Who Chooses and Who Benefits? The Design of Public School Choice Systems

Public school choice has evolved rapidly in the past two decades, as districts roll out new magnet, dual-language, and themed programs to broaden educational opportunity. We use newly collected national data to document that opt-in (voluntary) systems: (i) are the modal design; (ii) are harder to navigate; and (iii) have participation that is concentrated among more advantaged students. These facts suggest a striking inconsistency: districts have largely adopted centralized assignment algorithms to broaden access, but most rely on optional participation that fragments public education. We study the implications of this design choice in the Los Angeles Unified School District, the largest opt-in system in the country, combining nearly two decades of administrative data, randomized lotteries, and quasi-experimental expansions in access. Participation is highly selective, consistent with national evidence, and lottery estimates suggest that the students with the lowest demand for choice schools are the ones who gain the most from attending. Opt-in participation therefore embeds a selection mechanism that screens out high-return students and leaves many effective programs with unused capacity. To evaluate system-level implications, we estimate a structural model linking applications, enrollment, and achievement. Choice schools are vertically differentiated and generate meaningful gains, but the opt-in participation rule -through high application costs and negative selection on gains- prevents these benefits from reaching the students who need them most. Counterfactual simulations make the design stakes clear: information and travel-cost reductions have limited effects, whereas reforms that change the participation architecture eliminate core inefficiencies and deliver the largest district-wide achievement gains. These results underscore that system design -not school effectiveness alone- shapes who benefits from public school choice and to what extent.