The economic causes and effects of migration from developing countries depend on patterns of self-selection that are difficult to observe. We estimate the degree of migrant self-selection—on both observed and unobserved determinants of income—for 99 developing countries using nationally representative survey data on 653,613 people. In low-income countries, people actively preparing to emigrate have 14 percent higher incomes explained by observed traits such as schooling, and 12 percent higher incomes explained by unobserved traits. The simulated income elasticity of emigration is positive in the aggregate (+0.23) despite being negative in subpopulations, an instance of Simpson’s paradox.