Interprets Qini results for binary and multi-arm treatments, automatically detecting treatment type from input data.
Usage
margot_interpret_qini(
multi_batch,
label_mapping = NULL,
alpha = 0.05,
decimal_places = 2,
model_names = NULL,
spend_levels = c(0.1, 0.4),
include_intro = TRUE,
baseline_method = NULL,
scale = "average"
)Arguments
- multi_batch
List from margot_policy() or margot_qini() with diff_gain_summaries
- label_mapping
Named list mapping model names to readable labels
- alpha
Significance level for confidence intervals (default: 0.05)
- decimal_places
Decimal places for rounding (default: 2)
- model_names
Character vector of models to process (optional)
- spend_levels
Numeric vector of spend levels to analyze (default: 0.1). If requested levels don't exist in the data, the function will use available levels instead.
- include_intro
Logical whether to include explanatory text about CATE and Qini curves (default: TRUE)
- baseline_method
Method for generating baseline when regenerating summaries: "maq_no_covariates" (default if NULL), "auto", "simple", "maq_only", or "none". If NULL, uses the baseline method from the original QINI generation.
- scale
Character string specifying the scale for gains: "average" (default), "cumulative", or "population". This affects how gains are interpreted in the summary.
Value
List with summary_table, qini_explanation, concise_summary, reliable_model_names, reliable_model_ids
Details
The function automatically detects available spend levels in the data. If requested spend levels are not available, it will use the closest available levels and warn the user. To see what spend levels are available, check the names of `multi_batch[[1]]$diff_gain_summaries`.
This function accepts output from either:
margot_policy() - which includes policy trees and Qini results
margot_qini() - which focuses solely on Qini curves and gains