In this paper we try to test the hypothesis that sickness absence spells depend on socio-economic factors related to occupation. Proper statistical methods have been used in order to control the effect of recurrences on the inferences that can be drawn. The data consist of two cohorts formed by workers from two companies making up a public transport firm, and covering the period 1994-1996. The analysis was carried out separately by company and by contingence. Using a longitudinal design we estimated several conditional models, in particular Markovian Andersen-Gill models. The results show that occupation appeared as the predictor of sickness absence spells with highest relative risk incidence. Mechanical and railway operators, maintenance assistants, and drivers notified highest incidence rates, after adjusting by personal and work characteristics; health related behaviours, and health antecedents. Other characteristics of high incidence rates were: being female, between thirty and sixty years old, less than nine years of tenure, married having a child; uncomfortable shift work; those with lower educational level, being smoker or ex-smoker, and having chronic disease antecedents. Given the social security system, it is very likely that some common diseases were actually work-related diseases. This fact prevents up from separating “voluntary” from “involuntary” absence spells. The findings with respect to occupation can be mainly explained by reference to “adverse” work conditions.

Key words: sickness absence spells, occupation, Andersen-Gill, Markovian models.
JEL classification: C41, I12, J22, C14.