统计结构
设\((\mathscr{X},\mathscr{B})\)为可测空间,\(\mathscr{P}\)为其上的一个概率分布族,则称三元组\((\mathscr{X},\mathscr{B},\mathscr{P})\)为统计结构,或称为统计模型。假如分布族\(\mathscr{P}\)仅依赖于某个参数(或者参数向量)\(\theta\),即\(\mathscr{P}=\{P_\theta:\theta\in \Theta\}\),其中\(\Theta\)为参数空间,则称此结构为参数(统计)结构,或成为参数模型;否则被称为非参数模型。
由简单统计结构可以派生出乘积结构。设\((\mathscr{X},\mathscr{B},\mathscr{P})\)和\((\mathscr{X}',\mathscr{B}',\mathscr{P}')\)为两个统计结构,则称\((\mathscr{X}\times \mathscr{X}',\mathscr{B}\otimes \mathscr{B}',\mathscr{P}\otimes \mathscr{P}')\)为二者的乘积结构,并记为\((\mathscr{X},\mathscr{B},\mathscr{P})\otimes (\mathscr{X}',\mathscr{B}',\mathscr{P}')\),其中 \[
\mathscr{P}\otimes \mathscr{P}'=\{P\otimes P':P\in \mathscr{P},P'\in \mathscr{P}'\}
\] 特别地,\(n\)个相同统计结构乘积结构被称为重复抽样结构,记为\((\mathscr{X},\mathscr{B},\mathscr{P})^n\)。