This paper studies educational technologies that use artificial and human intelligence to enhance writing instruction. In a large-scale experiment, providing students with an online platform to practice and enabling teachers to outsource grading/feedback tasks to artificial intelligence improved essay scores on a high-stakes exam. Adding human graders to circumvent potential limitations of automated systems improved the perceived quality of feedback, but did not affect the extent to which this enhanced educational technology improved achievement. Both technologies shifted teachers’ tasks to individualized instruction, suggesting that teachers used them as complements to the educational inputs they provide to develop complex writing skills.

Other materials: J-PAL Evidence to Policy | J-PAL Summary  |  AEA RCT Registry

Media Coverage: Nexo | Valor Econômico  | O Globo | Porvir

This article identifies the probability of causation when there is sample selection. We show that the probability of causation is partially identified for individuals who are always observed regardless of treatment status and derive sharp bounds under three increasingly restrictive sets of assumptions. We use experimental data from the Colombian job training program Jóvenes en Acción to empirically illustrate our approach’s usefulness.

Other materials: Ungated version.

This paper describes the results from an evaluation of a public policy that offers scholarships to current and former public high school students, so that they can attend technical and vocational education courses free of charge. We use a waiting list randomized controlled trial in four municipalities in a southern Brazilian State (Santa Catarina) to quantify the effects of the program on school progression, labor market outcomes and non-cognitive skills. Our intention-to-treat estimates reveal substantial gender heterogeneity two years after program completion. Women experienced large gains in labor market outcomes and non-cognitive skills. Employment rose by 21 percentage points (or approximately 33%) and the gains in earnings are of more than 50%. Also, women who received the offer scored 0.5σ higher on the synthetic index of non-cognitive skills and 0.69σ higher on an extraversion indicator. We find no effects on the male subsample. These findings corroborate the evidence on gender heterogeneity in the labor market effects of technical and vocational education programs. We also perform a series of exercises to explore potential channels through which these effects arise.