I have been trying to estimate active duty and veteran populations in a specific MSA with a large military base with 1-year microdata. Using perwt I’ve been trying to measure increases and decreases in these two populations for the MSA with respect to the census population estimates. For example, if there is large net out migration in the MSA according to the PEP estimates, I would check if there was a decrease in populations related to the military. While doing this I realized that Military Community Demographics Reports • Military OneSource exists, but when I check the directional change of the active duty members, the change in the microdata does not mach with the change between years in these reports from the military.
I’ve been using vetstatd == 12 for active duty military and then using perwt to get the population. Is there something I’m not understanding about the microdata when comparing it with these military reports? I’m not so much concerned about the numbers being exact, I just want the directional change to be similar so I can be confident when I’m looking at other demographic statistics.
One particular challenge with using public use microdata for this sort of analysis is that the lowest level of geography that is consistently identifiable in these data is the PUMA (public use microdata area). These PUMA boundaries do not necessarily overlap with official MSA boundaries. Therefore, not all MSAs are identifiable in public use microdata. Additionally, the MSAs that are identified in IPUMS USA may not be perfectly identified. As is noted on the description tab of the MET2013 variable, there are errors of omission and commission associated with each MSA. Therefore, this may be one reason why some estimates using the microdata do not align with existing estimates using a different data source.
Beyond what @JeffBloem pointed out, you need to understand that the ACS weights are calibrated to match the population projections at pretty high levels (states, large counties, etc.). These projections are typically empirical in nature, and usually miss important local swings (the fracking booms and busts; large scale local construction, e.g., dams or plants, etc.). For these reasons, the Census Bureau explicitly advises against relying too much on the ACS totals, so @David_Kelley you are in a danger zone. I would expect that troop deployment would fall between the cracks – let’s just say that by nature of their work, DoD is unlikely to provide their counts to the Census Bureau.