Fairness Analysis

---

Fairness Question: Does the model perform equally well for high-urbanization areas vs. low-urbanization areas?

Group Definition:

• Group X: High urbanization areas (POPPCT_URBAN ≥ 50%)
• Group Y: Low urbanization areas (POPPCT_URBAN < 50%)

Evaluation Metric: Precision for “Long” outage predictions

Null Hypothesis (H₀): The model is fair across groups - precision scores for ‘Long’ outages are equal regardless of urbanization level.

Alternative Hypothesis (H₁): The model is unfair across groups - precision scores for ‘Long’ outages differ between high and low urbanization areas.

Test Statistic: Difference in precision scores (Group X - Group Y)

Significance Level: α = 0.05

Results:

• Observed difference in precision: -0.1796 (High urbanization areas have lower precision)
• P-value: 0.7018 (from 10,000 permutations)

Conclusion: With a p-value of 0.7018, which is much greater than 0.05, we fail to reject the null hypothesis. There is insufficient evidence to conclude that the model performs unfairly across different urbanization levels. The model appears to achieve fairness parity between high and low urbanization areas for predicting long-duration outages.

---