Social Vulnerability and Health Outcomes in America — Phase 1

Author: Mohammad Haider Date: April 2026 Phase: 1 of 4 — Foundation

---

The Question

If we know how socioeconomically vulnerable a community is, how well can we predict the health of the people who live there?

Health outcomes in America vary dramatically across geography. A diabetes diagnosis, a mental health crisis, or a chronic disease isn't just a medical event — it often correlates strongly with where a person lives and the conditions of their community.

This analysis joins two CDC datasets at the census tract level — the Social Vulnerability Index (SVI) and PLACES — to quantify that relationship.

What This Notebook Does

1. Downloads SVI and PLACES data from CDC (one-time, cached locally)
2. Joins them on census tract FIPS code
3. Computes correlations between vulnerability dimensions and health outcomes
4. Builds a national choropleth showing the geography of the disparity
5. Surfaces state-level patterns

Caveats Up Front

• This is observational data. We can describe associations but not causation.
• Both datasets are estimates with their own uncertainty bounds.
• Census tract is a coarse unit — there's variation within tracts we can't see.

Setup

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from pathlib import Path
import requests

# Display settings
pd.set_option('display.max_columns', 50)
pd.set_option('display.width', 200)
sns.set_theme(style='whitegrid', palette='deep')
plt.rcParams['figure.dpi'] = 100

# Project paths
PROJECT_ROOT = Path.cwd().parent if Path.cwd().name == 'notebooks' else Path.cwd()
DATA_RAW = PROJECT_ROOT / 'data' / 'raw'
DATA_PROCESSED = PROJECT_ROOT / 'data' / 'processed'
OUTPUTS = PROJECT_ROOT / 'outputs'

for p in [DATA_RAW, DATA_PROCESSED, OUTPUTS / 'figures']:
    p.mkdir(parents=True, exist_ok=True)

print(f'Project root: {PROJECT_ROOT}')
print(f'Pandas version: {pd.__version__}')

Project root: c:\Users\haidm\Desktop\social-impact-analysis
Pandas version: 3.0.2

Step 1: Download the Data

SVI 2022 (most recent): A CSV with 16 vulnerability indicators per census tract, plus four summary themes (socioeconomic, household, racial/ethnic minority, housing/transportation) and an overall ranking.

PLACES 2024 release (most recent): Estimated prevalence of 36 chronic conditions and behaviors per census tract.

Both datasets are public, no API key needed. We cache locally so we only download once.

Important: URLs occasionally change. If a download fails, check:

• SVI: https://www.atsdr.cdc.gov/placeandhealth/svi/data_documentation_download.html
• PLACES: https://www.cdc.gov/places/index.html

# CDC datasets are downloaded manually from:
#   SVI:    https://svi.cdc.gov/dataDownloads/data-download.html
#   PLACES: https://data.cdc.gov/500-Cities-Places/PLACES-Local-Data-for-Better-Health-Census-Tract-D/cwsq-ngmh
# Saved locally so the notebook is fully reproducible.

svi_path = DATA_RAW / 'svi_2022.csv'
places_path = DATA_RAW / 'places_2024.csv'

assert svi_path.exists(), f'Missing: {svi_path}. See URLs in cell comment above.'
assert places_path.exists(), f'Missing: {places_path}. See URLs in cell comment above.'

print(f'SVI: {svi_path.stat().st_size / 1e6:.1f} MB')
print(f'PLACES: {places_path.stat().st_size / 1e6:.1f} MB')

SVI: 61.1 MB
PLACES: 867.9 MB

Step 2: Load and Inspect SVI

# Load SVI - keep FIPS as string to preserve leading zeros
svi = pd.read_csv(svi_path, dtype={'FIPS': str, 'STCNTY': str})
print(f'SVI shape: {svi.shape}')
svi.head(3)

SVI shape: (84120, 158)

