I have been trying to understand predictive posterior distribution.The following expression is given here

$p(x^*|x)=\int_\Theta c\times p(x^*,\theta|x)d\theta=\int_\Theta c\times p(x^*|\theta)p(\theta|x)d\theta$

I understand that the conditional probability is defined as

$f(A|B)=\frac{f(A,B)}{f(B)} = \frac{f(B|A)*f(A)}{f(B)}$

Can someone explain how the predictive posterior distribution is written using conditional probability?


The first equality is the law of total probability. It is a special case of $f_X(x) = \int f_{X,Y}(x,y) \, dy$.

The second equality comes from computations with conditional densities similar to what you have written: $p(x^*, \theta \mid x) = p(x^* \mid \theta, x) p(\theta \mid x) = p(x^* \mid \theta) p(\theta \mid x)$, where the last step is due to $p(x^* \mid \theta, x) = p(x^* \mid \theta)$ by conditional independence of $x^*$ and $x$ given $\theta$.

  • $\begingroup$ $p(x_k, x_{k-1}\mid y_{1:k-1}) = p(x_k \mid x_{k-1},y_{1:k-1}) p(x_{k-1} \mid y_{1:k-1})$ Could you please explain how this is written? $\endgroup$ – Natasha Apr 20 '18 at 17:34
  • $\begingroup$ @Natasha This is the same as $p(u,v) = p(u \mid v) p(v)$, except with an extra conditioning on another variable $z$: $p(u, v \mid z) = p(u \mid v,z) p(v \mid z)$. $\endgroup$ – angryavian Apr 20 '18 at 17:36
  • $\begingroup$ Thank you!That clarifies the second part.I am still not able to relate,"It is a special case of $f_X(x) = \int f_{X,Y}(x,y) \, dy$" to the first equality.$f_X(x) $" is the marginal density.Does the term on the left refer to conditioning on the marginal density?Could you please elaborate? $\endgroup$ – Natasha Apr 20 '18 at 17:48
  • $\begingroup$ @Natasha Again, it is an extra conditioning, e.g. $f(x \mid z) = \int f(x,y \mid z) \, dy$. $\endgroup$ – angryavian Apr 20 '18 at 19:29

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.