Survival Analysis for the Survey of Income and Program Participationi
I-Li Lu and Lynn Weidman
93/01
ABSTRACT
Denote Tij as the spell duration of unit i's jth participation in some government sponsored
program Tij measures the length of time that an individual takes to exit the program. In this
paper, we use data extracted from the 1987 panel of the Survey of Income and Program
Participation (SIPP) to investigate the behavior of Tij. We study the effects of time and
explanatory variables on the hazard function h(t), which is the conditional density (probability)
that a unit exits the program at time t when in fact, this unit is still at risk of an exit at time t.
We discuss some complications that normally arise in data extracted from a longitudinal panel
survey and develop a general estimation procedure for the distribution of Tij. Our procedure
provides a solution for the following problem: (1) multiple occurrences, (2) random left
truncation, (3) random right censorship, (4) random number of repeated occurrences, and (5)
dependencies of truncation or censoring mechanisms on the spell durations. We use the EM
algorithm when left-truncated spells are present, and the number of spells for each unit is treated
as random in the analysis.
