Hi. I’m analyzing the CPS November supplement to see how the percentage of people reporting that they registered at a DMV changes over time nationally and some analysis state-by-state.
Which of these should I use when running regressions? They give slightly different answers for the tests of significance which may be important down the road, for reporting purposes if nothing else.
- svyset serial [pw=vosuppwt], strata(statefip)
- svyset serial [pw=vosuppwt]
- or just ignore svy altogether and type [pw=vosuppwt] in the regression command, e.g. logit dmv i.year [pw=vosuppwt]
The CPS microdata doesn’t include any sample design variables (strata or PSU clusters). Because of this, svyset is not necessary - you can just use the weights in each estimation command (so that’s your #3). This is not ideal because it is likely to somewhat underestimate standard errors, but unfortunately it’s the best we can do for basic monthly CPS samples.
Since you are using the Voter Supplement, you could use replicate weights to get more accurate standard errors. However these are not yet incorporated into IPUMS CPS. You can find the replicate weights for supplements at the Census Bureau website, and it is possible to join them to your IPUMS file, but it is a complicated process (an overview can be found here). Syntax for using replicate weights in Stata and SAS can be found here: IPUMS CPS. Suggested R syntax can be found here: How can I use replicate weights to create standard errors in R? - #10 by Matthew_Bombyk