2023 5yr PUMS Data by County

Social Explorer has those data up, as does the NHGIS.  What tables do you want (by number) and what are you using to get the data.  There are many cases where PUMAs span more than one county, and where counties have more than one PUMA.

Social Explorer and NHGIS have poverty ratio by race? Good to know, since I haven't been able to find that anywhere else.

Glenn is correct. NHGIS and Social Explorer redistribute the published ACS Detailed Tables (i.e., Summary File) data. I don't know of published ACS tables that cross-tabulate poverty ratio by race. In any case, NHGIS doesn't have any. I'd guess that Social Explorer doesn't either.

So you could estimate county-level characteristics from PUMAs, as Glenn indicates in his other post here: "If a small county's population comprises (e.g.) 30% of the containing PUMA's population, multiply the PUMA's characteristic of interest by 0.3 to get a county estimate." However, this approach makes the assumption that there is no variation in poverty/race correspondences among counties within a PUMA. A more sophisticated approach would take account of county-level race data and poverty data (for which the ACS does provide data, just not combined in one cross-tabulation), possibly using Iterative Proportional Fitting (IPF) to make county-based estimates in a way that's constrained by both the county-level info and PUMA-level info. I don't know offhand of a good general guide for this specific application, but I think there are multiple online resources covering IPF.

Glenn is correct. NHGIS and Social Explorer redistribute the published ACS Detailed Tables (i.e., Summary File) data. I don't know of published ACS tables that cross-tabulate poverty ratio by race. In any case, NHGIS doesn't have any. I'd guess that Social Explorer doesn't either.

So you could estimate county-level characteristics from PUMAs, as Glenn indicates in his other post here: "If a small county's population comprises (e.g.) 30% of the containing PUMA's population, multiply the PUMA's characteristic of interest by 0.3 to get a county estimate." However, this approach makes the assumption that there is no variation in poverty/race correspondences among counties within a PUMA. A more sophisticated approach would take account of county-level race data and poverty data (for which the ACS does provide data, just not combined in one cross-tabulation), possibly using Iterative Proportional Fitting (IPF) to make county-based estimates in a way that's constrained by both the county-level info and PUMA-level info. I don't know offhand of a good general guide for this specific application, but I think there are multiple online resources covering IPF.

Jonathan Schroeder said:

this approach makes the assumption that there is no variation in poverty/race correspondences among counties within a PUMA.

Right, this is a very simple method for estimation. I should have included that caveat.

In an ideal world, PUMAs would be defined to have broadly evenly distributed characteristics. Also, all states and counties would be rectangular, and no city would cross county lines.