Temptation to Trouble: Data Analysis by ZIP Code

“A child’s course in life should be determined not by the zip code she’s born in” – President Barack Obama, 2014

So exactly what is a ZIP Code, and can we get good data (from the Census Bureau and other major data products) at the ZIP Code level?

ZIP stands for “Zone Improvement Plan.” The United States Postal Service (USPS) introduced these familiar five-digit codes in the 1960s to reduce the amount of labor required to sort mail, thereby keeping the cost of a stamp low. But ZIP Codes have developed a life of their own. People often identify with their ZIP Code; if you’ve spent time in the Atlanta intown neighborhood of Lake Claire, you’ve probably seen one of these bumper stickers:

Figure 1: Lake Claire Bumper Sticker

And anyone who grew up watching TV in the United States in the 1990s can tell you the ZIP Code for Beverly Hills. ZIP Codes today are often used as stand-ins for neighborhood by everyone from policymakers to insurance companies (who have been accused of using them to segment out, and then charge higher rates for, urban customers). It’s tempting, then, to examine social patterns by ZIP Code. But there are a number of problems that complicate such analyses. (And by the way…for any youngsters out there…the Beverly Hills zip’s 90210).

The first issue: because the USPS designed ZIP Codes first and foremost for the ease postal delivery, they do not align well with other geographic units. ZIP Codes typically ignore city and county boundaries. Moreover, the “preferred” city name that USPS assigns to a ZIP Code is the city nearest to the post office. Consider this map, which compares “Atlanta” ZIP Codes (defined as ZIP Codes where USPS reports Atlanta as the preferred city) with Atlanta’s city limits, circa 4^th Quarter 2020, using Caliper Corporation’s ZIP Code file:

Figure 2: City of Atlanta vs. “Atlanta” ZIP Codes

Note that ZIP Code 30339, which lies entirely within Cobb County, has a preferred mailing city of Atlanta, GA (per USPS). Parts of Atlanta ZIP Codes poke into Clayton and Gwinnett counties as well. With the wave of annexations and incorporations we’ve experienced in the last two decades, it’s not unusual for a building to fall within the boundaries of one city yet to have a different city in its “preferred” USPS mailing address. Sandy Springs lobbied the USPS to have the preferred city for its ZIP Codes switched to Sandy Springs about a decade ago, to no avail.

The second issue: ZIP Code boundaries are not stable over time. As zip code boundaries are intended to optimize postal delivery, population shifts can often result in either the creation of new ZIP Codes or in border changes to existing ZIP Codes. Further, new ZIP Codes can be created to respond to complaints regarding the mismatch between community boundaries and ZIP Codes–particularly when a new city is created.

Fixes to the first challenge (bringing ZIP Codes into better alignment with other boundaries) exacerbate the second challenge (the lack of stability over time). In an earlier posting, we saw how the Census Bureau created tracts out of recognition for the need to measure change over time for stable geographies.

A third issue: when zip codes were drawn, there was no attention given to the underlying population (in terms of size or homogeneity)–in the case of Census tracts, considerations of this sort were involved. And while Census tracts were drawn with the input of local experts, ZIP Code design is all top-down. Again, the driving criterion was efficiency of mail delivery.

In response to the demand for data at the ZIP Code level, the Census Bureau has defined what it calls ZIP Code Tabulation Areas or ZCTAs. They are careful to point out that ZCTAs are not true ZIP Codes, but rather “generalized area representations of United States Postal Service (USPS) ZIP Code service areas.”

The main difference is related to how these geographies are drawn. Recall that in an earlier blog post, we discussed Census blocks and how they are drawn. And these blocks are used as the ‘atoms’ from which all other geographic units are formed. By contrast, ZIP Codes are essentially collections of mail carrier routes (that themselves may not be continuous). Consider this example. The Census block bounded by 6th and 7th streets NE to the south and north respectively and by Penn Ave NE and Myrtle St NE on the east and west is served by three different postal routes. Two of these routes belong to 30308, while one belongs to 30309:

Figure 3: Why it is Hard to Translate ZIP Codes to Census Geography: An Example from ZIPs 30308 and 30309

The Census Bureau solves this problem by assigning the ZCTA to the most frequently occurring ZIP Code within the given block, in this case 30308.

But there are other differences as well. Not all ZIP Codes get corresponding ZCTAs, either because they are primarily residential, they are used exclusively for PO boxes, or they do not meet the Census Bureau’s minimum area threshold of two square miles. For example, ZIP Code 30361, shown below in a USPS map, does not have a corresponding ZCTA. Instead, the area is folded into its surrounding ZCTA, 30309.

Figure 4: 30361, a ZCTA-less ZIP Code

Because ZCTAs are drawn to have as much fidelity as possible to their corresponding ZIP Codes, ZCTAs vary wildly in total population per area (per the 2020 Census). By comparison, Census tracts are much more uniform in size:

	ZCTAs	Census Tracts
Units with non-zero population	748	2,784
Median population	6,213	3,719
Mean population	14,320	3,848
Standard deviation of population	17,716	1,469

Unlike ZIP Codes (which can change at any time), ZCTAs remain fixed for a decade. While this allows for some comparison over time, it also means that the correspondence between ZCTAs and their underlying ZIP Codes will degrade over time.

When ZCTA boundaries are revisited each decade, they do change a moderate amount. This map compares 2010 and 2020 era ZCTAs in and near the city of Atlanta:

Figure 5: Comparing 2010 and 2020 Vintage ZCTAs

In sum, Census tracts are superior small-area units to ZIP Codes (and their ZCTA statistical counterparts) when it comes to examining sub-county patterns. Census tracts nest cleanly into counties, they are stable over time, and they are drawn so as to fit statistical parameters in terms of population size. But ZIP Codes retain one very important advantage over Census tracts: no one (well, almost no one) knows what Census tract they live in.