From raw public data to the stories that change how decisions are made.
PDI generates narrative documents from tract-level data. This excerpt was produced by the pipeline for a real census tract in south Madison.
Maria Garcia wakes at 5:40. Her shift at the food processing plant starts at 7:00. Her two children, ages 6 and 9, attend an elementary school 2.3 miles away. There is no before-school childcare in her census tract — it is designated a childcare desert, with zero licensed providers within walking distance.2 The nearest bus stop is a 12-minute walk, and during the AM peak, buses arrive fewer than 2 times per hour.3
Maria earns $38,000. Her neighborhood’s median household income is $42,000, placing it in the lowest income bin in the Atlas.4 Schools serving neighborhoods below $45,000 show an average chronic absence rate of 27.8% — compared to 17.6% in neighborhoods above $100,000. The infrastructure is the difference.
Each card maps a policy position to tract-level data — identifying which communities would benefit most, grounded in actual indicators, not composite scores.
·
Highest need:
Six methods. Each one grounded in peer-reviewed research. Each one solving a specific problem that simpler approaches get wrong.
The Census Bureau’s American Community Survey estimates come with a margin of error. At the tract level, a poverty rate of 12.4% might have a true rate anywhere from 4% to 20%. Acting on that number as if it were precise is acting on noise.
PDI flags this automatically. You see the confidence before you see the conclusion. CV < 0.15 = high, 0.15–0.30 = moderate, > 0.30 = low.
Most tools combine indicators into a single “vulnerability score.” The CDC Social Vulnerability Index predicts only 38.9% of COVID case variability (PMC, 2022). The Area Deprivation Index, when unstandardized, is 98.8% explained by just 2 of its 17 variables (Health Affairs Scholar, 2023). Environmental composite rankings shift by an average of 45 places across alternative specifications (PMC, 2023).
Is it poverty? Housing? Transit? Health? The score does not say. A program officer cannot tell whether housing is the problem.
Transit is the crisis. Housing burden is severe. Diabetes is low. A housing program and transit investment would reach this tract. A health clinic would not. The composite told you none of this.
Race and income are too correlated in the US to “control for” one while studying the other. The ICE measures both simultaneously:
Unlike a composite index, ICE measures a specific, named phenomenon: spatial polarization between privilege and deprivation. It does not average unrelated dimensions.
With 50 indicators, which ones belong together? Intuition says “poverty and education are both socioeconomic.” But they may not move together in every geography. EFA examines the correlations and finds clusters that actually co-occur.
None assumed in advance. An analyst who averaged poverty and vehicle access would be mixing independent dimensions. The average would mean nothing.
Drawing a line at the 80th percentile is arbitrary. Move it to the 75th and you add dozens of tracts. LISA identifies clusters by actual spatial pattern — which tracts are surrounded by tracts like them.
The HH cluster is where policy should focus — not because an analyst drew a line, but because the pattern is statistically significant.
Policy needs thresholds: “at what poverty rate does chronic absence spike?” Segmented regression finds the breakpoint.
Below $100K median household income, chronic absence rises sharply — a 10.4pp gap at the threshold. Above $100K, attendance stabilizes.
This is directly actionable. A composite score cannot produce it because it has already averaged income with a dozen other variables.
| Source | Category | Status | Level |
|---|---|---|---|
| Census ACS 5-Year | Demographic | Live | Tract |
| TIGER/Line | Geographic | Live | All |
| CDC PLACES | Health | Live | Tract |
| EPA EJScreen | Environment | Live | Tract |
| USDA Food Access | Food | Live | Tract |
| BLS LAUS | Employment | Live | County |
| WI DPI | Education | Live | District |
| HUD CHAS | Housing | Planned | Tract |
| Eviction Lab | Housing | Planned | Tract |
| HRSA HPSA | Health | Planned | County |
| GTFS | Transit | Planned | Route |
| NCES | Education | Planned | School |
| Platform | Open Source | National | API | Narrative | Raw-First |
|---|---|---|---|---|---|
| Census Reporter | Yes | Yes | Yes | No | Yes |
| COI 3.0 | Docs | Yes | Download | No | Composite |
| National Equity Atlas | No | Metro | No | No | Dashboard |
| Opportunity Insights | Code | Yes | Download | No | Yes |
| PDI | Yes | Yes | REST+SSE | Go tmpl | Yes |
PDI began as the Madison Equity Atlas — a 22-layer GIS platform analyzing 125 census tracts in Dane County, Wisconsin. The Atlas produced Five Mornings in Madison, 70 evidence cards, a founding partnership proposal, and a field guide for decision makers.
The Atlas proved that when tract-level data is structured right, it produces stories that move people and evidence that informs campaigns. PDI generalizes that methodology to run for any county in America.
Go + Python + PostgreSQL/PostGIS · 12 REST endpoints + SSE streaming · Apache-2.0
GitHub → DojoGenesis/policy-data-infrastructure