Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

L4-10 · Causal machine learning, transport, and fairness

Define a causal learning target and explain the additional assumptions needed for transport and fairness claims.

Level 4 icon Level 4 · Undergraduate
Objective. Define a causal learning target and explain the additional assumptions needed for transport and fairness claims.
Before this lesson. L4-04, L4-06, L4-08, L4-09, L3-09
Hands-on lab. Confounding and intervention

Learn

Causal machine learning can estimate outcome regressions, treatment probabilities, or treatment effect variation. Flexible prediction methods do not create identification. Start with a defined population, intervention, outcome, and time horizon. Then state exchangeability, consistency, positivity, and any other assumptions required for the chosen target.

Transport asks whether results from one population or environment apply to another. If only the distribution of measured effect modifiers changes, valid stratum effects can sometimes be reweighted to the target distribution. This requires stable relevant mechanisms and enough source data in target strata. Unmeasured mechanism changes can defeat this strategy.

Fairness is partly a choice of social objective. Equal predictive error rates and counterfactual fairness are different criteria and can conflict. A causal fairness claim needs a justified graph and normative decisions about acceptable paths. Removing a protected attribute from inputs does not remove its information from correlated variables.

Practice

Solution

The weighted effect is 0.25(2)+0.75(-1)=-0.25. This is a target-population causal average under the assumptions. It does not establish fairness, which needs a stated criterion and further justification.

Check your understanding

What must come before fitting a causal effect learner?

Show the answer

C. A defined causal target and justified identification assumptions

A learner estimates quantities within a causal analysis. The intervention target and identification assumptions determine whether its estimates answer the intended causal question.

Sources

This lesson draws on Hernan & Robins, 2020Schölkopf et al., 2021Peters et al., 2015.

Next: L5-01 · States and the Born rule

References
  1. Hernan, M., & Robins, J. (2020). Causal Inference: What If. https://miguelhernan.org/whatifbook
  2. Schölkopf, B., Locatello, F., Bauer, S., Ke, N. R., Kalchbrenner, N., Goyal, A., & Bengio, Y. (2021). Towards Causal Representation Learning. https://arxiv.org/abs/2102.11107
  3. Peters, J., Bühlmann, P., & Meinshausen, N. (2015). Causal inference using invariant prediction: identification and confidence intervals. https://arxiv.org/abs/1501.01332
  4. Bareinboim, E. (2025). Causal Artificial Intelligence: A Roadmap for Building Causally Intelligent Systems. https://causalai-book.net/