The genesis of this paper is a pair of articles that appeared in The Journal of Economic and Social Measurement in December 1985. Although this journal issue was devoted to the design and development of the Survey of Income and Program Participation, or SIPP, most of the articles had broader implications for analysis of panel data in general.

The first of the two articles, by David Byron McMillen and Roger Herriott, was entitled, "Toward a Longitudinal Definition of Households." The authors distinguished among a static definition of a household--e.g., everyone living at a given address in December of 1980; a dynamic definition of a household--a potentially changing collection of individuals over the course of 1980; and an attribute-type definition of a household--the potentially changing set of people who shared an address at any time over the course of 1980 with a particular individual. These three approaches have very different implications of measurement of household-level phenomena such as poverty or welfare receipt. The authors concluded that both dynamic and attribute-type definitions were desirable, to answer different sorts of questions.

The second article, by Greg J. Duncan and Martha S. Hill, was entitled, "Conceptions of Longitudinal Households--Fertile or Futile?" These authors argued that there is no satisfactory way to define a longitudinal household, and that the only appropriate analysis was based on an attribute-type definition, or as they preferred to call it, on an "individual-based" approach. They claimed that in principle, use of a dynamic definition of households could give misleading results, because the events that move households in and out of poverty, or on and off welfare, often change the identity of the household beyond recognition. Whether that occurred in practice, however, was an open question. In this paper, I report results from the 1984 Survey of Income and Program Participation that indicate that the implications of choosing between an individual- or household-based approach for estimating the length of food stamp receipt are virtually nil.