This analysis applies allocation theory to order US schools for limited measles vaccine doses by their marginal return to herd immunity, instead of by lowest coverage. Across a population of 36,031 US schools, the commonly binding case is that two thirds to three quarters of susceptible kindergarteners attend schools where each additional dose yields increasing marginal herd-immunity benefit. When allocation follows this marginal-return ordering rather than a lowest-coverage-first rule, the authors report a gain of 1.92 times in indirect protection and 1.16 times in total protection.
The theoretical optimum that orders by marginal herd-immunity return requires an assumption that each school has an equal chance of seeing an introduction (case). The study interrogates that assumption using outbreak records where exposures and susceptible headcounts can be compared directly.
The authors examine pre-outbreak records from a 2025–26 Upstate South Carolina outbreak and find that exposed schools are over-represented among those with the largest susceptible headcounts. Specifically, schools that experienced exposure were over-represented 5.0-fold in the top decile of susceptible headcount (95% CI 3.0 to 7.3). As a negative control, plain enrollment (which carries school size information but not susceptibility) did not show the same over-representation, arguing that exposure was associated with susceptibility rather than simply school size.
An independent outbreak in Clark County, Washington reproduced a similar scaling of exposure risk with susceptible headcount, supporting the South Carolina finding. However, the authors note that only two US jurisdictions publish the records necessary to perform this test, so how steeply introduction risk scales with susceptibility at a national level remains unidentified.
Given the uncertainty about how introduction risk scales with susceptibility nationally, the authors evaluate seven allocation rules for vaccine targeting. The point-estimate-optimal rule (the theoretical marginal-return ordering) is the least robust under uncertainty: in the worst-case scenario within the uncertainty set the theoretically optimal rule can lose 86% of the attainable benefit. This fragility persists across different ways of drawing the uncertainty set, indicating the problem is structural rather than a quirk of a particular parameter choice.
By contrast, allocation rules that incorporate a degree of hedging against measured risk perform more robustly when introduction risk is uncertain. The authors quantify that a light hedge informed by measured risk variables retains roughly 90% of attainable benefit at the primary budget level, as computed using the two jurisdictions that publish the requisite outbreak-susceptibility columns.
The study frames robustness in terms of worst-case losses relative to the unattainable ideal under complete information. The point-estimate-optimal allocation maximizes expected return under the equal-introduction assumption but is sensitive to deviations from that assumption. When introduction risk actually concentrates in high-susceptibility schools, as suggested by the two outbreak records, the point-estimate-optimal rule's advantage diminishes or reverses. The reported 86% loss in the worst case illustrates how much potential benefit can be forfeited if exposure risk is unmeasured and allocation ignores it.
The practical implication is that optimization on a well-measured variable (for example, susceptible headcount or coverage) while leaving exposure risk unmeasured will frequently produce a fragile point estimate that performs poorly under plausible alternative scenarios. Robustness metrics and stress-testing across uncertainty sets are therefore critical when designing allocation policies.
The authors recommend a modest hedging approach: blend the point-estimate optimum that targets marginal herd-immunity returns with measured indicators of exposure risk when such indicators are available. Using outbreak-derived columns from the two jurisdictions, a light hedge preserved roughly 90% of the benefit at the primary budget, substantially reducing the worst-case losses seen with the unhedged theoretical optimum.
Because only a small number of jurisdictions publish the detailed outbreak exposure and susceptible headcount data needed to assess how introduction risk correlates with susceptibility, jurisdictions planning allocation should consider both the potential for concentrated introductions and the limits of their data. Where exposure risk is not measured, conservative hedging reduces the risk that allocation will underperform during real-world introductions.
The authors state that data and code used in the analysis are available on GitHub (https://github.com/BentoLab-Cornell/measles-allocation). They also note a major limitation: only two US jurisdictions publish the records necessary to test how introduction risk scales with susceptibility, so the national relationship between susceptibility and introduction risk is not identified from the available data. Other specifics of the modeling choices, the seven allocation rules evaluated, and the exact operationalization of the hedge were described in the source analysis; readers should consult the preprint for full methods and supplementary materials.
The preprint was posted August 24, 2026, and the authors declare no competing interests.