Create a policy-tree leaf reporting table — margot_table_policy

Returns a tidy leaf table using the package reporting convention: the selected action is shown separately from the signed treatment-control contrast. Positive T-C values favour treatment and negative values favour control.

Usage

margot_table_policy_tree(
  object,
  model_name = NULL,
  depth = NULL,
  weights = NULL,
  digits = 3L,
  ci_level = 0.95,
  label_mapping = NULL,
  source = c("auto", "display_tree", "heldout_cv"),
  include_selected_action_difference = FALSE,
  include_value_contribution = FALSE,
  baseline = c("control_all", "treat_all")
)

Arguments

object: A margot_causal_forest()-style object, or a margot_policy_tree_cv object.
model_name: Optional model name, with or without the model_ prefix. Required for display-tree tables.
depth: Integer tree depth. If NULL for a CV object, the selected depth map is used when available.
weights: Optional evaluation weights for display-tree tables.
digits: Integer; rounding used in formatted columns.
ci_level: Confidence level for display-tree score intervals.
label_mapping: Optional named list used for display labels.
source: Character. "auto" chooses "heldout_cv" for margot_policy_tree_cv objects and "display_tree" otherwise.
include_selected_action_difference: Logical. Include the selected action minus alternative action score contrast. Defaults to FALSE.
include_value_contribution: Logical. Include baseline-dependent value contribution columns. Defaults to FALSE.
baseline: Character. Baseline for optional value contribution columns.

Value

A tibble with one row per reported leaf or held-out leaf-action summary.

Details

Let $\Gamma_{ja}$ denote the action score for observation $j$ under action $a$, let $L$ denote a policy-tree leaf, and let $E_L$ denote the evaluation observations routed to that leaf. For binary actions $C$ and $T$, the reported evaluation-sample contrast is $$ \Delta_L = \frac{\sum_{j \in E_L} w_j\{\Gamma_{jT} - \Gamma_{jC}\}} {\sum_{j \in E_L} w_j}. $$ The selected action is reported separately. For the fitted tree, it is the action stored in the terminal node, learned on the training observations $S_L$ routed to that node: $$ \pi(L) = \arg\max_{a \in \{C,T\}} \frac{\sum_{j \in S_L} w_j \Gamma_{ja}}{\sum_{j \in S_L} w_j}. $$ Held-out CV tables therefore report actions learned on training folds and signed contrasts computed on held-out evaluation rows; they do not reselect actions from held-out means. Between-leaf differences in $\Delta_L$ describe variation in the magnitude of the score contrast; they are not the decision rule used by the policy tree.