Question about replicate weights and inference

Hi,

Hopefully quick question. If I want to do inference, such as changes in income between 2018 and 2019, if I want to account for the fact that households are repeatedly sampled in a 4-8-4 month pattern, all I need to do is use the replicate weights as shown here (IPUMS CPS)? Anything else I need to be aware of?

-Dan

Sorry, to clarify, I want to work with the Basic Monthly file. This file doesn’t appear to have replicate weights. So do I just cluster on “WTFINL?”

-Dan

You’re right that the basic monthly samples do not include replicate weights. For most analyses with the basic monthly data, just weight your analysis using WTFINL. There is no cluster or strata information in the data. Keep in mind that since this is a complex survey design with clustering and stratification, but information about the design is not available in the public use microdata, the standard errors will be estimated with bias (with uncertain direction). IPUMS testing has found that using replicate weights with ASEC data usually increase estimated standard errors slightly. So your standard errors are likely to be slightly underestimated.

If you are using the CPS data longitudinally, there are a separate set of weights created by IPUMS that can be used in your analysis. More information on those is here: https://cps.ipums.org/cps/cps_linking_documentation.shtml#linked_weights

Thank you. Can you clarify what you mean by using the CPS data “longitudinally.” Let’s say I’m tracking a variable across years or across months within a year? Does that qualify as “longitudinally.”

-Dan

Longitudinal analysis follows individuals through time. If you are just calculating an aggregate statistic, like the unemployment rate, over multiple months, then it is not longitudinal analysis.

Thank you!