It sounds like you are asking about “-1” values for PULINENO in the original Census Bureau CPS data, please correct me if I am wrong. The original Census data is a flat person-level file; however, not all households in the original data have an interview, resulting in empty households being represented by a single “person” that are NIU (blank) for all person-level variables, including PULINENO. If you run a crosstab of PULINENO and HRINTSTA (Interview Status) in the original data, you will find that all PULINENO == -1 cases are non-interview households. By default, IPUMS CPS data come as rectangularized person-level files; because there are no person records for non-interview households, these cases with invalid/blank LINENO records will not be included. However, if you were to download a hierarchical file from IPUMS CPS you would see these empty households (because there is one “person” record per non-interviewed household, I suspect the number of household records with NUMPREC values of 0 would be equal to the number of PULINENO values of -1 in the original data).