Zip Code Caveat: Bias Due to Spatiotemporal Mismatches Between Zip Codes and US Census–Defined Geographic Areas—The Public Health Disparities Geocoding Project

• DOI: 10.2105/ajph.92.7.1100
• Publication Date: 2008-12-02T00:04:35Z
• Authors: Nancy Krieger; Pamela Waterman; Jarvis T. Chen; Mah-Jabeen Soobader; S. V. Subramanian; Rosa Carson
• Keywords: Geocoding; Zip code; American Community Survey; Code (set theory)

Use of zip codes in US public health research is on the rise. As February 2002, 230 articles were indexed by code PubMed,1 all published since 1989. Fifty-two these (23%) involved use censusderived socioeconomic data (e.g., median household income) to investigate effects position specified outcomes (article citations are available request from authors).

To date, discussions regarding for have focused chiefly whether codes' larger population size (average: 30 000) and potentially greater heterogeneity would attenuate estimates gradients detected using comparison with obtained via census tract (average population: 4000) or block group 1000) data.2–7 Unacknowledged literature, however, fact that differ tracts groups other important ways, including spatiotemporal definition stability.

Unlike tracts, defined Bureau Census as “small, relatively permanent statistical subdivision[s] a county . designed be homogeneous respect characteristics, economic status, living conditions,”8(ppG-10–G-11) “administrative units established United States Postal Service most efficient delivery mail, therefore generally do not political area boundaries.”9(pA-13) Spanning single building company high volume mail large areas cut across states, “carrier routes one may intertwine those more codes” such “this conceptual than geographic.”10

To “overcome difficulties precisely defining land covered each code,”11 created new entity built blocks: 5-digit tabulation (ZCTA), first used 2000 census.12 Of note, ZCTAs sharing same necessarily cover (Table 1 ▶),13 so self-report addresses medical records cannot assumed correspond census-defined ZCTAs.

TABLE 1

—Technical Definitions Distinctions Between Zip Codes Code Tabulation Areas (ZCTAs)

Even before introduction ZCTAs, there 2 types discontinuity could conceivably affect studies linking census-derived data: (1) changes routes—and hence affected code—and (2) discontinuation addition nondecennial years.14–16 1997 2001 alone, Post Office added approximately 390 nationwide discontinued 120 (oral communication, Meg Ausman, Data Center, 5, 2002). One implication persons correctly geocoded did exist preceding decennial census.

Findings Public Health Disparities Geocoding Project17 illustrate potential problems code–census mismatches, even dating creation ZCTAs. This project was assess which area-based measures at levels geography (census tract, group, code) appropriate monitoring inequalities health. states (Massachusetts Rhode Island) 1990 used. Records 1999 firm whose accuracy we ascertained (96%),18 firm, following standard practice, returned recent geocodes available.

Cancer incidence rates addressed. We found Massachusetts (474 listed census), 17 376 (10.4%) 166 730 cancer cases occurring during 1987 1993 193 included census; 15 774 (90.8%) changed after census.19–21 By contrast, Island (70 only 0.7% (148) 19 766 matched census.

In case colon Massachusetts, moreover, impact excluding linked substantial. code–level analyses yielded contrary observed reported literature (Tables 2–4 ▶ ▶).22

TABLE 2

—Incident Colon Cancer Counts Geographic Level: 1987–1993

TABLE 3

—Colon Incidence Rates, Stratified Area-Based Socioeconomic Measures, Among Persons With Least Most Resources, Along Age-Adjusted Comparisons (Incidence Rate Ratio Relative Index Inequality): ...

TABLE 4

—Area-Based Measures Cutpoints Used Analysis

Given growing interest geographic data,23,24 urge researchers, when records, pay careful attention mismatches between well boundaries years established. projects programs should likewise alert issues stemming replacement census.