本页提供题录、可核实摘要与阅读线索。AI 解读未完成全文核验，可能有误；具体假设、证明和结论请核对原文。仅依据书目生成的内容属于阅读导航，不是对论文结论的确认。
摘要保留来源文本，未经模型改写。材料不足时保留待解读状态；引用采用 OpenAlex 单条关联记录，不合计版本，缺失不视为零，也不代表论文质量。
正在核对对应版本的资源…
The problems of existence, uniqueness and location of maximum likelihood estimates in log linear models have received special attention in the literature (Haberman, 1974, Chapter 2; Wedderburn, 1976; Silvapulle, 1981). For multinomial logistic regression models, we prove existence theorems by considering the possible patterns of data points, which fall into three mutually exclusive and exhaustive categories: complete separation, quasicomplete separation and overlap. Our results suggest general rules for identifying infinite parameter estimates in log linear models for frequency tables.