Stop using zip codes for geospatial analysis (2019)

What ZIP codes actually are

• Described as mail-sorting constructs, not geographic polygons: abstract sets of delivery points along routes.
• Can be:
  • Non-contiguous areas.
  • A single point (e.g., a large company).
  • A single line (highway routes).
  • Overlapping or ambiguous when forced into polygons.
• They reflect USPS operational structure and logistics, not geography or political boundaries; they also change over time as routes change.

Why ZIP codes are problematic for spatial analysis

• Treating ZIPs as polygons is called a “category error” and can produce misleading results, especially for:
  • Rural vs urban comparisons, where a single ZIP can mix dense town centers with large rural surroundings.
  • Demographic or socioeconomic analysis where internal variation is large.
• Census “ZIP Code Tabulation Areas” (ZCTAs) are only approximations and may include overlaps, missing regions, and temporal drift.
• Relates to the Modifiable Areal Unit Problem: statistics and patterns change with how you draw boundaries.

Arguments that ZIPs are “good enough”

• Widely known by the public, easy to collect on forms, and embedded in addresses.
• Often “uniform enough” and “contiguous enough by travel time” for coarse analyses, marketing, bulk mail, sales-tax lookup, and sports blackouts.
• Seen as a practical first step when you need aggregation but lack more precise spatial data.

Alternatives proposed

• Census units (blocks, tracts, counties, CSAs): better-defined geographies and often population-normalized, but harder to collect from users and to explain.
• Spatial grids like H3:
  • Hexagonal cells with consistent neighbor relationships and tunable resolution.
  • Good for counting people/phenomena in areas and joining disparate datasets.
  • Rectangular lat/long bins are possible but bring design choices and edge issues.
• Exact addresses + geocoding, then aggregating to chosen units.
• Custom regions (e.g., DMAs, custom market areas) built from population + internal data.

Data, privacy, and international issues

• ZIP+4 and Canadian postal codes can be extremely granular, verging on household or building identifiers, raising re-identification risk.
• Postal code concepts differ by country; some boundaries are proprietary or not widely known, and user familiarity varies.
• Address and postal-code validation is messy, with numerous edge cases and conflicting datasets.