We propose a complete framework for data-driven difference-in-differences analysis with covariates, in particular nonparametric estimation and testing. We start with simultaneously choosing confounders and a scale of the outcome along identification conditions. We estimate first heterogeneous treatment effects stratified along the covariates, then the average effect(s) for the treated. We provide the asymptotic and finite sample behavior of our estimators and tests, bootstrap procedures for their standard errors and p-values, and an automatic bandwidth choice. The pertinence of our methods is shown with a study of the impact of the Deferred Action for Childhood Arrivals program on educational outcomes for non-citizen immigrants in the US.