Coverage of confidence intervals based on conditional probability
Coverage of confidence intervals based on conditional probability
复制标题
基于条件概率的置信区间覆盖
DOI：
10.1088/1126-6708/2000/11/036
复制
发表时间：
2000
影响因子：
5.4
通讯作者：
J. Schultz
中科院分区：
文献类型：
--
作者：
M. Mandelkern;J. Schultz
我们认为Roe-Woodroofe建设的泊松分布变量的情况下，置信区间的平均值是一个已知的背景和一个未知的非负信号的总和。RW构造基于条件pdf，并且具有条件覆盖，使得间隔仅对于重复实验的全集的子集包含具有规定概率的信号的真值。这些区间没有通常意义上的覆盖，但可以通过修改来使其具有这种覆盖，而这种修改不会影响这种有吸引力的构造的可信度和其他可取的特征。一个类似的修改可以用来提供覆盖最近讨论的应用这种方法的高斯边界问题。在这两种情况下的结果是等价于一个结构，其中定义的轮廓是baidian上限（均匀先验）的观察附近的物理界和baidian（或频率）的中心置信区间渐近。
We consider the Roe-Woodroofe construction of confidence intervals for the case of a Poisson distributed variate, where the mean is the sum of a known background and an unknown non-negative signal. The RW construction is based on a conditional pdf and posseses conditional coverage such that the intervals contain the true value of the signal with prescribed probability only for subsets of the full set of repeated experiments. The intervals do not have coverage in the usual sense but can be made to have such with a modification that does not affect the believability and other desirable features of this attractive construction. A similar modification can be used to provide coverage to a recently discussed application of this method to the gaussian-with-boundary problem. In both cases the result is equivalent to a construction in which the defining contour is the bayesian upper limit (uniform prior) for observations near the physical bound and the bayesian (or frequentist) central confidence interval asymptotically.