Diagnose QINI Gain Discrepancies — margot_qini

Compare QINI gains from plotted data, direct maq calculations, and diff summaries to help identify discrepancies between different calculation methods.

Usage

margot_qini_diagnostic(
  mc_result,
  model_names = NULL,
  spend_levels = 0.1,
  tolerance = 0.01,
  verbose = TRUE
)

Arguments

mc_result: Results from margot_causal_forest or similar with qini_data and qini_objects
model_names: Character vector of models to check (NULL = all models)
spend_levels: Numeric vector of spend levels to check (default: 0.1)
tolerance: Numeric tolerance for flagging discrepancies (default: 0.01)
verbose: Logical for detailed output (default: TRUE)

Value

A tibble comparing gain values from different sources with columns: - model: Model name - spend: Spend level - source: Source of the gain value ("plot_data", "maq_direct", "diff_summary") - cate_gain: CATE gain value - ate_gain: ATE/baseline gain value - difference: CATE - ATE difference - se: Standard error (where available) - discrepancy: Logical flag if values differ by more than tolerance

Examples

if (FALSE) { # \dontrun{
# Run diagnostic on all models
diag_results <- margot_qini_diagnostic(mc_result)

# Check specific models at custom spend levels
diag_results <- margot_qini_diagnostic(
  mc_result,
  model_names = c("model_anxiety", "model_depression"),
  spend_levels = c(0.1, 0.2, 0.5)
)
} # }