参数法和非参数法估计组合样中被测组分含量的置信区间

Parametric and Bootstrap Estimations of Confidence Intervals for Analyte Concentrations of Composite Samples

本文以散装生铁块的随机样本分析结果为基础 ,应用 Monte Carlo模拟法考察了组合样本中 Si、Mn、C、S、P含量 (或份样平均值 )的分布情况 ,研究了组分含量的置信区间 ,对参数法和非参数法的结果进行了比较。结果表明 ,对近似符合正态分布 (或 t-分布 )的原始总体或当组合样中的份样数目较多时 ,两种方法所估计的置信区间相近。但对偏离正态分布 (或 t-分布 )的总体且组合样中的份样数目较少时 ,非参数法较参数法的结果更符合实际。

Based on the initial analytical results of pig irons, Monte Carlo technique was used for the simulation of composite sampling. The confidence intervals of the concentration of Si, Mn, C, S, P in composite samples were studied with parametric and bootstrap methods, and the estimated results were compared. It shows that if the parent distribution complies with the Gaussian (or t distribution) discipline or if the composite sample contains a large number of increments, the estimated confidence intervals for two methods are in accordance with each other. However, if the parent distribution deviates greatly from the Gaussian distribution (or t distribution) and the composite sample contains a small number of increments, bootstrap method gives more reasonable confidence intervals than parametric method.