0
$\begingroup$

How can I write the unnormalized posterior

$ f(p_1, p_2 | Y) = (z_1-1)*log(p_1) + (n_1-z_1-1)*log(1-p_1) + (z_2-1)*log(p_2) + (n_2-z_2-1)*log(1-p_2) $

in terms of the log odds-ratio $\alpha$ and the log odds-product $\eta$?

$\alpha = log \left(\frac{p_2/(1-p_2)}{p_1/(1-p_1}\right) $ and $\eta = log \left(\frac{p_2}{1-p_2} \times \frac{p_1}{1-p_1}\right)$?

where z_1, n_1, z_2, and n_2 are constants.

$\endgroup$
2
$\begingroup$

Note that we have $$ \alpha = \log p_2 - \log(1-p_2) - \log p_1 + \log(1-p_1),$$ $$ \eta=\log p_2 -\log (1-p_2)+\log p_1 -\log(1-p_1).$$ Then $$\eta-\alpha = 2(\log p_1 - \log (1-p_1),$$ $$\eta +\alpha=2(\log p_2 -\log(1-p_2).$$

EDIT: previously I wrote that your unnormalized posterior couldn't be written in terms of $\alpha$ and $\eta$. However I just noticed that $$\frac{\eta - \alpha}{2}=\log \frac{p_1}{1-p_1}$$ so that $$p_1 = \frac{e^{(\eta-\alpha)/2}}{e^{(\eta-\alpha)/2}+1},$$ and similarly $$\frac{\eta + \alpha}{2}=\log \frac{p_2}{1-p_2}$$ so that $$p_2 = \frac{e^{(\eta+\alpha)/2}}{e^{(\eta+\alpha)/2}+1}.$$ Now since your unnormalized posterior is a function of $p_1$ and $p_2$, and of the constants $z_1,n_1,z_2,n_2$, you can using the above expressions write the posterior in terms of $\alpha$ and $\eta$.

$\endgroup$
  • $\begingroup$ Note to the OP: please reread this, after the EDIT, since I figured out you can rewrite as you wanted in terms of $\alpha,\eta$. $\endgroup$ – coffeemath Dec 9 '12 at 2:55

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.