Do the determinants of employment duration vary across employment spells?

This paper looks at employment spells beyond the first, to check if multiple employment separations improve the matching of skills between workers and firms. We investigate whether the factors that are relevant for the first spell remain relevant for later spells. The results point to a range of outcomes. Certain variables such as the unemployment rate gap or part-time employment remain strong consistent predictors of the separation hazard across spells. In contrast, the coefficients of some variables such as the education or family status of the worker change sign and significance as the number of spells increases. This pattern is explained using a theoretical model that connects the cross-spell impact of a variable to its impact on the underlying distribution of the tendency to separate across workers. Variables associated with either a very low or a very high tendency to separate have an inconsistent effect across spells, because workers with a low tendency to separate do not appear in subsequent spells. Additionally, we find that the results vary across genders and generations.Classification codes: J00, C14, C12, C41.