Could I construct panel data using multi-year ACS data

Hi, there,

As far as I know, ACS surveyed different people in different years, meaning that it is not a panel design. However, in the ACS data I downloaded from IPUMS, it has variables like household serial number and personal serial number, which can be combined to uniquely identify individuals (I suppose). Can I use this approach to link individuals in multi-year ACS data to construct panel data? Or, simply put, is it legit or possible to contract individual-level panel data using multi-year ACS data? Thanks.

Household serial number and person number can be combined to uniquely identify individuals only within a single sample. Household serial numbers are reused across samples and similar numbers do not indicate that this is the same household. Unfortunately, it is not possible to construct individual-level panel data using the ACS. I recommend taking a look at the CPS (Current Population Survey) which contains many of the same variables as the ACS and is specifically constructed as a rotating panel of households.

OK. So basically there is no way to identify individuals across ACS years even though some people may have been surveyed in multiple ACS years, because they (and their households) were given different identification numbers in different ACS years. And the legitimate way to use multi-year ACS data is to treat them as cross-sectional data?

Yes, that is correct. SERIAL is a unique variable created by IPUMS that does not allow for identifying individuals across sample years. You may be able to find more information on how the original Census serial numbers are constructed on the Guidance for Data Users Page. Moreover, the sampling procedure ensures that no households are included in the ACS for multiple years in a row. Section 4.3 of the ACS Design and Methodology Report notes that the sub-frame used to select addresses into the sample is rotated each year. Since there are five sub-frames in total, the same household comes up for selection into the sample only once every five years.