Nominal weights don't say which inputs actually move the score: a metric that barely differs between districts moves nothing, whatever its weight. To measure realized influence, we fit a linear surrogate model to each score across the 28 districts and explain it with SHAP. Each bar is an input's average absolute SHAP value: how many points it typically pushes a district above or below the average, shown as a share of all inputs combined. Inputs are correlated (e.g. poverty, income, and chronic disease move together), so credit among related inputs is approximate.
The five components are weighted 20% each, which is a judgment call. We re-ranked the districts under 5,000 random alternative weightings (each weight usually between about 5% and 40%). The alternative rankings agree closely with the published one (average Spearman correlation 0.965; 95% of them at least 0.905). The top district stays #1 in 68% of them. A typical district's rank moves within a band of about 5 places. Dropping any one component entirely leaves the ranking at a correlation of 0.931 or higher.
| District | Published rank | Range under other weights (90%) |
|---|---|---|
| Como Park (St. Paul) | 1 | 1–2 |
| Summit Hill (St. Paul) | 2 | 1–3 |
| Highland Park (St. Paul) | 3 | 2–5 |
| Longfellow (Minneapolis) | 4 | 4–8 |
| Macalester - Groveland (St. Paul) | 5 | 3–10 |
| Calhoun Isle (Minneapolis) | 6 | 5–10 |
| Saint Anthony Park (St. Paul) | 7 | 3–11 |
| Northeast (Minneapolis) | 8 | 5–10 |
| Hamline - Midway (St. Paul) | 9 | 5–10 |
| Nokomis (Minneapolis) | 10 | 5–13 |
| Southwest (Minneapolis) | 11 | 8–12 |
| Union Park (St. Paul) | 12 | 7–20 |
| West Seventh - Fort Road (St. Paul) | 13 | 12–16 |
| West Side (St. Paul) | 14 | 12–18 |
| North End (St. Paul) | 15 | 14–18 |
| Greater East Side (St. Paul) | 16 | 12–23 |
| Powderhorn (Minneapolis) | 17 | 13–21 |
| Battle Creek - Conway - Eastview - Highwood Hills (St. Paul) | 18 | 14–23 |
| Payne - Phalen (St. Paul) | 19 | 17–21 |
| Summit - University (St. Paul) | 19 | 14–23 |
| Frogtown (St. Paul) | 21 | 17–22 |
| Dayton's Bluff (St. Paul) | 22 | 20–26 |
| University (Minneapolis) | 23 | 19–26 |
| Near North (Minneapolis) | 24 | 23–27 |
| Camden (Minneapolis) | 25 | 23–28 |
| Central (Minneapolis) | 25 | 20–26 |
| Downtown (St. Paul) | 27 | 18–27 |
| Phillips (Minneapolis) | 28 | 27–28 |
If the score captures things people value, housing prices should reflect them. Home values and rent feed the Economic Profile component, so this uses the other four components only. Across the 28 districts, a higher non-price score goes with higher home values (Spearman 0.734, p < 0.001 by permutation test). Part of that is income, because richer areas both pay more and score higher. Holding household income fixed, the link weakens to 0.373. Rent shows almost no relationship (Spearman 0.111, p = 0.57). The Opportunity component actually runs opposite to rent (-0.303). So the score lines up with what buyers pay, but with 28 districts it's a sanity check, not proof.