CPS NBER Hours discontinuity 2017

Hi all,

I am working with the IPUMS CPS Monthly files as well as the NBER CPS Monthly files from 2013 to 2017. In the NBER CPS Monthly data, I noticed there is a sharp discontinuity in the number of hours worked starting in 2017 (total hours worked jumps by several orders of magnitude). This seems to be true when using the NBER weights pwsswgt, pwcmpwgt, and pworwgt (I did remember to divide by 10,000 for the implied decimal places). In the IPUMS CPS monthly data, I do not see the same discontinuity in hours starting in 2017 using the wtfnl weight.

I suspect I am forgetting about some data quirk. Can anyone explain to me why there is a discontinuity in hours for the NBER CPS Monthly but not in the IPUMS CPS Monthly? It would also be helpful to know where the IPUMS CPS variable WTFNL comes from (i.e. does it correspond to a specific weight in the NBER Microdata).

The variable in the NBER files corresponding to the IPUMS variable WTFINL is PWSSWGT. I don’t know exactly what’s causing the discontinuity in the NBER files, but I have a couple suggestions for investigating:

  1. Sum the weights for each sample. If there is no discontinuity in the sum of weights, then the issue is with the hours worked variable.

  2. Check the hours worked variable to see if there was a change in NIU or other codes. For example, the NIU code may have changed from 999 to 9999 which would make a huge difference in the average, if you didn’t account for this. I don’t see any obvious code changes for the hours worked variables in 2017, but I may have missed something.

If these checks don’t reveal anything, you may want to reach out to the CPS data team at the Census Bureau at dsd.cps@census.gov. If you find something that leads you to believe there is an issue with the IPUMS data, please reply to this post or send an email to ipums@umn.edu.

1 Like

Thanks for the reply! It definitely seems to be an issue with the weights themselves. When I look at the unweighted sum of hours, there is not a discontinuity. I also took your advice and summed the weights, and the sum of the weight totals jumps sharply in 2017. A good thought on your second suggestion, but I already drop missing/ outlier hours.

I will contact the CPS team to see if they can shed some light on the issue. I will post back here with results assuming I hear back from them.

Thanks again!

Alright, it’s been a minute, but it seems we have a resolution here.

I can confirm that there was nothing amiss with the IPUMS cps monthly individual weight. After talking with NBER and Census, it seems like the issue is a result of a slight inconsistency in the treatment of implied decimal places in the NBER monthly data. Prior to 2017, the weights in the NBER data have already been adjusted to account for the implied decimals (i.e. already divided by 10000). From 2017 onwards, the weights have not been adjusted to account for the implied decimal places in the NBER monthly data.

You can prove this to yourself (like I did) by merging the IPUMS cps monthly data with the NBER cps monthly data and comparing wtfinl from IPUMS with pwsswgt across different years. This pattern seems to be present for all weights in the NBER cps monthly data.

The data team at NBER said they would work on fixing this, but that it could take a bit. Short term solution is to make the adjustment yourself. Long term, be mindful that a fix could be implemented in the future, which means checking for this slight inconsistency when working with the NBER data.

Glad to provide further details for interested parties.