CPS tobacco supplement drops all august observations when tsmker == 9 (indeterminate) is dropped - why?

I am using CPS Tobacco supplement responses from 2006-2007 and 2010-2011. First I ran this command to drop any observations that were not in merged outgoing rotation groups (MORG) becasue they do not get surveyed with the tobacco supplement - everything is normal up to here:

drop if hourwage == 99.99 | age<=15 | age>=65 | (empstat !=10 & empstat != 12) | classwkr<20 | classwkr >28 | (mish !=4 & mish !=8)

Then when I go to drop all observations for which the answer to tsmker is “indeterminate” (aka all those for which tsmker == 9) something strange happens. What happens is that all the August observations in both time periods are dropped. I don’t see why all the August observations would be indeterminate. Please let me know if you have ideas.

This is a tricky question and is due to some interesting quirks to the survey design of CPS supplements.

The first detail to note is that the code for “Indeterminate” in TSMKER is ==00, not ==09. The 09 code identifies observations not in universe. This will become an important detail.

Next, when you drop all cases that have HOURWAGE==99.99 (that is, “not in universe” for HOURWAGE), you are dropping everyone except those who are in the outgoing rotation groups. What this means is that your sample now consists of only individuals who are in their 4th or 8th month in the CPS sample. Meanwhile the universe statement for the TSMKER variable states that only those in their 5th through 8th month in the CPS sample are eligible. So, only those in their 8th month in the CPS sample will have non-NIU values for TSMKER. Additionally, note that before dropping observations with TSMKER==09, all individuals surveyed in August have TSMKER==09. This indicates that there are no individuals who are in their 8th month in the CPS sample in the month of August and are participating in the Tobacco Use Supplement.

This observation may seem strange, but is entirely due to the design of the Tobacco Use Supplement (TUS). The supplement is implemented in three month waves that can be pooled together to increase the sample size and precision on state-level calculations. In order to avoid counting households twice when pooling the waves together, households already included in the given TUS wave are coded as NIU in any subsequent month. Specifically, this means that if a household is in their 8th month in the CPS sample in August of a given year, they must have also answered questions in May of that same year. As it turns out, the TUS waves that include August also include May of the same year. So any household in their 8th month in the CPS sample in August will be coded as NIU. Therefore, when you drop all cases that are TSMKER==09 every observation from August is dropped.

Thank you for the reply! That makes sense.