Using a combination of Dutch survey and administrative data, we show that survival expectations do in fact predict actual mortality in models that control for income and education level. This predictive power disappears, however, when controls are introduced for self-rated health status and...