ownership services), will increase the data needed for measuring poverty. SIPP, with its focus on income data, is in a position to respond to these needs; the March CPS, which must always be geared primarily to the requirements of the nation's main labor force survey, is not. Hence, we recommend that SIPP become the basis of the nation's official income and poverty statistics in place of the March CPS. This change should take effect when the slated improvements to SIPP are introduced in 1996.
A decision to use SIPP to produce the official poverty rates means that the SIPP design and questionnaire must be reviewed to determine if modifications are needed to enhance the survey's ability to provide accurate statistics under the proposed measure. (A panel that recently evaluated SIPP made a similar recommendation about using SIPP for income and poverty statistics [Citro and Kalton, 1993:85-87], and many of its recommendations on the SIPP design and questionnaire are relevant.)
In regard to the overall SIPP design, we are concerned that the Census Bureau's decision for 1996 to have new samples (''panels") introduced every 4 years, each of which is followed for a 4-year period, may be problematic for providing a reliable time series of annual poverty statistics because of biases that result from attrition from the samples over time. Every 4 years there may be a break in the time series because of the introduction of a new sample; in addition, because there is no overlap between the samples, it will be difficult to evaluate whether the changes in the poverty rate are real or not.
Such a nonoverlapping design also limits the usefulness of SIPP to analyze important policy changes, such as changes in welfare programs or health care financing: if policy changes take effect near the beginning or end of a 4-year sample, there is limited information available either before or after the change to adequately evaluate its effects. The SIPP evaluation panel recommended that SIPP samples be followed for 4 years but that a new sample be introduced every 2 years. Poverty rates under this design may also be affected by attrition and other biases, but, with the sample overlap, it will be possible to evaluate and, one hopes, adjust for the effects. Also, under this design, a new sample is in the field every 2 years, which should facilitate analysis of policy changes.28
It is important to carry out methodological research that can lead to yet further improvements in SIPP data quality for purposes of poverty measurement. A high priority is research to improve the population coverage in SIPP (and other household surveys), especially among lower income minority groups, particularly young black men (the Census Bureau has such research
The disadvantage for longitudinal analysis of the overlap design recommended by the SIPP panel is that the sample size is half that of the design of 4-year samples with no overlap; however, for the estimation of annual poverty statistics, the total sample size of the overlap design, added across the two samples in the field each year, is the same as that of the nonoverlap design.