Discrepancies between number of obs in 2018 ASEC and March BMS and problem with matches in 2021

I extracted CPS ASEC data from 2018 to 2021 and CPS Feb-March-April BMS from 2018 to 2022 in separate stata files.

I am finding that:

i- after excluding the ASEC oversample (drop if asecoverp ==1), the ASEC extract in 2018 has 114,291 obs but the March CPS has 122,956 obs. Any reason why the samples are different? Neither seems to have duplicates (duplicates report year month hrhhid hrhhid2 lineno)

ii- the match between ASEC and March CPS in 2021 works using year month cpsidp as linking keys but does not work using year month marbasecidp. For the 2018 to 2020 cpsidp or marbasecidp generate the same number of successful matches. Did marbasecidp change in 2021?

Thank you!

Let me answer your questions in order.

  1. The IPUMS CPS team is reasonably confident that the cause of this discrepancy is the 2018 split panel survey. In 2018, about 5,000 households in the ASEC sample were randomly assigned to receive an alternative questionnaire which replaced the health insurance coverage questions with the ones that were used prior to the 2014 CPS redesign. According to the accuracy statement, about 4,200 housing units were determined to be eligible for interview and about 3,500 interviews were obtained. This lines up with the 8,665 observations from the 3,494 households that constitute the gap in sample sizes. Since these households received the same March Basic questionnaire, but a different ASEC questionnaire, they are not included in the ASEC files harmonized by IPUMS. The team is considering adding these files at some point in the future.

  2. MARBASICIDP is currently not available for March 2021. The CPS team is currently working on making it available to users. As an alternative, you can use CPSIDP to link individuals between the March Basic and ASEC surveys.

If you’re interested in getting data on the 2018 split panel households, you can access the microdata files directly through the Census Bureau website.