If I use IPW to reweight ACS data, how do I change the replicate weights?

1

Hello! Right now I’m trying to use the three-year 2009-2011 ACS sample, and I have a question about weighting. I would like to delete all of the observations that have allocated characteristics, and then use inverse probability weighting (IPW) to reweight the remaining observations, so that they match the overall composition of the dataset in terms of race, sex, and state. I have done this before with Census 2000 long form data, and it has seemed to work well.

However, the ACS uses replicate weights. I’m not very familiar with them…

So here’s the issue. If I delete all of the observations that have imputed answers, and then try to reweight, then how do I reweight the replicate weights? Since there is a standard person-weight also given for each observation (variable “pwgtp,”) I’m tempted to just reweight the person-weight and not change the replicate weights at all. However, I have no idea whether or not this is really a valid method to use.

Can you offer any advice? Or do you know how I could find an answer to this question? I really appreciate any ideas you have!

2

The creation of the replicate weights by the Census Bureau and their use in variance estimation is described here. While replicate weights are considered the ‘gold standard’ for variance estimation in the ACS, using the person weight is a valid method and may be significantly easier in this instance.