Statistics for ML #99 — Causal Inference: DAGs, Do-Calculus
Published:
Causal Inference: DAGs, Do-Calculus
| Post #99/100 in the Statistics for ML series — Md Salek Miah | Statistician & ML Researcher | SUST, Bangladesh. |
Causal Inference moves beyond correlation to answer: “Does X cause Y?” This is the central question in public health policy.
DAGs (Directed Acyclic Graphs)
A DAG encodes causal assumptions. Arrows represent direct causal effects. Used to identify confounders, mediators, and colliders.
Tools: Dagitty (web-based), dagitty R package — used in our DHS research for confounder selection.
Do-Calculus (Pearl’s Framework)
\(P(Y | do(X=x)) \neq P(Y | X=x)\)
The do-operator represents intervening to set X, vs observing X.
Key Concepts
| Concept | Description | Example |
|---|---|---|
| Confounder | Causes both X and Y | Wealth → ANC visits AND delivery location |
| Mediator | On causal path X→M→Y | ANC → knowledge → SBA |
| Collider | Caused by both X and Y | Conditioning on collider opens non-causal path |
library(dagitty)
library(ggdag)
# DAG for SBA research
dag <- dagitty('dag {
Wealth -> ANC_visits
Wealth -> SBA
ANC_visits -> SBA
Education -> Wealth
Education -> SBA
Rural -> ANC_visits
Rural -> SBA
}')
# Find minimal adjustment set for ANC -> SBA
adjustmentSets(dag, exposure="ANC_visits", outcome="SBA")
ggdag(dag, layout="circle") + theme_dag()
Series Index | Post #99/100 | Md Salek Miah | saleksta@gmail.com