Skip to contents

Development Version of QINI Analysis with Integrated Interpretation

Source: R/margot_qini_dev.R
margot_qini_dev.Rd

Computes QINI curves and related metrics using the test set from margot_causal_forest_dev(). Integrates functionality from margot_qini() and margot_interpret_qini() into a single coherent function.

Usage

margot_qini_dev(
  cf_results,
  outcome_vars = NULL,
  baseline_method = "maq_no_covariates",
  use_evaluation_forest = TRUE,
  spend_levels = c(0.05, 0.1, 0.2, 0.4),
  n_bootstrap = 200,
  confidence_level = 0.95,
  seed = 12345,
  verbose = TRUE
)

Arguments

cf_results

Results from margot_causal_forest_dev()

outcome_vars

Character vector of outcomes to analyze. NULL = all.

baseline_method

Method for baseline curve: "maq_no_covariates" (default), "simple", "maq_only", "auto", or "none"

use_evaluation_forest

Logical. Use evaluation forest for DR scores if available (default: TRUE)

spend_levels

Numeric vector of budget levels for gain summaries (default: c(0.05, 0.1, 0.2, 0.4))

n_bootstrap

Integer. Bootstrap iterations for inference (default: 200)

confidence_level

Numeric. Confidence level (default: 0.95)

seed

Integer. Random seed (default: NULL)

verbose

Logical. Print progress (default: TRUE)

Value

List of class "margot_qini_dev" containing:

qini_curves

Data frame with QINI curve data for all outcomes

qini_objects

List of maq objects by outcome

gain_summaries

Gain summaries at specified spend levels

qini_metrics

Summary metrics (peak gain, area under curve, etc.)

metadata

Analysis parameters and data info

Details

This function: - Uses the test set from the causal forest for honest evaluation - Can use evaluation forests for computing DR scores if available - Provides multiple baseline methods for comparison - Computes gain summaries at specified budget levels - Includes bootstrap inference for uncertainty quantification

The baseline methods are: - "maq_no_covariates": Uses maq with target.with.covariates = FALSE - "simple": Constant treatment effect assumption - "maq_only": Standard maq baseline - "auto": Tries methods in order until one succeeds - "none": No baseline curve

Examples

if (FALSE) { # \dontrun{
# Generate test data and fit forests
test_data <- margot_simulate_test_data()
cf_results <- margot_causal_forest_dev(
  data = test_data$data,
  outcome_vars = c("Y1", "Y2", "Y3", "Y4"),
  treatment = "A"
)

# Compute QINI curves
qini_results <- margot_qini_dev(cf_results)

# Analyze specific outcomes with custom parameters
qini_results_custom <- margot_qini_dev(
  cf_results,
  outcome_vars = c("Y1", "Y4"),
  baseline_method = "simple",
  spend_levels = c(0.1, 0.25, 0.5)
)
} # }