位置—尺度分布族基于逐次定数截尾样本的拟合优度检验

发布时间：2018-06-12 21:41

本文选题：拟合优度检验 + 逐次定数截尾样本　；参考：《浙江工商大学》2012年硕士论文

【摘要】：拟合优度检验是统计基础的重要组成部分,而且和实际应用密切相关。Pakyari等(2011)讨论了基于逐次定数截尾样本位置-尺度分布族的拟合检验,给出了数个检验统计量。Monte Carlo模拟显示,在一些早截尾情况下,给出的统计量才优于Balakrishnan等(2004)给出的统计量T 基于变换和加权的思想,对Pakyari等(2011)给出的两个统计量Tm:n(1)和Tm:n(2)进行了修正,并用Monte Carlo模拟研究了修正统计量的功效。结果表明：在简单零假设下,修正统计量的功效在多数情况下高于Tm:n(1)和Tm:n(2)的功效。特别的,对于早期截尾的方案,所有修正统计量的表现都好于Tm:n(1)和Tm:n(2)；当观察到的失效数m较大时,统计量WMTm:n(1)WMTm:n(2)在多种截尾情况下好于Tm:n(1)和Tm:n(2)。当检验分布的参数未知时,修正也是有效的。此外,本文对位置-尺度分布族基于逐次定数截尾样本的拟合检验,给出一个新的检验统计量W,并且导出了简单零假设下W的精确分布。Monte Carlo模拟表明,在简单零假设下,W的功效在多数情况下高于T。特别的,当检验正态分布时,W的功效总是高于T；当检验Gumbel分布时,W的功效在晚截尾情况下总是大于T。在复合零假设下,检验正态分布时,W在晚截尾情况下表现也好于T
[Abstract]:Goodness of fit test is an important part of statistical foundation, and it is closely related to practical application. Pakyari et al. (2011) discusses the fitting test based on successive censored sample location-scale distribution family, and gives several test statistics. Monte Carlo simulation shows that, In some early truncation cases, the statistics given are better than those given by Balakrishnan et al. 2004. based on the idea of transformation and weighting, the two statistics given by Pakyari et al. The effect of modified statistics is studied by Monte Carlo simulation. The results show that under the simple zero hypothesis, the effect of modifying statistics is in most cases higher than that of TM: nu 1) and TM: nu 2). In particular, for the early truncation schemes, the performance of all modified statistics is better than that of Tm: n-1) and Tm: nu 2s; when the observed failure number m is larger, the statistic WMTm: n 1 is better than that of Tm: n1) and Tm: nn1), when the observed failure number m is larger, the statistics are better than Tm: n1) and Tm: nn1). The correction is also effective when the parameters of the test distribution are unknown. In addition, a new test statistic W is given for the location-scale distribution family based on the fitting test of successive censored samples, and the exact distribution of W under the simple zero hypothesis is derived by the Monte Carlo simulation. Under the simple zero hypothesis, the efficacy of W is higher than that of TW in most cases. In particular, the efficacy of W is always higher than that of Twhen the normal distribution is tested, and the efficacy of W is always greater than that of Twhen the Gumbel distribution is examined. Under the compound zero hypothesis, when the normal distribution is tested, the performance of W is better than that of T in the case of late truncation.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：C81

【相似文献】