Inside Decentralization: How Three Central American School-based Management Reforms Affect Student Learning Through Teacher Incentives Ilana Umansky Emiliana Vegas Despite decentralization reforms of education systems worldwide, there is little empirical evidence about the processes through which decentralization can improve student learning. This article contributes to the understanding of how decentralization reforms can improve learning and shows how education reforms, even when not conceptualized as affecting teacher incentives, can generate important changes for teachers that, in turn, affect student learning. The goal is to broaden the conception of how education reforms affect teachers by influencing teacher incentives and to explore how to design and implement these reforms to maximize their beneficial effects on teaching and learning. This article shows that education reform design should consider the potential impact on teaching quality, even when reforms are not specifically intended to alter the incentives that teachers face. She finds that greater teacher autonomy in implementing projects and designing teaching plans is associated with better student outcomes when school decisionmaking power is close to the level of the teacher. After the war ended, the central government acknowledged the success of these schools in providing education cost-effectively in remote areas and in 1991 decided to expand the program. If a school's choice or assignment into the program is based on expected benefits from participation, such as a school's need or likelihood of success, as is almost certainly the case in the Central American reforms, then the characteristics of those who participate will not be comparable to the characteristics of those who do not. These include evaluation of participating schools before and after the reform; matched comparison, in which a comparison group is chosen to match the observed characteristics of the treatment group; propensity score matching, which involves constructing a comparison group based on its conditional probability of receiving treatment given a set of observable characteristics; and natural experiments, which are naturally occurring experimental conditions due to quirks, isolated changes, or idiosyncrasies in EDUCO Schools Report Significantly More School-level Control Than Comparison Schools in Few Areas Area Ordinary least squares Propensity score matching Determine salary Determine teacher incentives Evaluate teachers Give teacher incentives Hire and fire administration Hire and fire director Hire and fire teachers Spend school money Teachers association activity Teacher supervision 0.05 0.10 20.01 0.03 0.33*** . Conclusions Previous research on educational decentralization has explored the impact of various decentralization reforms on indicators such as student learning and completion.