2
$\begingroup$

I missed the day of class where we went over likelihood functions, and I had a quick question. If $X_1,...X_n$ are i.i.d. $ {\Gamma}(\alpha,\beta)$ r.v.s, I'm trying to find the likelihood function for $\alpha$ and $\beta$. From what I can tell, the likelihood function is defined as $f(x_1;\alpha,\beta)*...*f(x_1;\alpha,\beta)$. If i'm not mistaking, $f(x_1;\alpha,\beta)$ should be the same as the pdf for the gamma distribution (although not a pdf, on account of x being a fixed value here?), so would the likelihood function in this case be ${(\text{pdf-of-}\Gamma)}^n$? Further, I'm a bit confused on what the support for this function should be; in this case x is fixed, am I correct in supposing that the support is then $\alpha>0, \beta>0$? I hope this makes sense - I get the feeling I might be very confused about the whole thing. Thank you for any help!

$\endgroup$
3
$\begingroup$

If $X$ follows a gamma distribution with shape $\alpha$ and scale $\beta$, then its probability density is

$$p_{\alpha, \beta}(x) = \frac{ x^{\alpha-1} e^{-x/\beta}}{\Gamma(\alpha) \beta^\alpha }$$

Sometimes this is re-parameterized with $\beta^{\star} = 1/\beta$, in which case you will need to change this accordingly.

The likelihood function is just the density viewed as a function of the parameters. So, the log-likelihood function for an IID sample $X_1, ..., X_n$ from this distribution with realized values $x_1, ..., x_n$ is

$$ L(\alpha, \beta) = \sum_{i=1}^{n} \log \big( p_{\alpha, \beta}(x_i) \big) = (\alpha-1) \sum_{i=1}^n \log(x_i) - \frac{1}{\beta} \sum_{i=1}^{n}x_i - n\alpha \log(\beta) - n\log( \Gamma(\alpha) )$$

which can be maximized jointly as a function of $\alpha, \beta$ to get the MLE.

$\endgroup$
2
  • $\begingroup$ How do we differentiate gamma$\alpha$ to get the value of $\alpha$ ? $\endgroup$ – idpd15 Sep 3 '15 at 12:27
  • $\begingroup$ @idpd15 The derivative of the logarithm of the Gamma function is the Digamma function, which can be approximated for certain input values using the Euler-Mascheroni constant. Though, I wasn't able, though, to solve the system of two equations formed by the derivatives of logL relatively to $\alpha$ and $\beta$. $\endgroup$ – Hakim Dec 26 '16 at 16:59

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.