It earns a place when
- Use causal methods when the decision is about intervention, policy or counterfactual impact, and make assumptions explicit.
- You can evaluate it against a credible baseline.
- Its output fits the decision and data constraints.
Machine learning models · Phase 6 · Lesson 51
Causal inference and structural causal models
Distinguish prediction from intervention and use causal graphs to reason about confounding.
Lesson goal
Distinguish prediction from intervention and use causal graphs to reason about confounding.
The core idea
Prediction asks who tends to have an outcome. Causation asks what would change if we intervened, such as giving a treatment to a person.
Mental model
Picture it this way. Prediction asks who tends to have an outcome. Causation asks what would change if we intervened, such as giving a treatment to a person. The important question is what assumption this picture makes, and whether that assumption fits the data.
Mathematical core
A structural causal model assigns equations to variables and a graph encodes assumptions. The backdoor idea asks whether conditioning on a sufficient set blocks non-causal paths.
Worked example
A customer who receives a discount may be more likely to churn because high-risk customers were targeted. Comparing treated and untreated groups without adjustment confuses targeting with treatment effect.
When to use it
Do not make it the default when
- Do not call feature importance causal, and do not believe a causal estimate without checking identification, measurement and treatment versions.
- A simpler model has not been tested.
- The data or target definition is still unclear.
Failure modes
Watch for this. Unmeasured confounding, selection bias, post-treatment adjustment and positivity violations can invalidate an estimate.
When a result looks surprisingly good, inspect the split, target timing, error slices and data-generating process before celebrating.
Practice
Use this as a small experiment rather than a recipe to copy blindly. Change one thing, record the result and explain the change.
Do this. Draw a causal graph for an intervention you care about. Mark treatment, outcome, confounder, mediator and collider, then state which variables you would adjust for and why.
Retrieval check
Answer from memory first. The buttons reveal feedback, but the durable step is explaining why.
1. What does an intervention question ask?
2. What is confounding?
3. Why can adjusting for a collider hurt?
Transfer prompt. Describe one real problem where this model or idea would be a sensible candidate. Name the target, the main risk and the metric you would inspect.
Primary source
Causal Inference: What If. Use the source for the deeper treatment after you can explain the lesson's core idea without looking.