Lock the execution settings for a Margot LMTP analysis — margot_lmtp_estimator

`margot_lmtp_estimator_spec()` records the arguments that determine an LMTP fit and protects them with a content hash. The specification belongs to Margot and requires no companion package. It deliberately records execution settings alone: the causal question, causal estimand, identification assumptions, and policy rationale remain in the study protocol.

Usage

margot_lmtp_estimator_spec(
  trt,
  outcomes,
  policies,
  seed,
  baseline = NULL,
  time_vary = NULL,
  cens = NULL,
  compete = NULL,
  outcome_type = c("continuous", "binomial"),
  id = NULL,
  folds = 5L,
  bounds = NULL,
  learner_profile = c("ensemble", "glm"),
  trim = 0.999,
  weight_column = NULL
)

Arguments

trt: Character vector naming the exposure at each policy node.
outcomes: Character vector naming the terminal outcomes.
policies: Named logical vector. Each name is a policy arm and each value is the `mtp` setting passed to `lmtp` for that arm. Continuous shifts, including a natural-course arm represented by `shift = NULL`, ordinarily use `TRUE`.
seed: Single whole-number estimation seed.
baseline: Optional character vector of baseline covariates.
time_vary: Optional time-varying covariate specification passed to `lmtp`.
cens: Optional character vector of censoring indicators.
compete: Optional character vector of competing-event indicators.
outcome_type: Outcome model, `"continuous"` or `"binomial"`.
id: Optional participant identifier column.
folds: Number of cross-fitting folds.
bounds: Optional common outcome bounds passed to `lmtp`.
learner_profile: Registered learner profile, `"glm"` or `"ensemble"`.
trim: Pooled density-ratio quantile cap passed to [lmtp::lmtp_control()].
weight_column: Optional data column containing non-negative analysis weights. The values remain in the analysis data rather than the specification.

Value

An object of class `margot_lmtp_estimator_spec`.

Details

The specification can name several terminal outcomes. When it is passed to [margot_lmtp()] with `reuse_density_ratios = TRUE`, Margot fits each policy-specific treatment and censoring density-ratio process once and reuses it across those outcomes.