Using for instance the 1990 5% Census, I have come across observations with a 0 value for PERWT. This means that these observations are merely dropped from weighted regressions. I therefore wonder there is an interpretation to this, or whether it’s a mistake.
In 1990-2000, PERWT values of 0 were a function of the complex sample design used by the Census Bureau. These weights were generated by the Census Bureau so IPUMS cannot offically speak to how they were created. I believe these 0 weights were used to account for oversampling, but if you would like more official information you can contact the Census Bureau.