Time-Invariant variables?


I was wondering whether some of non-identifier variables could be time-invariant in IPUMS USA?
Checking for this condition numerically is intensive, avoiding this computation would be really useful.

Can you clarify your purpose for needing time-invariant variables? The samples in IPUMS USA are cross sections, so there are not multiple observations per household or individual (with the exception of the multigenerational longitudinal panel). In general, birthplace (BPL), RACE and SEX are likely to be time invariant for an individual, among other variables.

Thank you for your reply. I wanted to know whether EDUC is time-invariant.
I recently read a working paper where EDUC did not have any time subscript. I thought it was a mistake and I just wanted to make sure.

As I mentioned, for IPUMS USA there are not multiple observations per person. They only have one record, so can only have one value of EDUC (and every other variable). If you are using the IPUMS CPS data, respondents are interviewed multiple times. In those cases, EDUC can be different for an individual’s different interviews.

Thank you, I misleadingly thought that unique identifiers created from serial, person and sample could represent representative individuals across years. I add this related link for anyone falling in the same trap.

1 Like