Can census tracts from 1980, 1990, and 2000 be linked?

I’m working on a project involving property values at the census tract level across decades. I have the NHGIS shapefiles for these tracts (and 2010 and 2020, though these are less important).

From what I understand, census tracts are designed to be matchable decade to decade, but I’ve only found relationship files for 2000 onwards.

Is there still a way to match 1980 and 1990? An officially correct method is ideal, but something like matching future tracts to the prior tracts their centroids fall into would be acceptable as well. Can you advise on how to approach this?

Thank you,

  • Harry
1 Like

In principle, census tracts are intended to be fairly consistent across time, but there are usually many changes from one census to another, and the changes are not restricted to splits and mergers. The standard approach to this problem is now to use some form of “areal interpolation” to estimate how characteristics for a set of source zones (e.g. 1980 tracts) would be distributed among a set of target zones (e.g., 2010 tracts), and then use the target zones as the analytical units.

NHGIS provides many resources to help with this. For statistics covered on the census “short form” (e.g., age, sex, race/ethnicity, housing tenure), we provide geographically standardized time series tables, which report interpolated 1990-2020 characteristics for 2010 tracts. To standardize data for subjects not covered in our time series tables, we also provide geographic crosswalks designed for allocating data from one year’s units to another’s.

As explained on that crosswalks page, we highly recommend starting from the smallest possible units. E.g., if you want to do an analysis at the tract level, you don’t have to start with tract-level data from all years. Instead, the same data are often available for smaller units such as block groups or blocks, and if you allocate from the smaller units, your estimates can be much more accurate.

The smallest units for which the Census provides data on property value are blocks in 1980 and 1990, block group parts in 2000, and block groups in later years (from the American Community Survey). We have crosswalks for all of these units, except not for 1980. We plan to add crosswalks from 1980 units, but I can’t project a timeline for that yet. We recently started releasing shapefiles and tables for 1980 blocks through this page. Currently, we provide boundaries for a limited set of metro areas, with new releases planned every few months to cover more areas. If your areas of interest are covered in the available shapefiles, I’d recommend trying to spatially overlay the 1980 block boundaries or centroids with 2010 tract boundaries to determine how to allocate 1980 housing counts among 2010 tracts. Then you could use our crosswalks for all other years.

Alternatively, there’s another product called the Longitudinal Tract Database (LTDB) that provides a crosswalk from 1980 tracts to 2010 tracts, but as I said, starting from tract data is often less accurate than starting from blocks, and the LTDB interpolation weights rely heavily on a simple areal-weighting approach that can also reduce accuracy significantly. Still, that’s currently the best “simple” option available to include data from 1980 for many areas… until we’ve added shapefiles and/or crosswalks for smaller 1980 units.

1 Like