Presents techniques for evaluating the performance of “job agents” — a new class of California civil servant established to help the hard-core unemployed and the disadvantaged get jobs. Job agents are to be paid on an incentive basis, their rewards depending on improvements in clients’ earnings achieved through the agent’s help. To gauge those improvements, it is necessary to predict what the client’s employment status and wages would have been without the agent’s help. A model is devised for making those predictions, and an adequate control group is found in a set of survey data compiled on the income and employment patterns of a representative sample of U.S. families over several years. Fitted to the model, these data allow estimation of a client’s employment situation without the agent’s help. The techniques described also allow interim incentive payments to a job agent when a client is placed.