Effects of Distributional Assumptions in Stochastic Frontier Models on Efficiency-Based Unit Targeting

Key Points

• This research aims to systematically assess how different distributional assumptions in stochastic frontier models affect efficiency-based intervention selection.
• Conducted extensive simulations varying sample size, inefficiency signal strength, and true inefficiency distribution.
• Introduced relative predictive improvement and selection accuracy as key performance metrics.
• Analyzed the stability and limitations of selection lists under different distribution assumptions.
• Found that changes in distribution assumptions did not significantly alter the selection lists among similar assumptions.
• Identified that total estimated improvement effects were sensitive to distribution settings, especially under weak inefficiency signals.
• Demonstrated the need to evaluate and report both selection list stability and uncertainty in improvement effect estimation.