Hi,
I’m conducting a difference-in-differences regression using repeated cross-sectional data from 7 ASEC survey years (2015-2021). I’m using Stata.
I want to know how to use svyset to incorporate survey weights into what I’m doing.
Is “svyset [iweight=asecwt]” sufficient if I don’t intend to use replicate weights? Description of asecwt: IPUMS CPS: descr: ASECWT.
Must I specify the psu and strata for this particular survey? Must I do anything to account for the fact that I’m taking data from different survey years?
Also, when I run estimates with svy: regress (etc.), I am unable to cluster my errors by state (my control and treatment groups for the DID regression are different groups of states). Is this an issue?
Thank you
The survey design variables (psu and strata) are not publicly available for CPS data. To generate standard error estimates for ASEC data, you should instead use replicate weights. You can download all 160 person-level replicate weights by including the variable REPWTP in your data extract; the household replicate weights are available from REPWT. This IPUMS CPS page on replicate weights provides sample code for using these in statistical packages.
Regarding clustered standard errors, providing analytical advice is beyond the scope of IPUMS User Support. You might be interested in this World Bank blog on clustered standard errors. In general, I recommend reviewing literature in your field and/or connecting with a survey statistician.