求参数置信限的一种方法

本文介绍一种找参数精确置信限和置信区间的一般方法,关键想法是在样本空间中定义一个序。本文主要考虑单参数统计模型，对序的优良性（即相应的置信限的优良性）作了讨论，指出应以得分函数的大小为依据在样本空间中定义序。还证明了用最大似然估计定义序得到的置信限是一个单调函数的唯一零点，从而通过解方程算出置信限。

A General Way for Obtaining Confidence Limits

This paper study the way for obtaining confidence limits for parameters by defining an ordering on the sample space, and we focus on statistical models with a real parameter. It is convinced that we should define an ordering on the sample space according to the score function. And it is established that if we define an ordering on the sample space according to the maximum likehood estimate, the resulted confidence limit is the unique root of a strictly increasing function, and we can compute the confidence limit by stochastic approximation.