	ST	STATE	ST_ABBR	STCNTY	COUNTY	FIPS	LOCATION	AREA_SQMI	E_TOTPOP	M_TOTPOP	E_HU	M_HU	E_HH	M_HH	E_POV150	M_POV150	E_UNEMP	M_UNEMP	E_HBURD	M_HBURD	E_NOHSDP	M_NOHSDP	E_UNINSUR	M_UNINSUR	E_AGE65	...	M_ASIAN	E_AIAN	M_AIAN	E_NHPI	M_NHPI	E_TWOMORE	M_TWOMORE	E_OTHERRACE	M_OTHERRACE	EP_NOINT	MP_NOINT	EP_AFAM	MP_AFAM	EP_HISP	MP_HISP	EP_ASIAN	MP_ASIAN	EP_AIAN	MP_AIAN	EP_NHPI	MP_NHPI	EP_TWOMORE	MP_TWOMORE	EP_OTHERRACE	MP_OTHERRACE
0	1	Alabama	AL	01001	Autauga County	01001020100	Census Tract 201; Autauga County; Alabama	3.793569	1865	368	733	114	700	114	402	184	19	19	125	74	205	78	153	89	363	...	65	0	13	0	13	103	75	0	13	8.1	7.2	11.2	7.3	4.3	3.8	2.4	3.4	0.0	2.0	0.0	2.0	5.5	4.2	0.0	2.0
1	1	Alabama	AL	01001	Autauga County	01001020200	Census Tract 202; Autauga County; Alabama	1.282174	1861	396	680	97	544	101	239	144	51	39	95	56	129	48	129	85	293	...	13	0	13	0	13	135	119	8	16	21.5	10.9	56.0	11.2	0.1	0.3	0.0	2.0	0.0	2.0	0.0	2.0	7.3	6.3	0.4	0.9
2	1	Alabama	AL	01001	Autauga County	01001020300	Census Tract 203; Autauga County; Alabama	2.065364	3492	593	1431	213	1305	227	773	335	33	38	313	128	244	114	175	97	470	...	19	0	13	0	13	148	182	0	13	10.3	5.9	25.1	13.0	1.3	1.7	0.3	0.5	0.0	1.1	0.0	1.1	4.2	5.2	0.0	1.1

3 rows × 158 columns

# The columns we care about for Phase 1:
# - FIPS: census tract identifier
# - STATE, COUNTY: human-readable location
# - RPL_THEMES: overall percentile ranking (0-1, higher = more vulnerable)
# - RPL_THEME1-4: percentile rankings for each theme
#     Theme 1: Socioeconomic Status
#     Theme 2: Household Characteristics
#     Theme 3: Racial & Ethnic Minority Status
#     Theme 4: Housing Type & Transportation

svi_cols = ['FIPS', 'STATE', 'COUNTY', 'E_TOTPOP',
            'RPL_THEMES', 'RPL_THEME1', 'RPL_THEME2', 'RPL_THEME3', 'RPL_THEME4']

svi_clean = svi[svi_cols].copy()
svi_clean.columns = ['fips', 'state', 'county', 'population',
                     'svi_overall', 'svi_socioeconomic', 'svi_household',
                     'svi_minority', 'svi_housing']

# CDC uses -999 to indicate missing data
for col in ['svi_overall', 'svi_socioeconomic', 'svi_household', 'svi_minority', 'svi_housing']:
    svi_clean[col] = svi_clean[col].replace(-999, np.nan)

print(f'Tracts with valid overall SVI: {svi_clean["svi_overall"].notna().sum():,}')
svi_clean.describe()

Tracts with valid overall SVI: 83,342

	population	svi_overall	svi_socioeconomic	svi_household	svi_minority	svi_housing
count	84120.000000	83342.000000	83342.000000	83358.000000	83617.000000	83359.000000
mean	3936.015133	0.499994	0.499987	0.499978	0.499391	0.499965
std	1731.508866	0.288680	0.288681	0.288680	0.288881	0.288711
min	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000
25%	2718.000000	0.250000	0.250000	0.250000	0.249400	0.250000
50%	3761.000000	0.500000	0.499950	0.500000	0.499400	0.500000
75%	4946.000000	0.750000	0.750000	0.749975	0.749800	0.750000
max	38907.000000	1.000000	1.000000	1.000000	0.995400	1.000000

Step 3: Load and Inspect PLACES

PLACES comes in long format — one row per (tract, measure) pair. We'll need to pivot it to wide format for analysis.

places_raw = pd.read_csv(places_path, dtype={'LocationID': str})
print(f'PLACES shape (long format): {places_raw.shape}')
print(f'Unique measures: {places_raw["MeasureId"].nunique()}')
places_raw[['LocationID', 'MeasureId', 'Measure', 'Data_Value']].head(5)

PLACES shape (long format): (3047284, 24)
Unique measures: 40

	LocationID	MeasureId	Measure	Data_Value
0	01073010900	DISABILITY	Any disability among adults	44.9
1	01073011207	DEPRESSION	Depression among adults	22.7
2	01077010100	DISABILITY	Any disability among adults	40.3
3	01081040202	ARTHRITIS	Arthritis among adults	23.5
4	01081040300	OBESITY	Obesity among adults	35.5

