IPUMS doesn’t have any specific recommendations on how to adjust your standard errors in pooled analysis of the CPS. One option is to only use households in a single rotation group, as you mentioned. Unfortunately that reduces your sample by almost 90%. I recommend reading some papers in your field using CPS data and see how they have approached this problem.
A couple other things to note on this:
-
There is another issue with standard error calculations with basic monthly CPS data, which is that the design variables (clusters and strata) are not available in the public use microdata. You can get fairly close to the actual standard errors by using the smallest geographic level available as strata and SERIAL as the cluster in your estimation. See Davern et al. (2006, 2007) for details on this.
-
You might also consider clustering your standard errors at the level of household, using CPSID as the identifier, since this will treat all observations of a given household as having correlated errors. You may also consider using household fixed effects. The particular model you use depends on the assumptions you’re willing to make about the data. Some good sources on adjusting standard errors for clustering include Wooldridge’s Econometrics of Cross Section and Panel Data, and most other graduate econometrics textbooks. Good sources for variance estimation in complex surveys include Cochran (1977), Sampling Techniques, and Kish (1995), Survey Sampling.