-1
$\begingroup$

Suppose the joint posterior density for parameters $(\theta_1, \theta_2)$ can be expressed as \begin{align} \Pr(\theta_1, \theta_2 \mid y) &\propto Pr(y \mid \theta_1, \theta_2) *Pr(\theta_1, \theta_2)\\ &=Pr(y \mid \theta_1, \theta_2) *Pr(\theta_1 \mid \theta_2)*Pr(\theta_2) \end{align} These 2 forms are equivalent, right? Then, can we factor and express the joint posterior as: \begin{align} \Pr(\theta_1, \theta_2 \mid y) &= (\Pr(\theta_1 \mid \theta_2, y)*Pr(\theta_2 \mid y)) \\ &\propto Pr(y \mid \theta_1, \theta_2) *Pr(\theta_1 \mid \theta_2) * Pr(\theta_2 \mid y) \\ &\propto Pr(y \mid \theta_1, \theta_2) *Pr(\theta_1 \mid \theta_2)*Pr(y \mid \theta_2 )* Pr(\theta_2) \end{align}

I think I followed the rules of factorization but because of $Pr(y \mid \theta_2)$ I can't equate the first derivation with the second up to a proportionality constant. $$ Pr(y \mid \theta_1, \theta_2) *Pr(\theta_1 \mid \theta_2)*Pr(\theta_2) \propto Pr(y \mid \theta_1, \theta_2) *Pr(\theta_1 \mid \theta_2)*Pr(y \mid \theta_2 )* Pr(\theta_2) $$

Can someone point me where I made a mistake?

Improve this question
$\endgroup$
2
  • 1
    $\begingroup$ In your second equation, I don't see how you can write $Pr(\theta_1|\theta_2,y)$ as being proportional to $Pr(y | \theta_1, \theta_2) \; * \; Pr(\theta_1 | \theta_2)$. $\endgroup$
    – Vimal
    Commented Sep 14, 2015 at 6:13
  • $\begingroup$ Yes I see, I've fixed it in my answer. $\endgroup$
    – ilyas patanam
    Commented Sep 14, 2015 at 6:34

1 Answer 1

Reset to default
0
$\begingroup$

As Vimal points out my mistake was writing, $$Pr(\theta_1 \mid \theta_2, y)*Pr(\theta_2 \mid y) \propto Pr(y \mid \theta_1, \theta_2) *Pr(\theta_1 \mid \theta_2) * Pr(\theta_2 \mid y)$$ When factoring, \begin{align} (\Pr(\theta_1 \mid \theta_2, y)*Pr(\theta_2 \mid y)) &= \Pr(\theta_1 \mid \theta_2, y)*Pr(\theta_2 \mid y) \\ &= \dfrac{Pr(y \mid \theta_1, \theta_2) *Pr(\theta_1 \mid \theta_2)}{Pr(y \mid \theta_2)} * Pr(\theta_2 \mid y) \\ &= \dfrac{Pr(y \mid \theta_1, \theta_2) *Pr(\theta_1 \mid \theta_2)}{Pr(y \mid \theta_2)} * \dfrac{Pr(y \mid \theta_2) Pr(\theta_2)}{Pr(y)} \\ &\propto Pr(y \mid \theta_1, \theta_2) *Pr(\theta_1 \mid \theta_2) * Pr(\theta_2) \end{align} Thus, both approaches factoring the prior and factoring the posterior are equivalent.

Improve this answer
$\endgroup$

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.