What is repeated cross-sectional data?

[Phuc Nguyen]

From reading the guide of Stata , I saw that:

didregress can be used with repeated cross-sectional data, where we sample different units of observations at different points in time

I am wondering what does "different units of observations" mean here. From my point of view, the "repeated cross-sectional data" differ from panel data because "repeated cross-sectional data" has the data for different firms over time while the firms are kept in all year in panel data. E.g.when doing a survey for 4 years from 1990 to 1994, cross-sectional data will be the data of firms A,B,C at year 1, firms A,C,D at year2... while panel data have the data for all these firms in all years. Is it a correct thought?

Thanks in advance.

[Weiwen Ng]

I don’t know if you’re in the US. But, the Medical Expenditures Panel Survey enrolls a cohort, and it keeps them in the survey for several (I think two) years. It’s actually a bit more complicated than that, because people enroll in different quarters, but you get the picture. This is like a cohort study. I don’t know if that’s exactly equivalent to panel data, but I think it is.

I believe that our National Health Interview Survey is a repeated cross section. That is, every year, they define the sampling frame, and then they survey people. The respondents in each year aren’t the same people - actually, by coincidence, there could be some repeat interviews, but the US has 300+ million people and the NHIS is maybe 15k respondents, so this is unlikely. Now, say you want the national average of some characteristic (e.g. mental distress) measured across years. You can use the NHIS for that. If the same question were on the MEPS, you could use that as well. But the NHIS doesn’t follow the same people across years.

[Phuc Nguyen]

Thanks Weiwen Ng , it seems that you are on the same page with me. I understand your example.