不同Halton抽样方法在混合Logit模型中的比较

摘要

Several Halton sampling methods which are widely used in mixed logit model parameter esti-mation are taken as the research objects ,including Halton ,scrambled Halton ,shuffled Halton and derandomized Halton sampling .Based on box plot of mean absolute percentage error ,parameter esti-mators of different Halton samplings are compared under different dimensions .Meanwhile ,standard error of hundreds of probability simulation is analyzed to judge the applicability of different methods . T he result show s that :under low-dimension ,all Halton sampling methods have almost no difference in parameter estimation results ,and shuffled Halton sampling is the same as pseudo random sam-pling ;with the increase of dimension ,distribution uniformity of sampling points from Halton sam-pling methods is gradually approaching to pseudo-random sampling ,w hile the scrambled Halton sam-pling is better than the other three Halton sampling methods .%以混合Logit模型参数估计中应用最广泛的四类 Halton抽样方法为研究对象，包括 Halton、变序Halton、乱序 Halton和去随机化 Halton ，借助平均绝对百分误差箱型图比较不同维度、不同抽样方法得到的参数估计值，同时结合数百次模拟概率标准差判断不同抽样方法适用性。结果表明，在低维度情况下，各Halton抽样方法参数估计结果相差不大，且乱序 Halton抽样性质与伪随机抽样无异；而随着维度的增加，几类 Halton的抽样点均匀程度逐渐与伪随机抽样接近，且变序Halton抽样优于其他3类Halton抽样。