Skip to content

Latest commit

 

History

History
59 lines (47 loc) · 2.22 KB

_ss.md

File metadata and controls

59 lines (47 loc) · 2.22 KB

title: Sufficient Statistic author: Keith A. Lewis institution: KALX, LLC email: [email protected] classoption: fleqn abstract: Sufficient for what? ...

\newcommand\RR{\bm{R}} \newcommand\CC{\bm{C}} \newcommand\FF{\bm{F}} \newcommand\ker{\operatorname{ker}} \newcommand\ran{\operatorname{ran}}

The set of bounded function on a set $X$, $B(X)$, is a Banach space with norm $|f| = \sup_{x\in X}|f(x)|$. The characteristic function of $A\subseteq X$ is $1_A$ where $1_A(x) = 1$ if $x\in A$ and $1_A(x) = 0$ if $x\not\in A$. The power set of $X$ is $\mathcal{P}(X) = {E\subseteq X}$. Let $\mathcal{O}(X) = {1_A:A\subseteq X}$.

Exercise. $B(X)$ is the norm closed algebra generated by $\mathcal{O}(X)$.

If $f\colon X\to Y$ is a function define $f^{\dashv}\colon\mathcal{P}(Y)\to\mathcal{P}(X)$ by $f^{\dashv}(B) = {x\in X:f(x)\in B}\subseteq X$, $B\subseteq Y$. We also write $f^{\dashv}1_B = 1_{f^{\dashv}(B)}$. This can be extended to a map $B(Y)\to B(X)$.

An algebra of sets on $X$ is a subset of $\mathcal{P}(X)$ closed under complement, union, and contains the empty set. If $\mathcal{A}$ is an algebra of sets on $X$ let $B(X,\mathcal{A})$ be the norm closed algebra generated by ${1_A:A\in\mathcal{A}}$.

Exercise. If $\mathcal{A}\subseteq\mathcal{P}(X)$ then the set of finite sums $\sum a_j 1{A_j}$, $a_j\in\RR$, $1_{A_j}\in\mathcal{A}$ is an algebra if and only if $\mathcal{A}$ is an algebra_.

Solution ...

Let $ba(X,\mathcal{A})$ be the collection of finitely additive set functions on $\mathcal{A}$, $α\colon\mathcal{A}\to\RR$ such that $α(A\cup B) = α(A) + α(B) - α(A\cap B)$ and $α(\emptyset) = 0$, $A,B\in\mathcal{A}$.

Lemma. The vector space dual of $B(X,\mathcal{A})$ is $ba(X,\mathcal{A})$.

Hint: Given $α^\in B(X,\mathcal{A})^$ define $α\in ba(X,\mathcal{A})$ by $α(A) = α^(1_A)$. Given $α\in ba(X,\mathcal{A})$ define $α^(1_A) = α(A)$ and use

Exercise. If $f = \sum_j a_j 1{A_j}$ is a finite sum then $f = \sum_k b_k 1_{B_k}$ where the $B_k$ are pairwise disjoint_.

A statistic is a measurable function $T\colon X\to Y$ be measurable.

When does a transformation $T\colon B(Y,\mathcal{B})\to B(X,\mathcal{A})$ come from a function $t\colon X\to Y$?