How does one model independent trials in statistics.

325 Views Asked by At

In my probability class, we covered the proof of the following result, known as the "strong law of large numbers":

Theorem. Let $(\Omega,\mathscr F,P)$ be a probability space, $\{X_n\}_{n\in\mathbb N}$ be a sequence of i.i.d. random variables with finite mean $E[X_n]=\mu<\infty$, and for every $n\in\mathbb N$, let $S_n=X_1+X_2+\cdots+ X_n$. Then, $S_n/n$ converges to $\mu$ with probability $1$.

As the professor was writing the result down, he mentioned that an important application of this theorem in statistics is that it shows that the empirical mean is a consistent estimator of the theoretical mean of random variables with finite mean. Thinking about this lead me to the following question:

Question. In statistics, what is the mathematical model for a finite collection of "independent trials"? More precisely, suppose we have a random variable $X:\Omega\to\mathbb R$ with a certain distribution $F:\mathbb R\to[0,1]$, how does one generate, given $n\in\mathbb N$ a collection $x_1,...,x_n\in\mathbb R$, which can be considered independent trials of that random variable?

My attempt at an answer so far:

Attempted Answer 1. One way could be to select $n$ elements in the sample space $\omega_1,...,\omega_n\in\Omega$ and consider the collection $X(\omega_1),...,X(\omega_n)$. After all, $X(\omega)$ is seen as a "realization" of the random variable $X$, and so we would obtain $n$ realizations. However, it is unclear how we could make these realizations independent.

Attempted Answer 2. Another idea would be to define a collection of i.i.d. random variables $X_1,...,X_n$ and then evaluate them at an arbitrary $\omega\in\Omega$, which would give the collection of trials $X_1(\omega),...,X_n(\omega)$. In this case, the independence is not a problem, but the random variables $X_i$ themselves seem cryptic: where do they come from? How do we define them from $X$?

1

There are 1 best solutions below

0
On BEST ANSWER

You need to use one big probability space. For $n$ i.i.d random variables, that probability space has the sample space $\mathbb{R}^n$ (assuming that random variables are real-valued), and the measure is simply the product measure $\underbrace{\mathbb{P}\times\ldots\times\mathbb{P}}_{n\text{ times}}$. $\mathbb{P}$ is the probability measure corresponding to a single random variable.

The same works for independent, but not necessarily identical distributed random variables, you instead use the product measure $\mathbb{P}_1\times\ldots\times\mathbb{P}_n$, where $\mathbb{P}_i$ is the probability measure corresponding to the $i$-th random variable.