Food Security Supplement-sample selection impact representativeness


We are analyzing CPS FSS data between 2015 and 2020 and have only included households who were completing the survey for the first time (i.e, MISH= 1,2,3, or 4) in order to ensure that we did not have the same households participate at two time points. I believe that because households are selected to participate randomly, even with dropping those cases, each year is a cross-sectional, independent, and representative sample of the US population, but wanted to check to be sure this is correct-specifically the representative part. Any insight on this would be appreciated.

Patrick Brady

Yes, each cohort in the CPS is a representative sample of the population. So keeping only this with particular MIS values shouldn’t be a problem. There are two things you should consider, though:

  1. Panel attrition occurs when households drop out of the survey after the first month. The CPS weights are adjusted in an attempt to account for panel attrition, by increasing the weights of households that remained in the panel and re-adjusting them to match demographic patterns in the country (you can read more details about the weighting process in the CPS Design and Methodology Report). Since different months-in-sample have different attrition rates, ideally you’d want to re-adjust the weights again if you are only including MIS 1-4. However I doubt this will change your results much.

  2. There is a phenomenon called “panel conditioning,” which describes the fact that households’ responses to questions change systematically as they move through the rotation pattern (as they are interviewed multiple times). This generally isn’t a huge problem but something you should be aware of. You can read more about this at the IPUMS CPS linking workshop page (Day 2 Presentations).