Abstract:

The problem of occurrence of recurrent (multiple) events occurs in many areas. In medical science, an example is the occurrence of chronic disease with a recurring course. In economic and social sciences, it is possible to analyse the time of entering and leaving the sphere of poverty, the time of subsequent guarantee or insurance claims, as well as the time of subsequent periods of unemployment. The method of determining risk intervals, i.e. the time at which an entity is exposed to the risk of an event, may differ. It depends on the adopted time scale. The aim of the article is to analyse the risk of subsequent registrations in the labour office depending on selected characteristics of the unemployed: gender, age, education and seniority. In the study, methods of survival analysis were used. The results obtained for four models were compared: Anderson-Gil model (the counting process model AG), Prentince-Williams-Peterson conditional models: total time (PWP-CP) and time gap (PWP-GT), and Wei, Lin and Weissfeld model (marginal model WLW). These models are the extensions of the Cox proportional hazards model. The AG model does not distinguish between first and next events. The number of events that occurred is important. The other three models are based on the stratified Cox model. They differ in the way risk intervals are determined. The strata are consecutive events. The study used anonymous individual data from the Poviat Labour Office in Szczecin.