Mechanics of using the adjacent year longitudinal weights

CPS
1

I’m working on a project to examine employment status and earnings trends across 2 consecutive years in the CPS ASEC. It seems like I should be using LNKFW1YWT for my analysis to weigh my calculations in terms of the first year. I’ve looked at several IPUMS resources on this but am having trouble understanding how this works mechanically once I match people across the two years. Do I simply merge the datasets and then use the LNKFW1YWT attributed to the first year’s data entries in my analyses? Am I correct that my tabulations weighted this way would give me the number of people fitting certain conditions out of all those who were eligible to match, but that calculations surrounding average earnings or something like that should be applicable to the population at large (or my universe of analysis) in year 1?

Thank you for any help you could provide!

Edit: Should I just be using ASECWT in the same way?

2

Merging two datasets based on CPSIDP and using LNKFW1YWT from the first year’s record is a valid approach to the analysis. The longitudinal weights adjust for the fact that those who actually do link are not necessarily perfectly representative of those who are eligible. This page gives more detail on how the weights were constructed.

The program used by IPUMS to generate these weights creates a set of population counts for the individuals in the first year who are eligible to link one year later (this is anyone in rotation groups 1-4). It then adjusts the weights for those who actually have a linked record a year later, so that the weights in the linked sample sum to the population count of those who were eligible to link. This is done within each of the subgroups noted on the page I linked above.

Any statistics calculated on this subsample will be approximately representative of the population in the first year. But note that the weights will sum to the total weight of people in rotation groups 1-4 (approx. 160 million in recent years), not the overall population (approx. 320 million). I think this is the same as what you wrote regarding tabulations (which sum weights) vs calculations (which use normalized weights).

Regarding ASECWT, this is a cross-sectional weight. LNKFW1YWT is also available for ASEC samples, in which case it is adjusted as I described above, but in this case it is based on ASECWT instead of WTFINL.