We use a flexible hazard rate model with unrestricted spell duration and calendar time effects to analyse a dataset including all Norwegian unemployment spells during the 1990s. The dataset provides a unique access to conditionally independent variation in unemployment compensation. We find that...