about 3 years (depending on the panel). Panel sizes have varied from 14,000 to 36,700. About 31,700 households were interviewed at the start of the 2001 panel (for details, see Weinberg, 2004). The SIPP collects data on a number of items that the ASEC does not, such as child care and other work-related expenses. The panel design allows one to calculate poverty over different durations (such as a month, a year, or multiple years) and to track how families are doing over the life of a 3- or 4-year panel.
The SIPP also has shortcomings. An important one is that many people drop out of the survey over the course of the panel, which likely introduces some bias into the poverty estimates over time. Studies have shown that low-income households are more likely to drop out of the survey than others. This bias could be overcome by reintroducing “overlapping” panels (a strategy that was dropped after the 1993 panel), in which a new 3- or 4-year panel is implemented every year. This approach would produce annual poverty estimates that come from three or four panels that are simultaneously in the field.
Some of the other shortcomings in the SIPP have been addressed in recent panels. Wage and salary information tends to be underreported in the SIPP, though an improved questionnaire implemented in the 2004 panel may reduce the magnitude of this problem. Prior to 2004, the SIPP also did not have state-representative samples in all states. While the 2004 panel does have an improved design to address this issue, the small samples in a few states will produce poverty estimates that are not very reliable for those states. Reintroducing overlapping panels may help address this problem too. Finally, while the SIPP collects information on taxes, the data are of poor quality. There are efforts now under way to model what families pay in taxes (and refunds they receive from the Earned Income Tax Credit) in the SIPP; these models are somewhat similar to CPS tax models.
John Czajka (Mathematica Policy Research) said that using data from the SIPP rather than the CPS has several advantages. He cautioned, however, that the SIPP still needs to be improved in a few ways. In addition to underreporting of earnings, he mentioned that data have to be released from the Census Bureau in a more timely manner. He noted that while an overlapping panels design is important for addressing the bias arising from people dropping out of the sample, it may involve making tradeoffs if, for example, each of the panels contain smaller sample sizes (which reduce the reliability of estimates from any given panel). Overlapping panels may need to be smaller because of the expense it takes to concurrently field multiple surveys.