As all observations in the 2018 5y ACS use the 2010 PUMA classification, is it still necessary to use the crosswalk between PUMA and PUMAMIG1?
The need to use the 2010 PUMA to the 2010 MIGPUMA1 crosswalk table will ultimately depend on what you are hoping to accomplish in your analysis. Note that even though the underlying PUMA classification does not change across all samples in the 2018 5-year ACS file, the relationship between PUMA and MIGPUMA1 boundaries is not necessarily one to one. Specifically, as this page notes, each MIGPUMA1 boundary corresponds to one or more PUMA boundaries. This aggregation of MIGPUMA1 boundaries exists to preserve respondent confidentiality in the ACS public use microdata.
Hi Jeff, thanks for the quick reply. The ultimate goal is to analyze migration flows on a small scale level. Hence, I will aggregate PUMA to the MIGPUMA1 boundaries using the crosswalk to compute the flows. That should work, right?
On a related note: The only way to compute migration flows that are fully comparable between the 2018 5-year ACS and the 2013 5-year ACS is to fall back to the county level, right? In principle, the consistent CPUMA0010 classification is available, but a) the boundaries are not 100% comparable and b) there is no CPUMA0010 one year ago question available. Hence, falling back to MIGCOUNTY1 and COUNTYFIP to compute migration flows is the only consistent way to compare migration in the two samples, right? Which, of course, comes at the price of a number of observations dropping out because the corresponding PUMAs cross county borders, eliminating the corresponding migration information.
To compute flows from a given MIGPUMA to a given PUMA, you wouldn’t need to do any aggregation. MIGPUMA1 will give you the previous MIGPUMA of residence, and PUMA will give you the current PUMA of residence. If you want to determine flows from MIGPUMAs to MIGPUMAs, though, then yes, you’ll have to use the crosswalk to determine the MIGPUMA of current residence based on the PUMA info.
PUMA and MIGPUMA definitions changed between the 2011 and 2012 ACS samples, so it’s true the that the 2013 5-year sample doesn’t use a consistent MIGPUMA definition that’s comparable to the 2018 5-year. One option would be to shorten your time frame and compare shorter periods by pooling 1-year samples, say, comparing the 2012-2014 1-year samples with the 2016-2018 1-year samples. (In general, you can pool 1-year samples simply by dividing weights by the number of years you pool.) Otherwise, yes, I think MIGCOUNTY1 may be the best option for longer periods, granting the limitations that you noted.
One more option you could consider: On the 2010 Migration PUMAs definitions page, IPUMS USA provides a crosswalk of 2000 to 2010 Migration PUMAs. It’s technically possible to use this information to construct “ConsMIGPUMAs,” but that’s a somewhat involved process, and may result in some very large consistent aggregation of units, so I’m not sure it’s worthwhile. But if you’re interested, here’s our PAA Extended Abstract detailing how we generated the 2000-2010 ConsPUMAs.
Thanks so much, extremely helpful!