Skip to contents

Guides

Source: vignettes/guides.Rmd
guides.Rmd

1. Design Principle

heteff intentionally keeps a small API surface. It does not replace the full grf interface. It standardizes input/output for routine heterogeneous-effect analysis.

2. Estimands

Observational Forest

Target:

τ(x)=E[Y(1)−Y(0)∣X=x] \tau(x) = E[Y(1) - Y(0) \mid X=x]

Survival Forest

Uses grf::causal_survival_forest() and predicts user-selected targets (for example RMST-type quantities through predict arguments).

Instrumental Forest

Conditional local IV quantity:

τIV(x)=Cov(Y,Z∣X=x)Cov(W,Z∣X=x) \tau_{IV}(x) = \frac{\operatorname{Cov}(Y,Z \mid X=x)} {\operatorname{Cov}(W,Z \mid X=x)}

3. Assumptions (High-Level)

Observational setting:

  • no unmeasured confounding after conditioning on baseline covariates
  • overlap in treatment assignment

Instrumental setting:

  • instrument relevance
  • exclusion and monotonicity assumptions as required by the design

Survival setting:

  • coherent censoring and event definitions
  • horizon/target choices aligned with the study question

4. Practical Workflow

  1. Define one analysis data frame.
  2. Fix columns for outcome/treatment/instrument/covariates.
  3. Fit one forest model.
  4. Export as.data.frame(fit) and rank_effects(fit) for downstream reporting.