Book Summary: The book of why, Dana Mackenzie and Judea Pearl

Key take aways

Without a casual model, data science is purely an exercise in data reduction. Statistics can only prove correlation but not causation. The human brain is a natural causal machine and is thus still relevant. Humans should go about identifying causality within our environment and delegate the ongoing monitoring to machines.

Components for consideration

Confounders

B and C are positively correlated but there are no direct causal links between the both of them. Without knowledge that A exists, B and C can be considered to be confounders.

  • A therefore B
  • A therefore C
  • C and B both occur together
  • B does not cause C
  • C does not cause B

Colliders

In the example A and C are considered colliders because both of them will result in B

  • A therefore B
  • C therefore B

Figuring out the coefficients

  • A therefore B
  • B therefore C
  • D therefore C

Front door techniques

Figuring out the cause model via path of A -> B -> C. This approach is only possible when you are able measure A, B and C

Backdoor techniques

Figuring out C by isolating B and D. This approach is used when not the entire causal chain is measurable.

Leave a Reply