Job assignment and bandit problems

The paper is concerned with dynamic job assignment when observed performance is an imperfect signal of the worker's type. When the rate of learning from past performance depends upon the particular job performed, promotion can be due to good performance only at a job for which the resulting probability of mistaking a low-ability type for a high-ability type is higher than for the job the worker is upgraded to. Income risk can be greater for old workers than for young workers. The length of the worker's active fife is relevant for job mobility notwithstanding optimal myopic procedures for job assignment. The dynamic perspective induced by learning can generate new forms of opportunism.