Applying sampling weights manually to observations

USA
1

Hi,

I am facing a question regarding sampling weight in the IPUMS USA data. I know that I can apply sampling weights in regressions using built in commands in various softwares. I wanted to ask if there is a possibility to manually apply sampling weights to each observation and create a weighted sample, which would then not require to use the perwt variable. For instance, multiplying the variable by sampling weight, or dividing it by sampling weight or any other such operation. I would be thankful for a response.

2

The short answer is no, but there are some cases where you can do this. You can multiply your variable of interest by the weights (normalized to sum to 1), which allows you to calculate a weighted average without weighting. This may also work with certain other calculations, such as weighted least squares. But it will not work for everything. For example, this will not allow you to calculate a weighted median. Also, any standard errors you calculate using these data will be incorrect.