Optimal sample allocation in stratified surveys for a single characteristic is well understood, yet multivariate allocation remains challenging. In the absence of a total order on the precision vector, a decision-maker must balance competing objectives through preference weightings whose effects on the resulting allocation are generally unclear. This paper develops an analytical sensitivity analysis for multivariate optimal allocation based on the Karush-Kuhn-Tucker conditions. We show that a stepwise weighted Chebyshev minimisation yields a unique efficient allocation and the normal vector to the Pareto frontier, from which marginal rates of substitution between survey objectives and shadow prices for cost and precision constraints follow directly. The method enables interpretable adjustments to weightings or cost limits when specified allocations do not meet survey purposes. Numerical results on an educational survey quantify the impact of budget and quality constraints on estimation precision and demonstrate how the trade-off information guides iterative refinement of preference weightings.

翻译：暂无翻译