Okay to do post-stratification of IPUMS using NCHS bridged-race population estimates?
If your goal is to “correct” for multiple-race persons in the data, then I would first suggest that you consider using the RACESING variable already available in the IPUMS data. This variable bridges multiple-race responses to single-race responses. The bridging in IPUMS is done using the same method as the NCHS. These estimates should match the bridged-race population estimates very closely. It is up to the researcher to determine if any remaining discrepancy between data sources requires a further adjustment such as post-stratification.
If you would still like to use the NCHS estimates, I would recommend first familiarizing yourself with the research design used for both IPUMS data and the bridged-race population estimates. If in your judgment the modified race estimates found in the NCHS data are more truly representative of the population necessary for your purposes, then there should be no issue with using a post-stratification method on the IPUMS data.
Hope this helps.
thank you for responding. I did not think of using post-stratification weighting to correct for race. My assumption was that NCHS a was the “correct” population estimate; therefore, I should post-stratify weight all estimates such as poverty, linguistic isolation, etc. Regardless of race. Is this ok?
If I understand correctly, you want to apply the rates that you calculated from the IPUMS data to the NCHS population estimates, with the assumption that the NCHS total population estimates are “correct” for a given geography. There is no issue with doing this.
However, keep in mind that the IPUMS data and the NCHS data both originate from the Census Bureau, so the population estimates are unlikely to differ substantially. For example, I took the latest NCHS SAS file and compared the State population estimates for 2010-2012 with IPUMS-USA. Most states differed by 0.1% or less in a given year, with a maximum difference of 0.5%. Similarly, the total U.S. populations differed by 0.01% or less. In other words, we are talking about very small differences.
I hope this helps.