KRIEGER ET AL. RESPOND

Abstract

As noted by Caretta et al, researchers should be cognizant of diverse types of spatiotemporal discontinuities that may affect the use of geographically delimited population data. Their letter, while alluding to—but never identifying any—minor factual errors in our brief, gives us an opportunity to address confusion about the use of zip codes (ZCs), zip code tabulation areas (ZCTAs), and census-derived area-based socioeconomic measures (ABSMs). We documented a type of bias specifically affecting ZC-linked ABSMs, resulting from changes in ZCs and boundaries between decennial censuses, a complication not affecting census tracts (CTs) or block groups (BGs). At issue is the use of ZC-level ABSMs to characterize the socioeconomic context in which individuals reside, whether for population-based incidence studies, case–control studies, randomized clinical trials, or other study designs. By suggesting that ZC data are no more or less problematic than CT or BG data, the letter by Caretta et al is misleading on 3 counts: Although CT boundaries can change from one census to the next, methods exist to compare CTs over time that cannot be used for ZCs.1,2 For example, CTs split in 2000 can be consolidated to the 1990 tract boundaries; CTs merged in 2000 can be disaggregated to BGs to reconstitute the 1990 CT boundaries; and “supertracts” can be created, incorporating several 1990 CTs or BGs to provide a match to a similar combination of 2000 CTs or BGs.2 These methods cannot be employed with ZCs, precisely because they lack stable census-defined boundaries. Even though a relatively small percentage of ZCs may cross over state lines, the point is that this phenomenon does occur; by contrast, it cannot occur with CTs or BGs.3 The stability of ZCTAs is the same as that of CTs, rendering the suggested positive feature of ZCs’ reflecting population change more quickly than CTs irrelevant to ZCTAs. More important, as we emphasized in our brief, the US Census Bureau has explicitly warned that ZCs cannot be directly matched to ZCTAs: ZCs and ZCTAs having the same numerical code can encompass different areas. Geocoding is thus required to ascertain an address’s ZCTA, thereby removing one often-cited advantage of using postal ZCs. One major implication is that design and use of administrative databases that currently obtain or provide only the ZC for linkage to census-derived ABSMs will need to be reconsidered, given the lack of congruence between ZCs and ZCTAs.