This paper reports the general results of research undertaken by Census Bureau staff. The views expressed are attributable to the author and do not necessarily reflect those of the Census Bureau.

The Survey of Income and Program Participation (SIPP) currently uses cross-classifications of age, race, sex, and householder/nonhouseholder status as controls in longitudinal estimation. The controls come from the Current Population Survey (CPS), which has its own controls based on post-censal estimates of age, race and sex. Previous research by Huggins and Fay [1988] ratio adjusted the SIPP 1984 sample that could be matched to IRS records. They adjusted the matched records to IRS-reported age, race, sex, and adjusted gross income. They did not control the nonmatched sample. Their adjustment produced a reduction in variances for most income and program participation variables.

Subsequent research by Dorinski and Huang [1994] applied demographic totals based on the CPS controls for age, race, sex, and ethnicity, to ratio adjust the estimates based on the SIPP sample that did not match to the IRS records. We combined the nonmatched and matched samples and then calculated estimates along with their variances. We found significant variance reduction, over previous research that did not adjust non-matched cases, for many of the variables examined.

Final results indicated large reductions in variances for many income and income related characteristics, with some variances affected adversely. Some variance estimates for Hispanics and to a lesser extent Blacks increased. Bias of the estimates studied either did not change or increased.

Due to some of the adverse results for Black, Hispanic, and program participation estimates, we decided to research the methodology on a more recent panel before adding it to the current SIPP weighting procedure. We chose the 1990 SIPP panel because it contained an oversample of households headed by Blacks, Hispanics, or females with no spouse present living with children under age 18. We focused on the respondents for calendar year 1990.

The next section outlines the methodology used. The succeeding sections discuss the differences from the 1984 panel research, variance results and effects of the new weighting on the bias. The final section presents recommendations.