Hi,
I am working on a project about internal migration in the U.S. using the American Community Survey. As I browsed the data, I found the countyfip and migcounty1 variables, which seem to be very useful for generating county-level migration flows. However, it seems that many of the counties in the variables are not identifiable—those are coded as zeros along with N/A cases—for confidentiality reasons and/or technical reasons.
However, I also see that there is a MIGPUMA1 variable without a problem with the unidentifiable cases, but there are many cases where multiple counties consist of a PUMA.
Browsing and Googling, I found this [posting] about a crosswalk from County to PWPUMA and saw that Jonathan Schroeder, thanks to him, uploaded a crosswalk from county to MIGPWPUMA for 2010 (for 2012-onward IPUMS data) definition; the filename is “ipums_migpwpuma10_county_crosswalk.csv”.
I was wondering if I could also get the crosswalk from county to MIGPWPUMA for 2000 definition for 2005-2011 IPUMS data.
Thanks!
Juno Kim
Thanks for this inquiry, Juno. I was able to reconstruct a crosswalk between 2000 counties and MIGPUMAs from some old code I have here. I’m sharing that crosswalk here, and, in case other users need this, I’m also sharing a crosswalk between 2000 counties and Place of Work (PW) PUMAs. (In 2000 definitions, MIGPUMAs do not uniformly match PWPUMAs, unlike the 2010 definitions.)
ipums_migpuma00_county_crosswalk.csv (64.5 KB)
ipums_pwpuma00_county_crosswalk.csv (68.8 KB)
NOTE: These crosswalks are valid only for 2000 definitions of counties. There have been a few substantial changes in county boundaries since 2000, and in those cases, these crosswalks will not give accurate current relationships. I don’t have any crosswalks from current county definitions to the 2000 MIGPUMAs or PWPUMAs.
Dear Jonathan,
Thank you so much for your help!
These crosswalks would help my project a lot.
Thank you for your help and I also appreciate your note, which is also valuable!
JK
Hi Jonathan,
Thank you so much for this. I’m working on a project where I need to compare 2024 definition County FIPS to PWPUMA with data ranging from 2015-2024. Do you suggest I find a crosswalk to have the county FIPS in 2000 definitions and then use the crosswalk? Also I know that PUMA definitions change every 10 years which includes some of my target years, so do you suggest I also crosswalk the more recent years to the 2012 definitions or do you have a more updated crosswalk? Also how does this crosswalk work when counties cross PUMA lines?
I would use the pwcounty FIPS in the ACS data but my research focuses on smaller economies so these are commonly ommitted from the ACS data because they have less than 100,000 workers.
Hi Ian, I’d start here: “my research focuses on smaller economies so these are commonly ommitted from the ACS data because they have less than 100,000 workers.” Consider the significance of this for your research: in general, ACS public-use microdata do not have a sample size adequate to support many types of analysis for areas with fewer than 100,000 residents. As such, it may be that you won’t be able to get statistically meaningful results for “smaller economies” by conducting your analysis using only ACS microdata.
Using a crosswalk will not help with this problem, and in fact, it would only make an analysis even less reliable. A crosswalk from PWPUMAs to counties can only indicate what portion of each PWPUMA’s population lies in each county within the PWPUMA. (That’s the case with the crosswalks I provided earlier.) It provides no further information to tell you exactly which respondents worked in each county. So in a case where a single PWPUMA corresponds to several small-population counties, you’d be dealing with two major sources of uncertainty: first, as noted above, the small sample basis may not be adequate to make accurate inferences about the population of interest, and second, you wouldn’t actually know which of the PWPUMA’s workers worked in a particular county.
You could assign a probability that a given worker worked in a particular county based on the populations of each county in a PWPUMA, and you might improve your model of that probability by using summary data about each county’s workers to impute which microdata records to assign to each county. (See for example the research of Nagle et al. 2014 for a method of using summary data to assign microdata records probabilities of residence in different small areas.) But there’d still be a lot of remaining uncertainty about which workers worked where and, of course, a small sample basis.
On top of these issues, the changes in county and PUMA boundaries within your period of interest add more complications. For example, in 2022, the Census Bureau switched from identifying eight counties in Connecticut to a new set of nine county equivalents with altogether different boundaries. There’s no way to crosswalk from one set to the other in a highly reliable way, so crosswalking across time would add substantially more uncertainty to an analysis. Similar problems exist for changes between the 2010 and 2020 PWPUMAs.
I’ll close by linking to a few resouces that could help with crosswalking, if you choose to go that route:
- Geocorr provides crosswalks between 2020 counties and 2020 PUMAs (as used in 2022 and later ACS microdata). You can also use it to get crosswalks to “Connecticut planning regions,” the new county equivalents in use starting in 2022.
- IPUMS USA provides composition files for 2020 PWPUMAs and MIGPUMAs indicating which counties and which PUMAs are in each PWPUMA/MIGPUMA.
- IPUMS NHGIS provides geographic crosswalks from 2010 small-area units to 2022 counties, and from 2020 and 2022 small-area units to 2010 counties, which could be helpful for some types of crosswalking. NHGIS also redistributes ACS summary tables, which include various summaries of worker characteristics by county, including many counties that aren’t identified by the PWCOUNTY microdata variable.