Replicate weights for ACS pre-2004

USA
1

I want to compare the standard errors of my regression coefficients to those when I bootstrap them using replicate weights. My current model uses probability weight (perwt) and clusters standard errors at the state level. Since my data include the ACS 2001-2003, I would like to know how to get the replicate weights for those three waves.

Guided by an earlier topic in this forum ([I want to combine ACS data 2000-2015 but also use replicate weights, which don't exist b4 2005. What should I do?]), I have found a document on creating replicate weights. The problem is I am a Stata user, and the sample codes for SAS, SUDAAN, or WesVar do not quite sink in. I would be grateful if you could help me with this!

2

Unfortunately, there are not replicate weights published before 2005. Replicate weights are published by the Census Bureau using internal information on the sampling process that is not publicly available. Therefore IPUMS cannot calculate replicate weights for new samples. Depending on your analysis, you could run the same estimates using both weighting methods on some samples post-2005, just to check the general robustness of the variance estimation.

Regarding clustering standard errors: instead of clustering at the state level, we recommend using the Stata svy commands with the CLUSTER and STRATA variables. More information can be found here. Syntax for setting the data is as follows:

svyset cluster [pweight=perwt], strata(strata)

Then use the svy: prefix for your regression commands. For example:

svy: regress y x1 x2 x3
1 Like