Are DATANUM, SERIAL, and PERNUM the same in single-year and 5-year ACS samples?

I previously downloaded the 2009 ACS 5-year sample, which did not include data on WKSWORK1; but WKSWORK1 is available for 2005-07 in the single-year ACS samples for 2005-07. Question: are the variables that uniquely identify cases (DATANUM, SERIAL, and PERNUM) the same in the 2009 ACS 5-year sample as in the single-year ACS samples for 2005, 2006, 2007? Can I accurately merge data from the single-year samples into the 5-year sample? (If so, I would link MULTYEAR in the 5-year sample with YEAR in the single-year samples.)

While you are correct that DATANUM, SERIAL, and PERNUM will uniquely identify an IPUMS record, the the Census Bureau adjusts the weight values when creating the multi-year samples. Therefore, merging data from a single-year file to a multi-year file would not give you accurate estimates. You could conduct your analysis as you desire by building your own 5 year file from the appropriate single year files.

I have a related question: Currently, there are some variables that are available in the single-year samples for 2013-2017 but not (at least yet) available in the 2017 5-year sample (e.g. CINETHH).

We generally use the 5-year samples because of the large N and because they have the official Census Bureau adjusted weights for the five-year pool.

It seems to me from the response above that it would be possible to merge just the CINETHH variable from the five one-year files (stacked up) to the five-year file by YEAR/MULTYEAR), SERIAL, and PERNUM. Is this correct? Or are the YEAR/MULTYEAR, SERIAL, and PERNUM variables only meaningful within a particular sample?

Seems to me that if we were able to merge in variables such as CINETHH that are not yet available in the five-year sample from the five one-year samples, we could still apply the Census Bureau adjusted weights for the five-year pool. Does that sound correct? The response above makes it sound like what I am suggesting here would not give accurate estimates, but seems like if they are in fact the same respondents, and the values of CINETHH are the same for individuals in the five-year and one-year samples, then applying the Census Bureau adjusted weights from the five-year sample would give the same estimates as if we waited for the CINETHH to be added to the five-year sample.

Which leads to another question: Will the CINETHH variable be added to the 2017 five-year sample eventually?


You are correct that the variable CINETHH (along with other household appliance variables) is currently available in the ACS 1-year samples from 2013 through 2017, but not currently available in the 2013-2017 ACS 5-year sample. With that said, the CINETHH variable will soon be made available in the 2017 5-year ACS sample by the IPUMS USA Team.

In the meantime, unfortunately, it is not possible to merge the 1-year ACS file with corresponding observations in the 5-year file. This is because SERIAL and PERNUM values are unique within IPUMS USA samples. This means that although the same individual exists in, for example, the 2017 ACS 1-year file and the 2017 ACS 5-year file this individual will have different values in SERIAL and PERNUM between these two files. The best way forward, besides waiting for these variable to be released in the 2017 ACS 5-year file, is to combine the five 1-year files from 2013 through 2017. As is noted above, you’ll need to adjust the sampling weights. A good approximation is to simply divide the sampling weight by 5 (or the number of ACS 1-year files included in the pooled sample).