8
$\begingroup$

Pearl et al. "Causal Inference in Statistics: A Primer" (2016) p. 69 Theorem 3.4.1 provides the Front-Door Adjustment formula: $$ P(y | \text{do}(X = x)) = \sum_z P(z | x) \sum_{x'} P(y|x', z)P(x'). $$ Does it mean $$ P(y | \text{do}(X = x)) = \sum_z \left[ P(z | x) \sum_{x'} \left\{ P(y|x', z)P(x') \right\} \right] $$ or $$ P(y | \text{do}(X = x)) = \left[ \sum_z P(z | x) \right]\cdot \left[ \sum_{x'} \left\{ P(y|x', z)P(x') \right\} \right]? $$

$\endgroup$
1
  • $\begingroup$ The last expression is not allowed, because $P(y|x',z)$ has $z$ in it, but that term has fallen out of scope of the $z$ summation. $\endgroup$ Commented Apr 2, 2020 at 14:40

1 Answer 1

6
$\begingroup$

It means $$ P(y|\text{do}(X=x)) = \sum_z \left[ P(z|x) \sum_{x'} \left\{ P(y|x',z) P(x') \right\} \right]. $$ This can be seen from equation (3.15) on p. 68 which is $$ P(y|\text{do}(X=x)) = \sum_z \sum_{x'} \left\{ P(y|x',z) P(x') P(z|x) \right\} $$ and can be rearranged to be $$ P(y|\text{do}(X=x)) = \sum_z \sum_{x'} \left\{ P(z|x) P(y|x',z) P(x') \right\}, $$ the latter ordering of conditional probabilities corresponding to the ordering in the equation at the top of this answer.

$\endgroup$
1
  • 1
    $\begingroup$ Right. $x'$ is just a dummy variable, and does not show up in the $P(z|x),$ so that that term can move in or out of the second sum with impunity. $\endgroup$ Commented Apr 2, 2020 at 14:31

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.