update

util/evaluation.py

@@ -16,6 +16,12 @@ def calculate_impact_ratio(selection_rates):
     impact_ratios = {category: rate / most_selected_rate for category, rate in selection_rates.items()}
     return impact_ratios
 
+def statistical_parity_difference(selection_rates):
+    """Calculate statistical parity difference."""
+    most_selected_rate = max(selection_rates.values())
+    spd = {category: rate - most_selected_rate for category, rate in selection_rates.items()}
+    return spd
+
 def calculate_four_fifths_rule(impact_ratios):
     """Calculate whether each category meets the four-fifths rule."""
     adverse_impact = {category: (ratio < 0.8) for category, ratio in impact_ratios.items()}
@@ -84,6 +90,7 @@ def statistical_tests(data):
         'Neutral': data['Neutral_Rank'].mean()
     }
     impact_ratios = calculate_impact_ratio(selection_rates)
+    spd_result = statistical_parity_difference(selection_rates)
     adverse_impact = calculate_four_fifths_rule(impact_ratios)
 
     # Friedman test
@@ -107,6 +114,7 @@ def statistical_tests(data):
         **pairwise_results,
         "Levene's Test for Equality of Variances": levene_results,
         "Pairwise Comparisons of Variances": pairwise_variances,
+        "Statistical Parity Difference": spd_result,
         "Impact Ratios": impact_ratios,
         "Adverse Impact (Four-Fifths Rule)": adverse_impact,
     }