US Census Bureau

What You're Getting

US Census Bureau provides the most granular US demographic and economic data: Decennial Census (10-year), American Community Survey (ACS, 1-year and 5-year estimates), Current Population Survey (CPS), Economic Census (5-year), County Business Patterns (annual), International Trade Statistics, Building Permits, and more. Data is organized by geographic unit (nation, state, county, tract, block group) and demographic/economic variables.

Ingestion Strategy

Use the Census API — free, no authentication required for basic use (key improves rate limits). ACS 5-year data is the workhorse for sub-state geography. Queries specify variables (e.g. B01001_001E for total population), geography (e.g. state, county), and dataset (e.g. acs/acs5).

import requests, pandas as pd

# ACS 5-year: median household income by county
url = "https://api.census.gov/data/2022/acs/acs5"
params = {
    "get": "NAME,B19013_001E",
    "for": "county:*",
    "in": "state:*",
    "key": "your_key"  # optional but recommended
}
r = requests.get(url, params=params)
df = pd.DataFrame(r.json()[1:], columns=r.json()[0])

Rate Limits & API Details

• No key: 500 requests/day; with free key: higher limits
• Key registration: api.census.gov/data/key_signup.html
• API base: api.census.gov/data/{year}/{dataset}
• TIGER/Line shapefiles for geographic boundaries: census.gov/geographies/mapping-files/time-series/geo/tiger-line-file.html

Schema Stability

Variable codes (e.g. B19013_001E) change between ACS releases — the table structure is re-published each year with some changes. Always pull the variable dictionary for the specific year you're loading. Geographic FIPS codes are stable; sub-county geographies can change with Decennial Census redistricting. ACS 5-year estimates are not comparable across overlapping periods — use non-overlapping periods for trend analysis.

Data Quality Gotchas

• Estimates with margins of error: ACS data includes MOE columns (e.g. B19013_001M) — always store and surface these, especially for small geographies
• Geography changes: county and census tract boundaries change with each Decennial Census; use GEOIDs carefully when building time series
• ACS 1-year vs 5-year: 1-year is current but only for areas with 65,000+ population; 5-year covers all geographies but is a rolling average
• Suppression: small cells are suppressed for privacy (shown as -666666666) — handle these explicitly

Tool Compatibility

• Python: census PyPI package; cenpy; direct API calls
• R: tidycensus (excellent, highly recommended)
• Airbyte: no official connector
• TIGER shapefiles: use geopandas for Python spatial analysis

Benchmarking Approach

US total population (ACS 5-year, B01001_001E): should match Census Bureau published estimate within 1%. County count: 3,143. After each December ACS release, verify new year data is available within 5 business days. Cross-check median household income for a known county against Census press release. Alert if MOE-to-estimate ratio exceeds 30% for more than 20% of county rows.