2
$\begingroup$

I would like some help with calculating the Fisher Information $I_o(\beta)$ and the expected information for a gamma distribution defined by

\begin{align*} f_X(x) = \frac{\beta^\alpha x^{\alpha - 1}e^{-\beta x}}{\Gamma(\alpha)} \; x > 0, \alpha >0, \beta > 0 \end{align*}

Where $\alpha$ is a known value and $\beta$ is the parameter of interest.

Attempt

I have attempted to calculate a likelihood function as follows:

\begin{align} L(\beta | X_i) &= \prod_{i = 1}^{n}f(x_i | \alpha, \beta) \\ &= \prod_{i = 1}^{n}\left( \frac{\beta^{\alpha}}{\Gamma(\alpha)}x_{i}^{\alpha - 1}\mathrm{exp}{\{-\beta x\}}\right) \\ &= \left(\frac{\beta^{\alpha}}{\Gamma(\alpha)}\right)^{n}\prod_{i = 1}^{n} \left(x_{i}^{\alpha-1}\right) \mathrm{exp}\{-\beta\sum_{i = 1}^{n} x_i\} \\ L(\beta | X_i) &= \left(\frac{\beta^{\alpha}}{\Gamma(\alpha)}\right)^{n}\left(\prod_{i = 1}^{n} x_{i}\right)^{\alpha-1} \mathrm{exp}\{-\beta\sum_{i = 1}^{n} x_i\} \end{align}

Thus the log-likelihood would be the following:

\begin{align} \ell(\beta | x_i) &= \ln\left((L(\beta | X_i)\right)) \\ &= n\alpha\ln(\beta) - n\ln(\Gamma(\alpha)) + (na-n)\ln(x_i) - \beta \sum_{i = 1}^{n} x_i \\ \end{align}

I understand that the information is found by taking the 2nd derivative of any of the likelihood functions where $I_o(\beta) = -\frac{\mathrm{d}^{2}{\ell}}{\mathrm{d}{\beta}^{2}} $

The derivatives calculated were as follows: \begin{align} &= n\alpha\ln(\beta) - n\ln(\Gamma(\alpha)) + (na-n)\ln(x_i) - \beta \sum_{i = 1}^{n} x_i \\ &= \frac{n\alpha}{\beta} - \beta \\ &= - \frac{n \alpha}{\beta^2} - 1 \end{align}

Thus the information would be

\begin{align} I_0(\beta) = \frac{n \alpha}{\beta^2} + 1 \end{align}

I am unsure what to do once I get to the expectation.

\begin{align} \mathbb{E}\{I_0(\beta)\} &= \mathbb{E}\left(\frac{n \alpha}{\beta^2} + 1\right) \end{align}

Would I have made a mistake within the derivation process? Any insight would be very much appreciated.

$\endgroup$

1 Answer 1

1
$\begingroup$

You almost got it right ! You just made a tiny mistake when computing the derivative of the log-likelihood, you should have had : $$\frac{\partial\ell}{\partial \beta} = \frac{n\alpha}{\beta} -\color{red}{\sum_{i=1}^n x_i} $$

From which it follows that $$\frac{\partial^2\ell}{\partial \beta^2} = -\frac{n\alpha}{\beta^2} $$

Next, to compute the Fisher information, all you have to do is to take the expectation of $-\frac{\partial^2\ell}{\partial \beta^2} $. However, that expectation is with respect to the distribution of $x_i$, and there are no $x_i$'s in the expression of $\frac{\partial^2\ell}{\partial \beta^2}$, it is a constant ! Its expectation is thus equal to itself : $$ \mathbb E\left[-\frac{\partial^2\ell}{\partial \beta^2}\right] = \mathbb E\left[\frac{n\alpha}{\beta^2}\right] = \frac{n\alpha}{\beta^2}. $$

$\endgroup$
1
  • $\begingroup$ Thank you! I appreciate it. $\endgroup$ Commented Apr 10, 2023 at 3:55

Your Answer

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

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