3
$\begingroup$

I'm reading "The Book of Why" by Judea Pearl and I'm getting an answer for a problem that doesn't match what the book says. This is my first foray into Structural Causal Models and their use, so I want to be absolutely certain that there isn't a gap in my understanding when I try to solve the problem; hence this question.

Here's the problem that I'm stuck on:

The Data

Bert: $EX = 9$, $ED = 1$, $S_1 = 92500$.

Caroline: $EX= 9$, $ED = 2$, $S_2 = 97000$

The equations

According to more data in the table, the following equations hold:

$$S_{ED} = 65000 + 2500 EX + 5000 ED + U_S\\ EX = 10 - 4 ED + U_{EX}$$

where $U_S$ and $U_{EX}$ are background variables that are different for Bert and Caroline.

The Question

The question requires that we calculate the value of $S_1(Bert) - S_1(Caroline)$ (basically subtract from 92500, the value of $S_{ED}$ for Caroline if her $ED = 1$). The problem is posed between pages 273 and 282.

My confusion

The book says that the final value should be 5000. My numbers are either 500, if I assume that the value of EX remains 9 like before (which would be wrong because the value of EX should be updated according to the second equation?), or -9500, if I simply plug in the new value of ED and solve the equations.

Am I missing something here? It's very possible as this whole exercise is couched in the realm of operating with causal model diagrams, and I might have missed an update step to the equations...

$\endgroup$
8
$\begingroup$

You are not missing anything, the correct answer is $-9500$!

First find the idiosyncratic characteristics of Caroline, by updating her background variables $U$ given the evidence $EX = 9$, $ED = 2$, $S = 97000$. This is the Abduction step,

$$ 97000 = 65000 + 2500(9) + 5000(2) +U_{s} \implies U_s = -500\\ 9 = 10-4(2) + U_{EX} \implies U_{EX} = 7 $$

Now we perform the action, which is setting Caroline's new education to 1, that is, $do(ED = 1)$, and properly mutilating the graph and structural equations. Finally we perform prediction under the new system (using the new updated information about the exogenous variable and the interventional graph),

$$ EX_1(Caroline) = 10-4*(1) + 7 = 13 \\ $$ And, $$ \begin{align} S_{1}(Caroline) &= 65000 + 2500\times EX_{1}(Caroline) + 5000(1) -500 \\ &=65000 + 2500(13) + 5000 - 500\\ &= 102000 \end{align} $$

Thus, $S_1(Bert) - S_1(Caroline) = 92500 - 102000 = -9500$.

PS: Pearl is aware of the typo and this will probably be in the errata and corrected soon.

$\endgroup$
  • $\begingroup$ This kind of statements should be backed up with some kind of source for the claim. $\endgroup$ – Tim Aug 13 '18 at 9:01
  • 1
    $\begingroup$ @Tim Personal communication and now twitter.com/yudapearl/status/1028828169028005888 $\endgroup$ – Carlos Cinelli Aug 13 '18 at 15:03
  • $\begingroup$ Thanks, maybe you could edit and put the link into the body of the answer? $\endgroup$ – Tim Aug 13 '18 at 15:07
  • 2
    $\begingroup$ @Tim this is just algebra, anyone can verify the correct answer by themselves. $\endgroup$ – Carlos Cinelli Aug 13 '18 at 15:08
  • $\begingroup$ I was referring to PS that was preciously the only part of the answer. $\endgroup$ – Tim Aug 13 '18 at 15:09

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.