# For Phase 1, focus on five flagship outcomes that span chronic disease,
# mental health, and behavioral risk factors
FOCUS_MEASURES = {
    'DIABETES': 'Diabetes',
    'MHLTH': 'Poor Mental Health',
    'CHD': 'Coronary Heart Disease',
    'OBESITY': 'Obesity',
    'CSMOKING': 'Current Smoking',
    'BPHIGH': 'High Blood Pressure'
}

# Pivot to wide: one row per tract, one column per measure
places = (places_raw[places_raw['MeasureId'].isin(FOCUS_MEASURES.keys())]
          .pivot_table(index='LocationID',
                       columns='MeasureId',
                       values='Data_Value',
                       aggfunc='first')
          .reset_index()
          .rename(columns={'LocationID': 'fips'}))

print(f'PLACES wide shape: {places.shape}')
places.head(3)

PLACES wide shape: (78815, 7)

MeasureId	fips	BPHIGH	CHD	CSMOKING	DIABETES	MHLTH	OBESITY
0	01001020100	41.4	7.2	15.6	13.3	17.9	39.4
1	01001020200	45.9	6.5	16.7	15.8	18.5	44.7
2	01001020300	42.9	7.3	16.1	13.9	18.6	40.3

Step 4: Join the Datasets

Both datasets use 11-digit FIPS codes. We do an inner join to keep only tracts present in both.

df = svi_clean.merge(places, on='fips', how='inner')
print(f'Joined dataset: {df.shape[0]:,} census tracts')
print(f'Population covered: {df["population"].sum():,}')
df.head(3)

Joined dataset: 78,815 census tracts
Population covered: 313,597,796

	fips	state	county	population	svi_overall	svi_socioeconomic	svi_household	svi_minority	svi_housing	BPHIGH	CHD	CSMOKING	DIABETES	MHLTH	OBESITY
0	01001020100	Alabama	Autauga County	1865	0.3635	0.4880	0.6149	0.3810	0.1478	41.4	7.2	15.6	13.3	17.9	39.4
1	01001020200	Alabama	Autauga County	1861	0.4155	0.4409	0.0628	0.7545	0.6461	45.9	6.5	16.7	15.8	18.5	44.7
2	01001020300	Alabama	Autauga County	3492	0.4843	0.4160	0.7481	0.4773	0.3658	42.9	7.3	16.1	13.9	18.6	40.3

df.to_csv(DATA_PROCESSED / 'svi_places_joined.csv', index=False)
print(f'Saved to {DATA_PROCESSED / "svi_places_joined.csv"}')

Saved to c:\Users\haidm\Desktop\social-impact-analysis\data\processed\svi_places_joined.csv

Step 5: The Headline Finding — Correlations

How strongly does each SVI theme correlate with each health outcome? Spoiler: very strongly, but the strength varies by condition.

svi_cols_list = ['svi_overall', 'svi_socioeconomic', 'svi_household', 'svi_minority', 'svi_housing']
outcome_cols = list(FOCUS_MEASURES.keys())

corr = df[svi_cols_list + outcome_cols].corr().loc[svi_cols_list, outcome_cols]

# Pretty labels
corr.index = ['Overall SVI', 'Socioeconomic', 'Household', 'Minority Status', 'Housing/Transport']
corr.columns = [FOCUS_MEASURES[c] for c in corr.columns]

fig, ax = plt.subplots(figsize=(11, 5))
sns.heatmap(corr, annot=True, fmt='.2f', cmap='RdBu_r', center=0,
            vmin=-1, vmax=1, cbar_kws={'label': 'Pearson correlation'},
            ax=ax, linewidths=0.5)
ax.set_title('Social Vulnerability vs. Health Outcomes\n(Pearson correlation across ~73,000 US census tracts)',
             fontsize=13, pad=15)
ax.set_ylabel('SVI Component')
ax.set_xlabel('Health Outcome')
plt.tight_layout()
plt.savefig(OUTPUTS / 'figures' / 'correlation_heatmap.png', dpi=150, bbox_inches='tight')
plt.show()

What this shows

TODO after running: Write 2-3 sentences interpreting the heatmap. Look for:

• Which SVI theme correlates most strongly with each outcome?
• Are there outcomes that DON'T track with vulnerability? (Often interesting to flag.)
• Any sign-flipping (negative correlations) that needs explanation?

Step 6: The Headline Scatter Plot

One image, one finding. Pick the highest-correlation pair and show it cleanly.

