Response error in longitudinal household surveys has been a continuing and important research topic for SIPP (the U. S. Census Bureau's Survey of Income and Program Participation). As in most surveys, SIPP response errors have been difficult to detect and difficult to reduce once we learn about them.

This paper describes an ambitious project whose goal was to reduce SIPP response error importantly by radically changing the way income data were collected in the interview. The paper covers the changed procedures, the experimental design used to test them, implementation issues, the complex uses of administrative records for estimating errors, the results of the evaluations and how the redesigned SIPP interview will address issues raised in the research.

Results indicate that most households will use personal records in reporting income and that record use can improve the quality of reporting income amounts over time. But the fundamental errors in SIPP income reports are failing to mention entire income sources. The new procedures neither anticipated nor corrected this flaw. Redesigned SIPP will encourage households to use their personal records for reporting income amounts. But we are still unsure how to correct the whole source underreporting problem.