- [[causal revolution]], [[causal inference]]
# Idea
[[Judea Pearl]] distinguished three levels of causal inference:
- **association**: $P(y|x)$
- seeing: what is?
- the probability of $Y = y$ given that we observe $X = x$
- **intervention**: $P(y|do(x))$
- manipulation/experimentation/doing: what if?
- the probability of $Y = y$ given that we intervene and set the value of $X$ to $x$
- **counterfactuals**: $P(y_x|x', y')$
- imagining/retrospecting: what if I had acted differently?
- the probability of $Y = y$ if $X$ had been $x$ given that we actually observed $x'$, $y'$
These three *levels* related to data science's three tasks: description, prediction, [[causal inference|causal inference]].
# References