Assess Variable Correlations for Policy Tree Analysis

Analyzes correlations among covariates used in policy tree analysis. This helps identify groups of correlated variables that may substitute for each other in tree splits, explaining apparent instability.

Usage

margot_assess_variable_correlation(
  model_results,
  model_name,
  correlation_threshold = 0.5,
  method = c("pearson", "spearman", "kendall"),
  plot = TRUE,
  label_mapping = NULL
)

Arguments

model_results

List returned by margot_causal_forest() (not bootstrap results). Must have been run with save_data = TRUE to access covariate data.

model_name

Character string specifying which model to analyze

correlation_threshold

Numeric threshold for considering variables correlated (default 0.5)

method

Correlation method: "pearson", "spearman", or "kendall" (default "pearson")

plot

Logical: Create correlation heatmap (default TRUE)

label_mapping

Optional named list mapping variable names to labels. If NULL, uses automatic transformation via transform_var_name()

Value

List containing:

• correlation_matrix: Full correlation matrix (with labelled row/column names)

• high_correlations: Pairs of variables with |r| > threshold (with labels)

• correlation_clusters: Groups of intercorrelated variables (with labels)

• summary_text: Narrative summary of findings (using labels)