Replicate weights reducing CIs

USA
1

Hi I have been analyzing the IPUMS USA data using replicate weights supplied.

In most cases the replicate weights seem to increase the SE and hence margin of error in comparison to American Factfinder – this is what I expect.

However in a few cases, the replicate weight variance actually seems lower than suggested by published margin of error.

Does this seem plausible or an indicator of a potential problem in the analysis/code?

2

Although it is typically the case that using replicate weights will increase the standard error around coefficient estimates, it is possible for the standard errors to decrease. Therefore, your observation seems plausible to me. More information about the use of replicates weights can be found on this page.