Weighted average of CPS EARNWEEK for a subset of states and months

Hi everyone,

I am trying to plot the distribution of EARNWEEK for a subset of states and pooling together several months of CPS Earner data, but I am not sure if I am handling EARNWT correctly. Specifically, I would like to get a connected scatterplot where I have (binned) EARNWEEK on the x axis and a weighted count of individuals on the y axis. I am working with 3 consecutive months of CPS Earners data and a subset of states. Here is what I do:

  • I divide observations into bins by EARNWEEK values
  • I construct a weighted average count of how many individuals are in a given EARNWEEK bin (I use Stata collapse to build a weighted count for every bin).

Now, I would like to obtain a weighted average count of how many individuals are in each bin, across the 3 months and states I am interested in. I use EARNWT / 3 (number of months I am interested in) to build the weighted average count.

Is that correct, given that CPS is representative at the state / US level but I am working with a subset of states?

On a separate note, if I were to replicate this analysis using EARNWEEK as reported in the ATUS survey, I believe I should still use EARNWT as weight. However, because of how the ATUS sample is constructed, I guess I should perform some sort of adjustment on it to account for the different ATUS sample. How should I adjust EARNWT in this case? Would EARWT * BWT (ATUS person base weight) be appropriate?

Thank you so much!

Yes, that is the correct method for adjusting the weights for pooling multiple months. The CPS sample is stratified by state, and the weights are adjusted after data collection to add up to each state’s population, so there should be no problem with your method.

Would you mind posting your second question as a separate thread in the Time Use forum?