This paper develops and applies a method for decomposing cross section variability of earnings into components that are forecastable at the time students decide to go to college (heterogeneity) and components that are unforecastable. About 60 % of variability in returns to schooling is...