Using a new nationally representative dataset, we find minor differences in test outcomes between black and white infants that disappear with a limited set of controls. However, relative to whites, all other races lose substantial ground by age two. Combining our estimates with results in prior literature, we show that a simple model with assortative mating fits our data well, implying that differences in children's environments between racial groups can fully explain gaps in intelligence. If parental ability influences a child's test scores both genetically and through environment, then our findings are less informative and can be reconciled with a wide range of racial differences in inherited intelligence.

We investigate learning by doing using detailed data from a major auto producer’s assembly plant. We focus on the acquisition, aggregation, transmission, and embodiment of the knowledge stock built through learning. We find that most knowledge was not retained by plant workers despite their importance as a learning conduit. This is consistent with the plant’s systems for productivity measurement and improvement. We further explore how learning at the hundreds of processes along the production line undergirds plantwide productivity. Our results shed light on how productivity gains accrue at the plant level and how firms apply managerial inputs to expand production.