In modern census samples the lowest geographic area identifiable in the public use files is the PUMA. Therefore, Cities are not directly identifiable but are able to be identified if the city boundaries are perfectly coterminous with the PUMA boundary. In other instances there is a protocol for identifying cities and this process is the cause of the errors recorded in CITYERR.
As the comparability tab of the CITY variable states: The CITY variable is identifiable in 1990 and later samples if the majority of each PUMA’s population resides within the city boundaries. Where a city is identified, it indicates that, for the PUMA in which the household resided, a majority of the PUMA’s population resided in the identified city, according to population counts from the reference census for the PUMA definitions (e.g., 2000 populations for 2000 PUMAs). A household might not in fact have resided in its identified city, but given its PUMA’s population distribution, it is likely that the household resided in the city. This protocol yields errors of omission (where a CITY code is not assigned to some residents of the corresponding city) and errors of commission (where a CITY code is assigned to some non-residents). As an index of mismatch for each CITY code, IPUMS uses the sum of percent omission error (the portion of a city’s population residing in excluded PUMAs) and percent commission error (the portion of the population in associated PUMAs that did not reside in the city).