The 2023.1 run incorporated revisions to census estimates for 2020. In some places, those estimates created a break in the timeseries with 2019, and that break in the series unduly influenced projected in-migration and out-migration. Our new projections include newly released birth and death data from the CDC, which helps to mitigate the influence of the revised 2020 data. This said, use demographic projections with caution. Because of changes to birth, death, and migration patterns during COVID in 2020 and 2021, it is uncertain whether those new patterns represent a change in trajectory (for example, people leaving larger metropolitan areas and working remotely becomes a new normal), or a temporary aberration, and the next few years will look more like the late 2010’s rather than 2020-2021. With data currently available, it is not yet apparent which trends will hold and which will break, and therefore there is more uncertainty in the projected population.
Revision to ZIP- and tract-level employment methodology
This datarun introduced new methodology for our ZIP and tract employment calculations. Our main source of employment data is QCEW, which publishes down to the county level, and we then model that county-level employment down to the ZIP and tract level. In our previous methodology, zip and tract distribution was based solely on DBUSA. While this distribution was generally in keeping with our previous distribution methodology based on Zip Business Patterns released by the Census, there were still shortcomings to this approach, particularly when there were mismatches in NAICS classification between QCEW and DBUSA.
Our new methodology seeks to address these shortcomings. In our new methodology, ZIP- and tract-level employment is grounded in the employment figures produced by LODES, which publishes data ZCTA and tract-level data at the 2-digit NAICS, providing a much more grounded high level estimates. DBUSA is then used to distribute employment below the 2-digit NAICS, but has much more rigid guidelines provided by the LODES data. Note that we do not attempt to match LODES exactly, since there are differences between LODES and QCEW – rather, we use LODES as a control to inform our distribution of QCEW employment, without matching LODES exactly.