Correlation Does Not Imply Causation
相关性通常是对称的，因果性通常是不对称的（单向箭头），相关性不一定说明了因果性，但因果性一般都会在统计层面导致相关性。这篇论文中提到了 “The Three Layer Causal Hierarchy”的概念。
| level | typical activity | example | |
|---|---|---|---|
| 关联（Association）$P(y | x)$ | seeing | How would seeing X change my belief in Y ? |
| 干预（Intervention）$P(y | do(x),z)$ | doing | what if I do X? |
| 反事实推断（Counterfactuals）$P(y_x | x^1, y^1)$ | imagining | What if I had acted differently? Was it X that caused Y? |
