Why it's so hard to match residence addresses to census blocks - and how to fix it.

How does one match addresses to other kinds of geography, for example associating a residence address (a common bit of information) with a parcel map or the boundaries of U.S. census blocks, block groups, and census tracts? While it might seem "easy" to convert an address to a latitude and longitude, and then match that with maps of census block boundaries, it turns out that in a significant number of cases, the residence "address" gets placed outside the correct block, either in a block that is empty of population, or one that is populated but where the address does not belong. These issues are not new, and have been consistently explored ever since widespread use of geocoding technology emerged in the early 2000s. Geocoded addresses have been applied in many areas of research, including for the detection of cancer clusters (health), measurement of environmental inequality (environment), and optimization of business locations (commerce). We are now at a turning point where vast amounts of spatial data are collected and the linking of disparate data is evermore necessary. We join mortality data for 1990 to 2022 from 6 states (California, Colorado, Florida, Massachusetts, Michigan, New Jersey) to census block geometries. With those data we explore the mismatch generated when an address point is located in an implausible census block, a problem we call overlay uncertainty. We (1) estimate error levels associated with this join, (2) reveal systematic problems with the ways spatial locations of addresses are misplaced, and (3) propose corrective measures and estimate the reduction in error that can be accomplished.