Suppose that the lifelengths ( in thousands of hours) of light bulbs are distributed Exponential($\theta$), where $\theta>0$ is unknown. If we observe $\overline x = 5.2$ for a sample of $20$ light bulbs, record a representative likelihood function. Why is it that we only need to observe the sample average to obtain a representative likelihood?

The likelihood is pretty straightforward to find: $$L(\theta \mid x_1,\ldots,x_{20})=\prod_{i=1}^{20}\theta e^{-\theta \overline x} = \theta^{20}e^{-20 \theta \overline x}$$

I am having an issue showing this is a sufficient statistic however, as I cannot seem to factor it in the form of $h(\overline x)g_\theta (T(\overline x))$


1 Answer 1


You've already done what you think you haven't done.

You should write it as $h(x_1,\ldots,x_{20}) g_\theta( (x_1 + \cdots + x_{20})/20),$ or in other words as $h(x_1,\ldots,x_{20}) g_\theta(T(x_1,\ldots, x_{20})).$

What's going on is camouflaged by the fact that $h(x_1,\ldots,x_{20}) = 1$ regardless of the value of $(x_1,\ldots,x_{20}).$

That often happens. (One situation where it does not happen is with an i.i.d. sample from the family of Poisson distributions.)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.