The aim of this paper is to provide a methodology and MATLAB programs to compute confidence intervals for the cell probability parameters in a high-dimensional discrete log-linear model when the maximum likelihood estimate of these parameters does not exist. To do so, we use the geometry of exponential families as well as recent methodology to identify the submodel for which the maximum likelihood estimate exists. We illustrate our results with both simulated and real world data.

中文翻译：

---

不存在 MLE 时离散对数线性模型的置信区间

本文的目的是提供一种方法和MATLAB程序来计算高维离散对数线性模型中细胞概率参数的置信区间，当这些参数的最大似然估计不存在时。为此，我们使用指数族的几何学以及最近的方法来识别存在最大似然估计的子模型。我们用模拟和现实世界的数据来说明我们的结果。