The author considers the problem of estimation in a panel data sample selection model, where both the selection and the regression equation of interest contain unobservable individual-specific effects. He proposes a two-step estimation procedure, which 'differences out' the sample selection...