I have this problem that I stumbled upon. Suppose the random variable $X$ follows a Gamma distribution with parameters $\alpha$ and $\beta$ with the probability density function for $x>0$ as

$$f(x)= \frac{\beta^\alpha}{\Gamma(\alpha)} x^{\alpha-1} \exp(-\beta x)$$

where $\Gamma(\alpha)$ represents the Gamma function with $\Gamma(\alpha)=(\alpha-1)!$ when $\alpha$ is a natural number.

Further suppose we know that for the random variable $X$, the parameter $\alpha=4$. We record the independent observations $X_1,X_2,\ldots,X_n$ as a random sample from the distribution.

And I must find the likelihood function for $\beta$, $L(\beta)$, given $\alpha=4$, the maximum likelihood estimator $β$ and show that this indeed is a maximum.

I found that the Maximum Likelihood is: $\beta= 4n/\sum x_i$ but i am not sure if my way of thinking is correct. Any help will be much appreciated

  • $\begingroup$ What is your argument? $\endgroup$ – Hans Engler Dec 12 '14 at 0:39
  • $\begingroup$ I found that likelihood function is: L(β)= Π (β^4 * xi^3 * exp(-βxi)/(3!), then worked out the log likelihood, differentiated it and equaled it to zero and found the Maximum Likelihood β as showed above. Hope this helps $\endgroup$ – rogerdom Dec 12 '14 at 0:43
  • 1
    $\begingroup$ That's the right approach, and the answer is correct. $\endgroup$ – Hans Engler Dec 12 '14 at 0:44
  • $\begingroup$ The fact that the derivative is zero at a certain point is not enough to prove that there is a maximum there. But see my answer below. $\endgroup$ – Michael Hardy Dec 12 '14 at 1:01
  • $\begingroup$ In order to show that there is a maximum i found the second derivative which is -4n/β^2 which is less than 0 thus is a maximum. Is that the full log-likelihood mentioned in your comment? I found the same but + 3Σ log xi - nlog3!. $\endgroup$ – rogerdom Dec 12 '14 at 1:08

\begin{align} L(\beta) & = \prod_{i=1}^n \frac{\beta^4}{\Gamma(4)} x_i^{4-1} \exp(-\beta x_i) \\[8pt] & \propto \beta^{4n} \exp\left(-\beta\sum_{i=1}^n x_i\right) \end{align} (The factor $\prod_{i=1}^n x_i$ does not depend on $\beta$ and so is a part of the constant of proportionality, as is $(\Gamma(4))^n$.) $$ \ell(\beta) = \log L(\beta) = C + 4n\log\beta -\beta\sum_{i=1}^n x_i $$ $$ \ell'(\beta) = \frac{4n} \beta -\sum_{i=1}^n x_i \quad \begin{cases} >0 & \text{if } 0<\beta<\dfrac{4n}{\sum_{i=1}^n x_i}, \\[6pt] = 0 & \text{if } \beta=\dfrac{4n}{\sum_{i=1}^n x_i}, \\[6pt] <0 & \text{if } \beta>\dfrac{4n}{\sum_{i=1}^n x_i}. \end{cases} $$

| cite | improve this answer | |

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.