Hello!
Are there any plans to release crosswalk files that link 1990 and 2000 blocks to 2020 blocks?
I’ve used the recommended block-to-block method to crosswalk Hispanic population counts from 1990 and 2000 to 2010 boundaries. In theory, as an initial approach, would it be possible to first crosswalk 1990 and 2000 data to 2010 boundaries, and then crosswalk these to 2020 boundaries by applying 2010-to-2020 weights to the 1990 and 2000 data that was aligned with 2010 boundaries?
Thank you!
Yes, we do plan to add these crosswalks in the future, but we don’t have a timeline set for that yet.
In the meantime, the method you describe should work well in most cases.
(That method can perform poorly in cases where a 2010 block is larger than both the source 1990 or 2000 blocks and the target 2020 block. In such cases, the method will aggregate data up from smaller blocks to the larger 2010 block and then disaggregate back down to smaller 2020 blocks, increasing the potential for misallocations. But I expect that this case doesn’t happen very often.)
Hi,
Great, thank you for the prompt response!
I’ll keep this in mind as I test the crosswalks for 2020. Would you recommend any specific adjustments or checks to help mitigate this issue in cases where it may arise?
First, you could identify the cases where the problem may arise by using the existing crosswalks to identify cases where a single 2010 block corresponds to multiple 1990 or 2000 source blocks and to multiple 2020 target blocks.
Second, for a mitigation strategy, you might consider how we developed our 1990-to-2010 crosswalks, which was a similar setting. That strategy makes use of the approach you described, which we call “indirect overlay” (crosswalking through an intermediate zonal definition). The section on indirect overlay describes its advantages and disadvantages. As an alternative approach, we applied “constrained direct overlay” and then combined that with indirect overlay through a “balanced overlay”.
I expect we’ll implement something similar for crosswalks to 2020 blocks, along with some simplifications to the (admittedly complex) weighting we use to balance the indirect and constrained direct overlay. That said, I can’t say how important all this modeling is! If your analytical unit is something larger than blocks (e.g., block groups or census tracts, etc.), then I expect it will generally make little difference whether you use indirect, constrained direct, or balanced overlay, given that, as you aggregate up to larger units, any errors in individual block allocations become smaller in size in relative terms and a have a smaller chance of affecting the aggregated statistics.
Note also that the Census Bureau’s new privacy protections for 2020 census data mean that 2020 census block data are considerably less reliable than in past years, and it’s uncertain when or how they’d be reliably fit for use.
Hi Jonathan,
Thank you for your detailed response. Most of the research is at the tract level or higher but wanted to follow the recommendation to crosswalk at block-to-block and I’ll be mindful of these potential issues going forward.
Thank you again!