Getting at Counties in PUMS

I am measuring disconnected youth (ages 16-24 neither in school nor working) in each state and working with the ACS 1-year PUMS to generate these estimates by state by year. No problem there; I can match state-level estimates reported by Measure of America But the goal is to generate county-level estimates. 

Now if I have read the documentation correctly, PUMAs are the smallest geography I can get at with the PUMS. This is probably why Measure of America relied on  "custom tabulations provided by special arrangement with the US Census Bureau" to generate their county estimates of disengaged youth.  

What would requesting custom tabulations such as this entail? I imagine there is a fee and a process to request such a tabulation. Has anyone had any experience with such requests and if yes, would be willing to share the essential steps? I suppose I am really looking for someone to tell me it can't be done without a hefty fee but do lie to me :)



  • Custom tabulations are $10,000, as far as I know. Never used these -- way too steep.

    If you have a justifiable research project and an affiliation with a university that has a research data center (, you can write a proposal and request access to the restricted ACS data, which includes accurate county IDs.
  • Hi Ani, I recently recreated the Measure of America estimate for opportunity youth in the Austin MSA using the Puma areas. You could use the Missouri State Data Center's Mabel tool to identify which PUMAs fall into each county. Seems like a large undertaking for the nation, but I think it's possible without paying for custom tables.
  • In reply to BAyres:

    Thanks Stas and Brittain; I'll look into both options. I'll start with Mabel and run a request through our RDC

  • In reply to Ani:

    Hi Ani,

    You won’t get far with the RDC request. County level tabulations are a no-no as the Census Administrator at your RDC will tell you.

    Counties with fewer than 100,000 residents are going to be combined with a neighboring county to form a PUMA in order to get to the 100,000 threshold. Can you live with PUMA’s that combine two or more smaller (less than 100,000 residents) counties?


    Warren A. Brown, PhD
    Senior Research Associate, Cornell Institute for Social and Economic Research
    Director, Cornell Program on Applied Demographics
    Research Director, Cornell Federal Statistical Research Data Center
  • In reply to Warren Brown:

    Thanks Warren, that is useful to know! I imagined it would be difficult to get at county estimates and that expectation is borne out by both your and Stas' feedback. I I have to, then I'd rather stay with PUMAs and skip the county-level estimates altogether.

    Since Brittain's pointer re: Mabel I have pulled the PUMA to county crosswalk and am going to tinker with it in the next few days to try and generate pseudo-county estimates. There are a few kinks I know I have to work out with the weighting once the crosswalk is used. I'll wrestle with it and then likely have some questions for the group.

    Once more, thanks everyone for your invaluable feedback.

  • In reply to Ani:

    How did it go? Hoping to do the same but for just Hispanics at the Metropolitan Statistical Area level for my dissertation. Would appreciate any tips you may have. This is my first time using the ACS.
  • In reply to Michelle Hawks Cuellar:

    Hi Michelle;

    It worked out well, and I can share my code with you but I used R not Stata. Sounds like your work will mirror Brittain's (she was focusing on the Austin MSA).
  • I'd recommend using 5 YR ACS data if looking at this small of a sample. Also, you might want to toi try and combine PUMAs into MSAs instead of counties. Sometimes the geography works out better.
  • In reply to Michelle Hawks Cuellar:

    it looks like the ACS data should have the metro area designations off the bat (

    If that does not work, the FAQ reference for nearly every geographic cross-walk imaginable is Missouri Census Data Center geocorr tool: The "from" geography will be PUMA, the "to" geography is CBSA... as far as I understand that tool... which I don't, so I always request both allocation factors, and then just see which one of the is 1, or sums up to 1 in the expected way. Then you can `joinby puma using geocorr_crosswalk, assert(match)`out of the ACS data.
  • In reply to Douglas White:

    I agree - we create our own 3 year estimates to look at Teen Idleness and Disconnected Youth in NYC and still find that some of the PUMA Ns are quite low.