Hello,
I am performing a state-level time series analysis of populations by education, race, and employment status using the WTFINL weights and basic monthly data (non-ASEC). I am pooling three years together to get a larger sample and presumably, a more accurate one. However, one area that I’m somewhat lost in is the use of MISH. There doesn’t seem to be clear guidance on whether or not to use only the first interview or all the interviews - even if it includes double counting. Is there any guidance available on that front?
Thanks
-Peter
Thanks for your post. The CPS is often treated as a repeated cross-section without any special adjustments for the panel component of the survey, though I recommend consulting any papers in your specific field using CPS to see how others have approached this problem. IPUMS does not have any specific recommendations on how to address repeated observations in pooled analyses of the CPS. A former colleague of mine shares a few ideas on this topic in a separate post; I will recap them here as well.
- You could keep only a single rotation group per sample to avoid repeat observations (as you note), but this reduces your sample size considerably
- Davern et al. (2006, 2007) have suggestions on estimating standard errors with the public use data in the absence of sample design variables, including using the household SERIAL number as the cluster in standard error estimation. Using CPSID, the household-linking key from IPUMS CPS, instead of SERIAL would treat observations of a given household as having correlated errors.