DAGs and Assumptions in Causal Inference

🔗 DAGs and Assumptions in Causal Inference

Directed Acyclic Graphs (DAGs) provide a powerful way to encode assumptions about how variables in your system are related.

They help us: - Identify confounders and mediators - Test identification strategies - Derive adjustment sets

---

🎯 Define a Simple DAG

Let’s say we’re estimating the effect of Treatment on Outcome, with a confounder X.

dag <- dagitty("dag {
  X -> Treatment
  X -> Outcome
  Treatment -> Outcome
}")

plot(dag)

---

📚 Adjustment Sets

Which variables do we need to control for to identify the causal effect of Treatment on Outcome?

adjustmentSets(dag, exposure = "Treatment", outcome = "Outcome")

## { X }

This shows that adjusting for X is sufficient.

---

🧠 More Complex DAG

complex_dag <- dagitty("dag {
  Age -> Treatment
  Income -> Treatment
  Age -> Outcome
  Income -> Outcome
  Treatment -> Behavior -> Outcome
  unobserved1 [unobserved]
  unobserved1 -> Income
  unobserved1 -> Outcome
}")

ggdag::ggdag(complex_dag, layout = "circle")

  ggtitle("DAG with Confounding and Mediation")

## $title
## [1] "DAG with Confounding and Mediation"
## 
## attr(,"class")
## [1] "labels"

---

✅ Summary

DAGs help us: - Encode our causal assumptions - Identify valid adjustment strategies - Prevent common pitfalls like conditioning on colliders

Learning to draw and analyze DAGs is a core skill for any causal analyst.

---

📖 References

• Textor et al. (2016). dagitty: Graphical Analysis of Structural Causal Models
• Pearl (2009). Causality: Models, Reasoning, and Inference