Causal internals

This page documents the causal pipeline for contributors.

For user-facing workflows, prefer causal wrapper entry points (dpmix.causal() / dpmgpd.causal()) over direct mcmc() calls.

Objects

A causal fit typically contains: - treatment-arm fit - control-arm fit - (optional) propensity score model output - derived estimands (ate(), att(), qte(), qtt(), cate(), cqte()) and their summaries

Computation outline

1. Preferred user path: fit via dpmix.causal(..., treat=...) or dpmgpd.causal(..., treat=...).
2. Internal contributor path: build a causal bundle (bundle()), optionally including a PS sub-bundle.
3. Internal contributor path: run MCMC (mcmc()), producing a causal fit object.
4. Compute marginal estimands (ate(), att(), qte(), qtt()) on training data; compute conditional estimands (cate(), cqte()) on newdata if supplied, otherwise training X.

Estimand contract notes

• cate() / cqte() are conditional estimands and require a model fitted with X (they are unavailable for no-X causal fits).
• ate() / att() / qte() / qtt() are marginal estimands and always use training data; any supplied newdata/y is ignored with a warning.
• R-mean is supported for mean-based estimands via type = "rmean" in cate(), ate(), att(), and through ate_rmean().

Important constraint

ImportantGolden rules (do not break)

1. Notation: mixture components use subscript j; restriction time uses L; Cox shorthand is ph.
2. User-facing output: the “GPD flag” must display as TRUE/FALSE (not ON/OFF).
3. Causal legality: the only unused/disallowed combo is SB + normal + GPD = FALSE.

Prereqs

• Required packages and data for this page are listed in the setup chunks above.

Outputs

• This page renders model fits, diagnostics, and summary artifacts generated by package APIs.

Interpretation

• Canonical concept page: Spec And Contracts
• Treat this page as an application/example view and use the canonical page for core definitions.

Next

• Continue to the linked canonical concept page, then return for implementation-specific details.