MIGRATE: A New Dataset of Annual Migration

Pretty slick research from a collaboration between some folks at Cornell and Berkeley just landed in Nature last Friday.

For a long time, researchers have been stuck using county-to-county data in the United States, which is very coarse. And proprietary data sets are often not available to researchers or show high bias. This paper introduces MIGRATE, a new dataset that covers 2010–2019 and captures flows between 47.4 billion pairs of Census Block Groups (CBGs) by anchoring proprietary/commercial data to Census population counts.

To understand the scale that's achieved with this new dataset, it helped me to look at the hierarchy of US Census Bureau measurement:

County: The standard unit for migration data. Counties can be massive (LA County has ~10 million people).

Census Tract: A subdivision of a county.

Census Block Group (CBG): The unit used in this study. It is a cluster of blocks that generally contains between 600 and 3,000 people. There are about 217,000 of these groups in the US.

Census Block: The smallest unit (equivalent to a city block bounded by streets).

The researchers claim that their MIGRATE dataset is approximately 4,600 times more granular than the publicly available 5-year county-level data and 18 million times more granular than state-level data. Crazy stuff.

This allows researchers to see hyper-local patterns that were previously invisible. A perfect example they mention is the "Climate Retreat" from California wildfires in 2017 and 2018 (Tubbs and Camp fires). If you looked only at standard county-level data, you would see almost flat out-migration rates. You might conclude that these massive fires didn't displace anyone.

But 77% of movers from the Tubbs fire and 54% from the Camp fire moved to other CBGs within the same county. Population declines were also "260% larger in magnitude for the Tubbs fire and 40% larger in magnitude for the Camp fire than those visible at the 5-year ACS level."

This paper should help a lot of US-based researchers, especially for those looking at climate-based displacement. For example, I covered a redfin article a while back on net outflows in flood-prone areas and they used inter-county data. Reports like this may have vastly underestimated climate displacement because it masks short-distance retreats.