40
$\begingroup$

I'm having trouble verifying why the following is correct.

$$p(x, y \mid z)= p(x \mid y, z) p(y \mid z)$$

I tried grouping the $(x, y)$ together and split by the conditional, which gives me

$$p(x, y \mid z) = p(z\mid x, y) p(x, y)/p(z)$$

However, this did not bring me any closer. I'm uncertain about what kind of manipulations are allowed given more than 2 variables.

Say an expression like:

$$p(a, b, c)$$

Then I know from the chain rule that I can break it down to:

$$p(a, b, c)=p(a \mid b, c) p(b, c) = p(a \mid b, c) p(b \mid c) p(c)$$

Is it allowed to split by the second comma:

$$p(a, b, c) = p(a, b \mid c) p(c) ?$$

And even more complicated and expression like: $$p(a|b,c)$$

Am I allowed to rewrite this expression by grouping (a|b) together to give me something like $$p(a|b,c)=p((a|b)|c)p(c)$$ And does this expression even make sense?

$\endgroup$
  • 2
    $\begingroup$ Some people object to $p((a|b)|c)$. See this for example $\endgroup$ – Henry Jul 28 '12 at 23:50
  • $\begingroup$ Actually $a$, $b$ and $c$ should refer to events rather than variables. $p(x,y|z)$ should be written as $P(x\bigcap y| z)$, etc. $\endgroup$ – A.G. Oct 15 '16 at 3:11
36
$\begingroup$

$\Pr(a,b,c)=\Pr(a,b\mid c)\Pr(c)$ is allowed.

You are simply saying $\Pr(d,c)=\Pr(d\mid c)\Pr(c)$ where $d = a \cap b$.

Combine this with $\Pr(a,b,c)=\Pr(a\mid b,c)\Pr(b,c)=\Pr(a\mid b,c)\Pr(b\mid c)\Pr(c)$ and divide through by nonzero $\Pr(c)$ to get $\Pr(a,b\mid c)=\Pr(a\mid b,c)\Pr(b\mid c)$.

$\endgroup$
  • 3
    $\begingroup$ Thanks, I had failed to see the obvious $p(a,b,c)/p(c)=p(a,b|c)$ $\endgroup$ – Jim Jul 29 '12 at 0:20

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.