Matching PUMAs 1990-2020

Hello IPUMs,

Im working on a project and need to calculate some migration statistics at the smallest level possible (so PUMAs looks like) for Atlanta GA. Because PUMA boundaries change often Im wondering if I can navigate this by constructing my own crosswalk? I found a 2000-2010 crosswalk on the website and Im wondering if by using GIS and census tract estimates(from the decennial cesus and ACS data) I can approximate a consistent PUMA area across 1990-2020 or at least 2000-2020? Im curious about the implications of this approach and the potential loss of percision.

Any thoughts, comments, helpful critique is appreciated.

Thank you,

Note 1: The 2020 PUMAs are not yet in use in any data product. We expect they’ll first be used in the 2022 ACS PUMS released next fall. All ACS microdata for 2012 through 2021 use the 2010 PUMAs.

Note 2: Information about prior residence in public use microdata does not use PUMA-level detail but rather “migration PUMAs,” which are aggregations of one or more PUMAs. IPUMS USA provides severable variables identifying migration PUMAs and other areas derived from them, and you’ll find some additional resources (maps, composition files, etc.) for migration PUMAs here.

Note 3: For general guidance on how we’ve constructed consistent PUMAs and how you might go about constructing your own, see this previous forum post. See also this post about relationships between migration PUMAs and our consistent PUMAs.

Lastly, I’d say that the approach you suggest could work well, but much depends on how well the areas you’re interested in match up across time, and whether consistently aligned areas would still be small enough to be useful for your analysis. In short, it seems worth a shot, but it’s hard to know till you’ve done some more digging!

1 Like


Thank you for the quick response! I really apprciate the feedback.

Regarding Note 2: I dont need to know necessarily where they came from, only when they moved in. I am using the “MOVEDIN” variable. Does this still apply? Is the respondent answer referring to the migration PUMA that live in currently and not the actual PUMA recorded?

That’s helpful! Migration PUMAs are used only for prior residence. PUMAs identify current residence, so yes, you could identify the current PUMA of residence for everyone in the MOVEDIN universe (i.e., all householders).

1 Like

Jonathan - thanks so much again.

Last question(s) for a while I promise -
Using the ACS 2020 5-yr sample, are the variables adjusted to 2020 if the respondent was surveyed at the beginning of the period? For example, if a respondent answered that they moved in to their current home 1 year ago and took the survey (ACS 2020 5-yr) in 2016 is the variable adujusted to say 4 years in the 2020 data extract?

The reason I ask is I did some quick summaries and I am finding (for the PUMAs I’m interested in -GA 04600, 02002, 01006, 01007) prety low move in rates in this 10 year period (2010-2020). I summed the variable “PERWT” for responses that reported they moved in within the last 10 years and is between 20-30% which is very low compared what other people have reported.

Any suggestions would be great, thanks.

There’s no adjustment to standardize the reference period. As explained in the MIGRATE1 description:

MIGRATE1 reports whether the person had changed residence since a reference point 1 year ago. Specifically, individuals age 1+ were asked if they had lived in the “same house” (non-movers) or a “different house” (movers) one year earlier.

The “reference point” will be 1 year before the respondent’s survey time, whenever that was, for all respondents in the sample, so a 5-year estimate of 1-year movers is like an averaging of all 1-year periods preceding each response.

I’m not sure what’s causing you to find unusually low move rates. It may have something to do with how you’re defining “moved within the last 10 years”, which is not something the ACS measures directly. MIGRATE1 reports only whether an individual lived in the same or different house 1 year before.

Ah, sorry, I forgot you were using the MOVEDIN variable! That variable pertains only to householders, not all persons. That could explain the low rates of movers you’re finding.

1 Like

Ahh I see, so what variable should I use to get totals? “HHWT”?

Yes, using HHWT would get you the proper estimate of total householders by MOVEDIN code. Because the question is only asked of the householder, there’s no way to directly estimate total population by MOVEDIN. MIGRATE1 is more general… for all population age 1 year and over.

1 Like