# Diabetes vs. SVI - typically the strongest signal
fig, ax = plt.subplots(figsize=(10, 6))

ax.scatter(df['svi_overall'], df['DIABETES'],
           alpha=0.08, s=8, color='#1f4e79', edgecolors='none')

# Trend line via rolling median (more robust than regression on this scale)
tmp = df[['svi_overall', 'DIABETES']].dropna().sort_values('svi_overall')
tmp['rolling_median'] = tmp['DIABETES'].rolling(2000, center=True, min_periods=500).median()
ax.plot(tmp['svi_overall'], tmp['rolling_median'],
        color='#c8102e', linewidth=3, label='Rolling median')

ax.set_xlabel('Social Vulnerability Index (percentile)', fontsize=12)
ax.set_ylabel('Diabetes Prevalence (%)', fontsize=12)
ax.set_title('More vulnerable communities have substantially higher diabetes rates',
             fontsize=14, pad=15)
ax.legend(loc='upper left')
ax.grid(True, alpha=0.3)

# Annotate the headline number
low_svi_diabetes = df[df['svi_overall'] < 0.2]['DIABETES'].median()
high_svi_diabetes = df[df['svi_overall'] > 0.8]['DIABETES'].median()
ax.axhline(low_svi_diabetes, color='gray', linestyle='--', alpha=0.4)
ax.axhline(high_svi_diabetes, color='gray', linestyle='--', alpha=0.4)
ax.text(0.02, low_svi_diabetes + 0.3, f'Low-vulnerability tracts: {low_svi_diabetes:.1f}%',
        fontsize=10, color='gray')
ax.text(0.98, high_svi_diabetes - 0.5, f'High-vulnerability tracts: {high_svi_diabetes:.1f}%',
        fontsize=10, color='gray', ha='right')

plt.tight_layout()
plt.savefig(OUTPUTS / 'figures' / 'diabetes_vs_svi.png', dpi=150, bbox_inches='tight')
plt.show()

print(f'\nDiabetes prevalence in low-vulnerability tracts: {low_svi_diabetes:.1f}%')
print(f'Diabetes prevalence in high-vulnerability tracts: {high_svi_diabetes:.1f}%')
print(f'Ratio: {high_svi_diabetes / low_svi_diabetes:.2f}x')

Diabetes prevalence in low-vulnerability tracts: 9.5%
Diabetes prevalence in high-vulnerability tracts: 15.6%
Ratio: 1.64x

Step 7: State-Level Disparities

National averages obscure regional patterns. Where is the SVI-health relationship sharpest?

# For each state, compute the within-state correlation between SVI and diabetes
state_corr = (df.dropna(subset=['svi_overall', 'DIABETES'])
                .groupby('state')
                .apply(lambda g: g['svi_overall'].corr(g['DIABETES']))
                .sort_values(ascending=True)
                .rename('correlation')
                .reset_index())

fig, ax = plt.subplots(figsize=(9, 12))
colors = ['#c8102e' if c > 0.7 else '#1f4e79' if c > 0.5 else '#888' for c in state_corr['correlation']]
ax.barh(state_corr['state'], state_corr['correlation'], color=colors)
ax.set_xlabel('Within-state correlation: SVI vs. Diabetes prevalence', fontsize=11)
ax.set_title('States where vulnerability most strongly predicts diabetes',
             fontsize=13, pad=15)
ax.axvline(0.5, color='gray', linestyle='--', alpha=0.4, linewidth=1)
ax.axvline(0.7, color='gray', linestyle='--', alpha=0.4, linewidth=1)
ax.grid(True, axis='x', alpha=0.3)
plt.tight_layout()
plt.savefig(OUTPUTS / 'figures' / 'state_correlations.png', dpi=150, bbox_inches='tight')
plt.show()

Phase 1 Summary

TODO after running: Write 4-6 sentences summarizing what we learned.

Suggested structure:

1. The headline number (e.g., "diabetes rates are X.Xx higher in the most vulnerable tracts")
2. Which vulnerability dimension matters most
3. Where the disparity is sharpest geographically
4. The explicit caveat about correlation vs. causation
5. What Phase 2 will explore

Next Steps (Phase 2)

• Build interactive choropleth map with Folium/Plotly
• Explore "twin tracts" — pairs with similar SVI but divergent outcomes
• Add urban/rural dimension (CDC's NCHS Urban-Rural Classification)
• Sensitivity analysis: do findings hold when we control for population age?