| HOLC Grade | Tracts (n) | Median Household Income | Mean All-Ages Poverty Rate (%) | Mean Under-5 Poverty Rate (%) | Under-5 "Child Poverty Premium" (pts) |
|---|---|---|---|---|---|
| A | 14 | $68,217 | 13.2 | 17.3 | 4.1 |
| B | 59 | $62,778 | 17.0 | 24.1 | 7.1 |
| C | 61 | $47,865 | 24.9 | 31.0 | 6.1 |
| D | 58 | $58,365 | 24.5 | 28.2 | 3.7 |
Preliminary Findings: Redlining and Under-5 Poverty in Baltimore City
Background
This post shares preliminary findings from my final project, which asks whether Baltimore City census tracts that received a low (“C” or “D”) grade on the 1937 Home Owners’ Loan Corporation (HOLC) “Residential Security Map” still show worse family economic outcomes today, and specifically whether that gap is larger for young children than for the population as a whole. This builds on an earlier dashboard I built using ZIP-code-level ACS data; this week’s analysis moves to the Census-tract level and merges in the actual historic HOLC grade for each tract, rather than relying on a hand-coded neighborhood classification.
The two data sources behind this analysis are both real, current data, not simulated: (1) 2019-2023 five-year American Community Survey (ACS) estimates pulled directly via tidycensus for all Baltimore City census tracts, and (2) the HOLC-to-2020-Census-tract crosswalk maintained by the Mapping Inequality project (American Panorama / University of Richmond’s Digital Scholarship Lab), which spatially assigns each present-day tract the HOLC grade that covers the largest share of its area. Of Baltimore’s census tracts, 192 have usable HOLC coverage and matching ACS data; those 192 tracts are the analytic sample below. This is genuinely preliminary: the numbers below are simple descriptive comparisons, not yet a regression or formal hypothesis test, and I flag one clear complication in the pattern rather than smoothing over it.
Variables
Dominant HOLC grade is the 1937 redlining grade (A–D) that covers the largest share of a tract’s area, used here as a proxy for historic disinvestment
Under-5 poverty rate is the share of children under age 5 whose family income falls below the federal poverty threshold (not the share of the population that is under 5 – a distinction that matters, since a tract can have lots of poor people without having lots of poor young children specifically).
All-ages poverty rate is the same calculation across the full tract population, included as a comparison point.
Data and Preliminary Results
Output 1: Summary Statistics by HOLC Grade
Output 2: Distribution of Under-5 Poverty Rate by Grade
Output 3: All-Ages vs. Under-5 Poverty Rate by Grade
Interpretation
I noticed two things, even at this preliminary stage. First, the under-5 poverty rate is higher than the all-ages poverty rate in every single HOLC grade. Children under 5 are consistently poorer than their neighborhoods as a whole, which is expected but useful to confirm directly in Baltimore’s own data rather than assume from the national literature. Second, grades A and B behave the way the “long shadow of redlining” hypothesis predicts: median income drops and both poverty measures climb steadily from A to B to C. Grade D breaks that clean pattern because it has a higher median income and lower mean poverty (both all-ages and under-5) than grade C, not the worst outcome as a strict monotonic story would predict.
My working explanation, which I plan to check formally once I add tract geometry back in, is gentrification in a handful of historically D-graded but now high-value tracts near the Inner Harbor and Locust Point/Federal Hill – areas that were graded “Hazardous” in 1937 largely because they were working-class immigrant harbor neighborhoods, and that have since gentrified sharply. A few very high-income tracts in that group could be pulling grade D’s averages up enough to mask a more severe pattern in the majority of D-graded tracts elsewhere in the city. Median income (less sensitive to outliers than the mean) already shows some of this: D’s median income ($58,365) sits below B but still above C, consistent with a mixed population of both very-high-income gentrified tracts and very-low-income tracts within grade D.
Next, I will test whether excluding likely-gentrified high-income outlier tracts from grade D changes the picture, and move from these descriptive comparisons to a regression of under-5 poverty rate on HOLC grade (controlling for tract population), which is the actual hypothesis test (H1/H2) proposed in my prospectus. Even without that formal test yet, the consistent “child poverty premium” across every grade (ranging from 3.7 to 7.1 percentage points) is itself a useful preliminary finding for ShareBaby’s targeting question: neighborhood-level poverty rates alone appear to understate where poor young children specifically are concentrated, in every part of the city regardless of historic grade.