The Survey of Income and Program Participation is a major household survey conducted by the Census Bureau, which is intended to be an important source of information on the economic situation of persons and households in the United States. In the current design a new sample panel is selected each year. A person in a SIPP panel is generally interviewed eight times over a period of 2 2/3 years, with each round or wave of interviewing collecting information for each month of a four month reference period. Although the survey has cross-sectional uses, a major interest is in longitudinal estimates. Under current procedures, a sample person who misses any interviews may be excluded from the longitudinal estimates, sometimes, as explained later in the paper, even for estimates for time intervals which do not overlap any missed interviews. Concern has been expressed by some data users, particularly the Food and Nutrition Service (FNS), over the detrimental effects on variances and biases of the exclusion of these sample cases, particularly the cases that miss some interviews but later return to sample. The purpose of this paper is to investigate the implications of this exclusion and to consider alternative approaches.

The structure of this paper is as follows. In order to fully understand the problem of the sample omitted from the longitudinal estimates, some knowledge is required of the SIPP design, the noninterview problem for this survey and the weighting procedures used. These areas are briefly reviewed in Section 2. In Section 3, using data from the 1984 SIPP panel, estimates are compared from four groups which partition the set  of people for whom at least one interview was obtained as part of this panel. These four groups, which a redefined more precisely in Section 3, are roughly the portion of the sample interviewed for all waves (group 1), the portion interviewed the first wave, but who eventually leave permanently (group 2), the portion interviewed the first and last waves but who miss some interviews (group 3), and the portion first interviewed subsequent to the first wave (group 4). The comparisons are for certain demographic and economic characteristics at the time of each sample person's first interview, and also for gross change estimates for these characteristics. The cross-sectional weighting system which, unlike the longitudinal system, assigns positive weights to all sample cases for any month that an interview is obtained, is used in these comparisons. Also in Section 3, first wave estimates for these characteristics obtained from the cross-sectional weighting system, which, as just noted, includes the sample cases that miss later interviews, are compared to the same estimates obtained from the longitudinal weighting system which, in the file used, excludes these cases. This comparison provides some insight on the effects of the longitudinal weighting adjustments in compensating for the exclusion of the sample that miss interviews.

Finally, in Section 4, the following possible modifications of the current longitudinal weighting procedures are discussed: inclusion of group 3 cases into the estimates after imputation for missing waves, which might also be combined with a weighting adjustment of the group 3 cases to compensate for exclusion of group 2 cases; inclusion of group 2 cases; inclusion of group 4 cases; and use of multiple sets of weights to cover different time intervals.