The Survey of Income and Program Participation (SIPP) is a longitudinal survey which began in October 1983. The survey is intended to measure extremely complex phenomena: detailed income sources, recipiency of Federal and state aid, weekly labor force status, health status and health insurance coverage, taxes, assets, and interest income. In addition to these types of information, the survey collects information on demographic characteristics of all household members. During each visit, interviewers ask questions from a core questionnaire and also, on most visits, ask variable sets of additional questions in the form of "topical modules" on particular issues, such as child care or educational financing.

Approximately 20,000 households formed the "1984 panel." Persons living at the selected addresses were initially interviewed (in four equal-size groups) between October 1983 and January 1984. The s ample of addresses was selected to be nationally representative of the civilian noninstitutional population of the United States. Persons whose usual residence was at one of the selected addresses were then scheduled for interview once every four months throughout the 2 2/3 years of the 1984 sample's life, that is, into the summer of 1986. In February 1985 and in February 1986, new smaller samples were introduced and a new sample will be introduced each year in the future. More details on both the structure and content of SIPP are available in SIPP Working Paper No. 8401 Update, "An Overview of the SIPP" by Nelson, McMillen, and Kasprzyk.

While other major surveys such as the Current Population Survey and the National Crime Survey, both conducted by the Census Bureau, return to the same address for each subsequent visit regardless of whether the occupants of the address change, the SIPP interviewer returns to interview the same persons who form the sample. Persons who move in with SIPP sample persons after the first interview, while they live with sample persons, are also included in the sample and interviewed. If persons move to a new address, they are followed and interviews are obtained at the new address (for more information about mover's procedures, see Jean and McArthur, 1984.)

Throughout the sample period efforts are made to continue to interview all persons who are ever part of the sample--even if they move to other parts of the country--with a few exceptions: persons who moved into households with sample persons after the first interview are not followed unless they moved with those sample persons; persons who are institutionalized, move outside of the United States, or move to an Armed Forces barracks are not followed; and children under 15 who move and are not accompanied by a sample person who is 15 years old or over are not followed.

Sample maintenance is an important issue in any survey operation. It is especially important when that survey is longitudinal; the Survey of Income and Program Participation (SIPP) follows its sample population through 2-2/3 years. This paper represents a continuation of the work presented during the 1985 annual meetings of the American Statistical Association (see McArthur and Short, 1985).

The goal of this work is to determine whether attrition from the survey is basically a random phenomenon or whether it is systematic. If attrition is indeed related to personal characteristics, it may be possible to use these results to improve field procedures or adjust weighting specifications. Insofar as improvements are not possible, quantification of individuals leaving the sample is necessary for analysis of succeeding panels of data. If, indeed, attrition is systematic then account should be taken of the degree to which this occurs in statistical analyses of these data. This is so whether the analysis is a simple description of characteristics in a cross-section of the data set involving a subsequent interview or if the analysis is a dynamic one involving several successive interviews at once. Just as individuals with particular characteristics may leave the sample with greater frequency, so individuals experiencing particular events may be more likely to leave the sample. For example, an individual who loses a job may be temporarily absent or may move out of the area, and thus, not be located by an interviewer for a subsequent wave or waves. If this occurs, any estimation of the probabilities of an event's occurrence over time will be biased to some extent by nonrandom attrition. For a theoretical discussion of this issue see Cox (1959), Gail (1975), Fisher and Kanarek (1974) and Williams and Lagakos (1977).

Thus, this paper describes our findings in several areas: interviewing patterns, reasons recorded for noninterview, and characteristics of sample members by their interviewing status, and finally some initial work studying the relationship between the occurrence of events, such as a change in marital status or employment status, and interviewing status.