I have been working with ipums usa data for awhile now and have read through all the sample selection information on the website, but I am still a bit unsure on which samples are best for my research objectives. I am looking at 4 years of data 1990, 2000. 2010, and 2015. I am looking at a smaller group of the population- household heads who do not own cars (vehicles available variable) and I want to compare their characteristics over time at the puma level. Specifically, I am unsure about which samples to use for 1990 and 2000. I originally thought I should use the 1% metro sample as my study area is 20 urban areas, then I wondered about using the 1% unweighted because it says it represents the general population better. But then the other argument is to use 5% samples when looking at smaller subgroups to have enough samples. I wondered how to choose the sample, and how to ensure the sample I choose is mostly consistent over time. Also for 2010 I am unsure about if to use the ACS or the 10% sample. I realize the sample choice would be different for different research, but I wondered if anyone could offer some advice. Thanks!
Generally speaking, it is better to use samples that implement similar sampling frameworks. However, since you are interested in performing your analysis at the PUMA level your choices are limited by the fact that the 1% files from 1990 and 2000 do not identify geographic detail at the PUMA level. (Instead, they use Super PUMAs.) So, you will need to use the 5% files from 1990 and 2000. In 2010 you should use the ACS files, because the 2010 10% file only consists of the “short form” census and has a limited set of variables. Finally, in 2015 you should again use the ACS.