College enrollment and completion rates are highly disparate by family income. Motivated by this fact, we consider the causal effect of family income on college enrollment and completion in the potential outcomes framework. In contrast to a standard regression framework, we begin from a default assumption that the causal effect of income is nonlinear and heterogeneous. In contrast to most work using the potential outcomes framework, we consider income as a continuous treatment condition. The relevant questions are what levels of income are helpful and for whom. Leveraging flexible machine learning estimators, we show two main results. College enrollment rates increase most steeply with income among children in low-income families, while college completion rates increase most steeply with income among high-income families. These distinct patterns for college enrollment and completion suggest that improvements to the financial circumstances of poor families might promote equity in college enrollment, but equity in completion may require other interventions. This talk is based on joint work with Jennie E. Brand.