基于二水平误差控制的序贯分支筛选法及应用

针对仿真实验中三水平序贯分支（sequential bifurcation，SB）难以处理定性因子的问题，提出了基于两水平的误差控制SB法，并在序贯概率比（sequential probability ratio test，SPRT）检验群组效应过程中，采用动态的初始仿真次数作为初始样本点，提高了SPRT的效率，减少了统计推断所需的仿真次数. 蒙特卡罗仿真的对比分析以及基于实际背景的仿真案例应用，验证了在保证统计推断效力的条件下，所提方法能明显降低SB所需的仿真次数，并成功处理包含定性因子的仿真模型.

Simulation factor screening and application based on two-level controlled sequential bifurcation

To tackle the screening problem that qualitative factors are difficult to identify by three-levels sequential bifurcation (SB) method within simulation experiment, this paper proposed a controlled error SB method based on two levels, in which dynamic initial simulation replicates was incorporated into effect test procedure. The new method can improve the efficiency of sequential probability ratio test (SPRT), and decrease the replicated required in statistical inference. The comparative analysis of Monte Carlo simulations, and the corresponding application results from a real-world case study indicate that the proposed method indeed greatly reduces replicates needed in SB procedure under ensuring the power of statistical power, and can handle qualitative factors of a simulation model successfully.