{\theta_0}-\frac{1}{\theta_1}\right)\sum x_i \right]\leq k \end{equation*} or equivalently when $\theta_1<\theta_0$ \begin{equation*} \sum x_i \leq \left(\frac{1}{\theta_1}-\frac{1}{\theta_0}\right)^{-1} \ln\left(\frac{\theta_0}{\theta_1}\right)^n k = k^* \end{equation*} where $k^*$ is a positive constant. Since $\sum x_i$ has a gamma$(\theta,n)$ distribution, $K$ can be set to give a particular size $\alpha$, using quantiles of the gamma distribution (this test can be based on $\bar{x}$).

The question is: what happens when $\theta_0<\theta_1$? I don't really see how this changes things ...


You reject $H_0$ iff $$ \left( \frac 1 {\theta_1} - \frac 1 {\theta_0} \right) \sum_i x_i \le \text{something}. $$ Multiplying both sides by the reciprocal of $\displaystyle\left( \frac 1 {\theta_1} - \frac 1 {\theta_0} \right)$, you get $$ \sum_i x_i \quad \left.\begin{cases} \le \\[10pt] \ge \end{cases}\right\} \quad \left( \frac 1 {\theta_1} - \frac 1 {\theta_0} \right)^{-1} \cdot\text{something}. $$

You get "$\le$" if the thing you multiplied by is positive, and "$\ge$" if it's negative.


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.