Treatment Effects: ATE, CATE, and Heterogeneous Effects

College Depth 85 in the knowledge graph I know this Set as goal
Unlocks 2 downstream topics
causal-inference treatment-effects heterogeneous

Core Idea

The Average Treatment Effect (ATE) measures mean causal impact across the population; Conditional Average Treatment Effect (CATE) varies by subgroups; Local Average Treatment Effect (LATE) applies to compliers in instrumental variable settings. Identifying these requires assumptions about confounding and selection mechanisms.

How It's Best Learned

Contrast ATE, CATE, and LATE with examples; understand which assumptions identify each and when each is relevant.

Common Misconceptions

ATE and CATE are conceptually different; ATE averages over the population while CATE allows effects to differ by observable characteristics.

Explainer

From your potential outcomes prerequisite, you know the fundamental setup: every unit i has two potential outcomes — Y(1) if treated and Y(0) if not treated. The individual causal effect is τᵢ = Y(1)ᵢ − Y(0)ᵢ, but you observe only one potential outcome per person. Since individual effects are never identified, empirical work targets averages — and which average matters depends on the policy question being asked.

The Average Treatment Effect (ATE) is E[Y(1) − Y(0)] averaged over the full population, treated and untreated alike. This is the right quantity for a universal policy question: if we were to randomly select someone from the population and treat them, what would happen on average? The Average Treatment Effect on the Treated (ATT) narrows focus to those who actually received treatment: ATT = E[Y(1) − Y(0) | D = 1]. For a voluntary job training program, ATT tells you the benefit to participants — people who chose to enroll. ATE would additionally include the counterfactual effect on people who didn't enroll, which may be different if selection into treatment was based on expected benefit.

The Conditional Average Treatment Effect (CATE) takes heterogeneity seriously. Rather than a single number, CATE maps each covariate vector X to a local treatment effect: τ(x) = E[Y(1) − Y(0) | X = x]. This is the right framework when effects are expected to vary across subgroups — a drug may work better for older patients, a training program may benefit low-educated workers more. In practice, estimating CATE requires data-rich methods: causal forests and other machine learning approaches partition the covariate space to estimate local effects, but they need large samples to be precise. The fundamental tradeoff is between aggregation (ATE is precise but hides heterogeneity) and granularity (CATE reveals heterogeneity but demands more data and stronger assumptions).

The Local Average Treatment Effect (LATE) arises in instrumental variables settings. When you use an instrument Z to address selection into treatment, the IV estimator does not recover ATE or ATT — it recovers the effect only for compliers: people who take treatment when Z assigns them to treatment and decline when Z does not. Always-takers (who take treatment regardless) and never-takers (who never take it) contribute no identifying variation. LATE can differ substantially from ATE if compliers are unusual. In a military draft lottery study, LATE measures the earnings effect for men who served because they were drafted but would not have enlisted voluntarily — this may not generalize to those who chose to serve.

The practical lesson is to specify the target parameter before analyzing data. "The effect of X on Y" is incomplete. "The ATE if the policy is universally applied" differs from "the ATT — the effect on current participants" which differs from "which subgroups benefit most (CATE)" which differs from "what does this particular instrument identify (LATE)." Most empirical papers report ATT or LATE rather than ATE, because their identification strategy only credibly identifies effects for a specific subpopulation. Being explicit about which estimand you are targeting — and why it matches your policy question — is the hallmark of careful causal inference.

Practice Questions 5 questions

Prerequisite Chain

Counting to 10Counting to 20Understanding ZeroThe Number ZeroCounting to FiveOne-to-One CorrespondenceCombining Small Groups Within 5Addition Within 10Addition Within 20Two-Digit Addition Without RegroupingTwo-Digit Addition with RegroupingAddition Within 100Repeated Addition as MultiplicationMultiplication Facts Within 100Division as Equal SharingDivision as Grouping (Measurement Division)Division: Grouping (Repeated Subtraction) ModelDivision: Fair Sharing ModelDivision as Equal SharingDivision as GroupingBasic Division FactsDivision Facts Within 100Two-Digit by One-Digit DivisionDivision with RemaindersRemainders and Quotients in DivisionDivision Word ProblemsIntroduction to Long DivisionFactors and MultiplesPrime and Composite NumbersEquivalent FractionsRelating Fractions and DecimalsDecimal Place ValueReading and Writing DecimalsComparing and Ordering DecimalsAdding and Subtracting DecimalsMultiplying DecimalsDividing DecimalsDividing FractionsMixed Number ArithmeticOrder of OperationsInteger Order of OperationsVariable ExpressionsCombining Like TermsOne-Step EquationsTwo-Step EquationsSolving Multi-Step EquationsEquations with Variables on Both SidesAngle Pairs: Complementary, Supplementary, and VerticalParallel Lines and TransversalsCorresponding AnglesAlternate Interior AnglesTriangle Angle Sum TheoremExterior Angle TheoremTriangle Inequality TheoremSimilar Triangles: AA SimilaritySimilar Triangles: SSS and SAS SimilarityProportions in Similar TrianglesRight Triangle Trigonometry IntroductionTrigonometric Ratios ReviewRadian MeasureConverting Between Degrees and RadiansThe Unit CircleGraphing Sine and CosineGraphing Tangent and Reciprocal Trigonometric FunctionsDerivatives of Trigonometric FunctionsAntiderivativesIndefinite IntegralsBasic Integration RulesRiemann SumsDefinite Integral DefinitionProbability Density Functions and Continuous DistributionsCumulative Distribution FunctionsContinuous Random VariablesNormal DistributionCentral Limit TheoremConfidence Intervals for MeansZ-Tests and T-Tests for MeansOne-Sample Z-Test for MeansOne-Sample and Two-Sample T-TestsOne-Way ANOVAF-Test and Joint SignificanceR-Squared and Model FitOmitted Variable BiasCausal Inference and the Identification ProblemPotential Outcomes and the Rubin Causal ModelTreatment Effects: ATE, CATE, and Heterogeneous Effects

Longest path: 86 steps · 424 total prerequisite topics

Prerequisites (3)

Leads To (1)