I've got a one major problem using the Neyman-Pearson lemma.

We're testing hypotheses $H_0: \theta \le \theta_0$ vs. $H_1: \theta > \theta_0$. Our $$f(x,\theta) = \frac{1}{\theta}e^{-\frac{x}{\theta}}$$

We use the Neyman-Pearson lemma and get the following expression for the critical value of the test: $$\phi(X) = \mathbf{1}_{\{f(x,\theta_1) > k f(x,\theta_0)\}}(x) + \gamma \mathbf{1}_{\{f(x,\theta_1) = k f(x,\theta_0)\}}(x)$$

Now I can get some expression in terms of $\frac{1}{n} \sum_{i=1}^n X_i > g(k)$ for the test, but I need a value for $k$ from the expression for $\phi(X)$.

How would I compute $$\mathbb{E}[\phi(X) |H_0] =\alpha$$ so that I can solve for $k$?


1 Answer 1


Note that $\frac{1}{n} \sum_{i=1}^n X_i > g(k) \implies \sum_{i=1}^n X_i >n g(k)=c(\text{say})$

Consider your $\phi(X)=\begin{cases}1 & \text{if}\,\,\sum x_i>c \\ 0 & \text{otherwise} \end{cases}$.

Since $\sum X_i$ is coninuous R.V. we have $P[\sum X_i=c]=0$.

Then $c$ is determined from $$\begin{align} \alpha &= \mathbb{E}[\phi(X) |H_0] \\ &=1\cdot P_{H_0}[\sum X_i>c]\\&=P_{H_0}[\dfrac{2\sum X_i}{\theta}>\dfrac{2c}{\theta_0}]\\&=P[\chi^2_{2n}>\dfrac{2c}{\theta_0}] \end{align}$$

  • $\begingroup$ Does the distribution of $X$ make any difference in the conversion to chi-squared? $\endgroup$
    – shimee
    Jun 19, 2013 at 19:07
  • $\begingroup$ @shimee: Yes it will. $\endgroup$
    – A.D
    Jun 19, 2013 at 19:12

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.