Skip to contents

Generate QINI Data for Visualization

Source: R/margot_generate_qini_data.R
margot_generate_qini_data.Rd

This function generates QINI curve data on-demand from a model result. It extracts tau_hat from the model, uses maq to generate the CATE curve, and creates a baseline following various methods.

Usage

margot_generate_qini_data(
  model_result,
  outcome_data,
  treatment,
  weights = NULL,
  baseline_method = c("maq_no_covariates", "auto", "simple", "maq_only", "none"),
  seed = 12345,
  verbose = FALSE,
  treatment_cost = 1,
  x_axis = c("proportion", "budget")
)

Arguments

model_result

A single model result from margot_causal_forest output

outcome_data

The outcome data (Y) for the model

treatment

The treatment assignment vector (W)

weights

Optional weights vector

baseline_method

Method for generating baseline: "maq_no_covariates" (default), "auto", "simple", "maq_only", or "none". See details.

verbose

Logical for verbose output

treatment_cost

Scalar treatment cost per unit. Default 1.

x_axis

Type of x-axis for QINI curves: "proportion" (default) or "budget". "proportion" shows proportion of population treated (0 to 1). "budget" shows budget per unit (0 to treatment_cost), matching maq's visualization.

Value

A list with components: - qini_data: data.frame with columns proportion (or budget), gain, curve - qini_objects: list with cate and ate maq objects (ate may be simplified) - x_axis: the x-axis type used ("proportion" or "budget")

Details

The baseline_method parameter controls how the no-prioritization baseline is generated:

  • "maq_no_covariates": Use maq with target.with.covariates = FALSE (default). Automatically falls back to simple baseline if maq fails.

  • "auto": Try maq with target.with.covariates = FALSE first, fall back to simple baseline if it fails

  • "simple": Always use simple baseline (straight line from (0,0) to (1, mean(tau_hat)))

  • "maq_only": Use standard maq with constant rewards (may fail with no fallback)

  • "none": No baseline curve

The simple baseline represents the expected gain under random allocation, where treating proportion p of units yields gain p * E[tau]. This method always succeeds and provides a robust fallback when maq fails.