Heterogeneous Treatment Effect Causal Impact Analysis
Also known as: HTE-CausalImpact, CATE causal impact, heterogeneous causal impact, subgroup causal impact analysis
Heterogeneous treatment effect causal impact analysis extends the Bayesian structural time-series causal impact framework to estimate not just the average effect of an intervention but how that effect varies across subgroups or individual units. By combining counterfactual prediction with conditional average treatment effect (CATE) estimation, it reveals which groups benefit most or least from an intervention.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use this method when you have a well-defined intervention affecting a time series or panel of units, and you suspect that the treatment effect differs meaningfully across subgroups defined by pre-treatment characteristics such as demographics, geography, or baseline outcomes. It is appropriate when you have sufficient pre-intervention data to fit a credible control model and when units have comparable background covariates enabling CATE estimation. Do not use it when pre-intervention series are too short for stable Bayesian estimation, when subgroup sample sizes are very small, or when subgroup definitions are chosen post-hoc without multiplicity correction, as these conditions severely inflate false-discovery risk.
Strengths & limitations
- Reveals treatment effect heterogeneity that a single aggregate estimate conceals, enabling targeted policy decisions.
- The Bayesian structural time-series backbone provides credible uncertainty quantification for the counterfactual at each time point.
- Flexibility to incorporate modern machine-learning CATE estimators (causal forests, BART) within the impact analysis pipeline.
- Applicable to observational data without a randomised control group, relying on parallel control series instead.
- Posterior predictive intervals naturally propagate uncertainty rather than relying on asymptotic approximations.
- Requires sufficient pre-intervention data — typically at least 52 observations for weekly series — to identify a stable structural time-series model.
- CATE estimation demands adequate sample size per subgroup; thin subgroups yield wide, uninformative credible intervals.
- The counterfactual validity depends on the quality of control covariates; if important predictors are missing, bias enters through the BSTS model.
- Combining two modelling layers (BSTS counterfactual + CATE learner) compounds model specification error and increases computational cost.
- Subgroup comparisons require multiplicity correction; without it, spurious heterogeneity is easily overreported.
Frequently asked
How is this different from standard causal impact analysis?
Standard causal impact analysis produces a single aggregate estimate of the intervention effect averaged over all units. The heterogeneous version additionally estimates how that effect varies across subgroups or individuals, using CATE estimators applied to the unit-level residual impacts from the BSTS counterfactual.
What CATE estimator should I use alongside the BSTS model?
Causal forests (grf package in R) and doubly robust learners are the most common choices because they are flexible, handle high-dimensional covariates, and provide honest uncertainty estimates. BART is also popular in Bayesian settings where full posterior inference on subgroup effects is desired.
How much pre-intervention data do I need?
As a practical rule, at least one full seasonal cycle of pre-intervention data is recommended for the BSTS model to capture trend and seasonality reliably. For weekly data that means roughly 52 weeks; for daily data, at least several months. Shorter windows risk underfitting the counterfactual model.
How should I handle multiple subgroup comparisons?
Pre-specify subgroups based on substantive theory before seeing post-intervention data, and apply a multiplicity correction such as Benjamini-Hochberg false-discovery rate control when reporting many subgroup effects. Post-hoc subgroup mining without correction is a major source of spurious findings.
Can I use this without a control time series?
In principle yes — the BSTS model can be fitted using only the pre-intervention portion of the treated series — but the counterfactual quality is much weaker without control series. Including high-quality control predictors is strongly recommended to reduce confounding in the counterfactual estimate.
Sources
- Brodersen, K. H., Gallusser, F., Koehler, J., Remy, N., & Scott, S. L. (2015). Inferring causal impact using Bayesian structural time-series models. Annals of Applied Statistics, 9(1), 247-274. DOI: 10.1214/14-AOAS788 ↗
- Athey, S., & Imbens, G. (2016). Recursive partitioning for heterogeneous causal effects. Proceedings of the National Academy of Sciences, 113(27), 7353-7360. DOI: 10.1073/pnas.1510489113 ↗
How to cite this page
ScholarGate. (2026, June 3). Heterogeneous Treatment Effect Causal Impact Analysis. ScholarGate. https://scholargate.app/en/causal-inference/heterogeneous-treatment-effect-causal-impact-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Causal Impact AnalysisCausal inference↔ compare
- Heterogeneous Treatment Effect Difference-in-DifferencesCausal inference↔ compare
- Interrupted Time SeriesCausal inference↔ compare
- Propensity Score MatchingResearch Statistics↔ compare
- Synthetic Control MethodCausal inference↔ compare