Understanding of the definition "statistical experiment" (Blackwell, 1951)

116 Views Asked by At

I'm interesting for the understanding of the definition "statistical experiment". Formally statistical experiment, statistical model or just experiment was defined as a triple $$\mathscr{P}=(\Omega,\mathscr{F}, \{ P_{\theta}:\theta \in \Theta \}),$$

where $\Omega,\mathscr{F}$ is a sample space, $\Theta$ is a parameter spase and $\{ P_{\theta}:\theta \in \Theta \}$ is family of probability measures, defined on the given sample space.

My question is: why it is a family of probability measures? What is/are an example(s) of this abstaction?

Especially we could interesting for the parameters estimation. Let me give an example as the (real-valued) mean estimation problem of the random variables, that are normally distributed with known variance equals one. More precise: we have some i.i.d. r.v.'s $X_1,X_2,...,X_n$ (given observations) with $X_1 \sim \mathscr{N}(\theta, 1)$. Then we could formulate related statistical experiment as $$\mathscr{P}=(\Bbb{R}^n,\mathscr{B}(\Bbb{R}^n), \{ P_{\theta}^{\times{n}}:\theta \in \Theta=\Bbb{R} \}).$$

So, we define our statistical experiment as the infinite set of possible distributions (i.e. as a set $\{\mathscr{N}(\theta, 1)^{\times{n}}:\theta \in \Bbb{R}\}$). Why do we need a set? Why we can not say, that we are looking for some estimation of the "true" distribution, which is unique?