Causal Inference¶

Infer the effect of changing X on Y from data by fixing a causal estimand and defending an identification design or assumption that separates that effect from noncausal association, then quantify its uncertainty and scope.

Version: v5 · 2026-09-19 · History
Domain-specific #: 140
Domain group: Formal Sciences
Origin domain: Experimental Design & Statistics
Subdomain: Causal Inference → Experimental Design & Statistics

Core Idea¶

Causal inference is the disciplined practice of learning the effect of changing a treatment, exposure, action, or variable on an outcome from necessarily incomplete data. It begins by defining a causal estimand: a contrast between outcomes under alternative treatment or intervention states for a named population. Because no unit or system is normally observed in every alternative state, the observed association does not by itself identify that contrast. The analyst must therefore state and defend an identification design or assumption—random assignment, conditional exchangeability, a valid instrument, continuity at a cutoff, parallel trends, or another explicit warrant—before an estimator can translate the observed data into a causal claim.

The abstraction is a specialized form of Statistical Inference: it still moves from finite, noisy observations to an uncertain conclusion, but fixes the target to a causal effect and adds an identification burden that ordinary estimation or prediction does not carry. It presupposes Counterfactuals because the target effect is defined by outcomes under alternative states, only some of which are observed. It does not receive a separate direct Causality edge because Counterfactuals already strictly presupposes Causality in the live topology.

How would you explain it like I'm…

Did It Really Cause That?

To know if a medicine really helps you, you'd want to see what happens if you take it and what happens if you don't — but you can only do one. Causal inference is the careful way of figuring out what something really does anyway. You need a good reason to trust your comparison, like flipping a coin to decide who gets the medicine.

Finding Real Causes

Causal inference is about finding out what really happens when you change something, like giving a medicine. The hard part is that you can never see the same person both with and without the medicine at the same time. Just noticing that two things go together isn't enough, because something else might explain it. Scientists use tricks like flipping a coin to decide who gets the medicine, or careful comparisons, to argue that the difference they see is caused by the change. Sometimes the honest answer is that we can't tell for sure.

Identifying Causal Effects

Causal inference is the practice of learning the effect of changing something — a treatment, action, policy, or variable — on an outcome, using data that are necessarily incomplete. It starts by defining exactly what effect you want: a comparison of outcomes under at least two possible treatments, for a specific population and time frame. Since you never see every possible outcome for the same person or system, an observed association only counts as a causal effect if you have an explicit argument, called identification, linking the data to the missing comparison. Randomized experiments create that argument by design. In observational data you must defend it — for instance, by adjusting for all important confounders, using an instrumental variable, comparing just above and below a cutoff, or assuming parallel trends. Only after identification do you choose an estimator, measure uncertainty, and run checks. A careful analysis may conclude the effect is only partly known or can't be identified at all.

Causal inference is the disciplined practice of learning the effect of changing a treatment, exposure, action, policy, or variable on an outcome from necessarily incomplete data. It begins with a causal estimand — a contrast between outcomes under at least two intervention states for a named population and horizon — and confronts the fact that the data never reveal every counterfactual outcome for the same unit. An association becomes causal only via an identification argument linking the observable contrast to the missing one. Randomization manufactures that warrant; observational strategies defend it: conditional exchangeability licenses confounder adjustment, a valid instrument isolates exogenous treatment variation, continuity supports regression-discontinuity comparisons at a cutoff, and parallel trends supports difference-in-differences. Each makes different claims about which noncausal paths are blocked and which effect (e.g., local or average) is recovered. Estimation follows identification, quantifying uncertainty and facing diagnostics, falsification tests, and sensitivity analysis; a responsible analysis may conclude that the effect is only bounded, locally identified, assumption-sensitive, or unidentified.

Scope of Application¶

The same methodological substrate appears across economics, epidemiology, medicine, political science, education, psychology, and digital experimentation. The particular treatments, outcomes, and identifying institutions change, while the sequence—estimand, identification warrant, estimator, diagnostics, uncertainty, and scope—remains intact.

Clarity¶

The decisive question is not whether a model contains causal language, but what makes its observed contrast equal the named causal estimand. A credible causal inference makes that bridge inspectable. If the bridge is only correlation, predictive accuracy, temporal order, or an unsupported story, the result is not causal inference.

Examples¶

A randomized trial identifies an average treatment effect because assignment makes the treatment groups exchangeable in expectation.
An observational study adjusts for a defended sufficient set of pre-treatment covariates under selection on observables.
A regression discontinuity design compares units near a threshold under a continuity assumption.
An instrumental-variable analysis uses relevant, exogenous variation subject to an exclusion restriction and scopes the result to the margin the instrument moves.

Relationships to Other Abstractions¶

Current abstraction Causal Inference Domain-specific

Parents (2) — outgoing typed relations

Causal Inference is a kind of Statistical Inference Prime

Causal Inference is Statistical Inference specialized to intervention or counterfactual effects whose recovery requires an explicit identification warrant.
Causal Inference presupposes Counterfactuals Prime

Causal Inference presupposes outcomes under alternative treatment states because their contrast defines the effect it seeks to identify.

Children (4) — abstractions linked here by typed relations

Instrumental variable Domain-specific is a kind of Causal Inference

Instrumental Variables are Causal Inference specialized to identification by relevant, exogenous variation reaching the outcome only through treatment.
Natural Experiment Domain-specific is a kind of Causal Inference

A Natural Experiment is Causal Inference specialized to found as-if-random assignment, a substantively defended identifying process, and a local effect.
Internal validity Domain-specific presupposes Causal Inference

Internal Validity presupposes a Causal Inference whose in-setting warrant it audits against confounding, selection, history, attrition, and other threats.
Selection on Observables Domain-specific presupposes Causal Inference

Selection on Observables presupposes the Causal Inference activity whose observational treatment-effect conclusion depends on conditional exchangeability.

Hierarchy paths (5) — routes to 5 parentless roots

Causal Inference → is a kind of Statistical Inference → is a kind of Inductive Reasoning

Show alternative paths (4)

Neighborhood in Abstraction Space¶

Causal Inference sits in a sparse region of the domain-specific corpus (70^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Causal Inference & Regression Modeling (15 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-25

Not to Be Confused With¶

Statistical Inference: broader; its target may be descriptive, predictive, or associational.
Causality: the productive asymmetric relation in the world, not the practice used to learn about it from data.
Counterfactuals: the alternative-state structure defining the target, not the complete identification-and-estimation enterprise.
Intervention: one way to define or manufacture a causal contrast; observational causal inference need not perform a literal intervention.
Identifiability: a property of whether a target is uniquely recoverable under a model and evidence; causal inference can also report bounds, sensitivity, or non-identification.

References¶

Hernán, M. A., & Robins, J. M. (2024). Causal Inference: What If. Chapman & Hall/CRC.
Imbens, G. W., & Rubin, D. B. (2015). Causal Inference for Statistics, Social, and Biomedical Sciences. Cambridge University Press.
Pearl, J. (2009). Causality: Models, Reasoning, and Inference (2^nd ed.). Cambridge University Press.

Notes¶

(New domain genus; queued for Claude style harmonization, FACT-anchor treatment, and independent citation verification.)