This paper uses random assignment to estimate the causal impacts on child skills of a widely emulated early childhood home visiting program. We show the feasibility of replicating it at scale. We estimate vectors of latent skills for individual children and compare treatments and controls. The program substantially improves child language and cognitive, fine motor, and social-emotional skills. We go beyond reporting treatment effects as unweighted item scores. We determine whether the program affects the latent skills generating correct answers to lists of test items and how the program affects the mapping from skills to item scores. Enhancements in latent skills explain most of the conventional treatment effects for language and cognition. The program operates primarily by improving skills and not by improving how effectively skills are used. The program barely changes the map from latent skills to item test scores.