1
$\begingroup$

As we know the expectation of a RV $X$ or a function, say $g(X)$, of $X$, both with pdf $p_{X}(x)$ is

$$ \begin{array}{*{20}{c}} {X \sim {p_X}(x):}&{E[X] = \int {x.{p_X}(x)dx} }\\ {g(X) \sim {p_X}(x):}&{E[X] = \int {g(x).{p_X}(x)dx} } \end{array} $$

So my question is, can we say that When $X$ is just an argument the mean of pdf $p(x)$ is the same as the expected value of $X$, i.e. $E[X]$; but for a function of this RV ,say $g(X)$, it is not true.

Also as mentioned in this topic:

"The expectation is the average value or mean of a random variable not a probability distribution."

So in general: The expected value of a RV is not always the same as the mean of the corresponding pdf. Is my interpretation correct?

$\endgroup$
  • $\begingroup$ There is no difference between expectation of $X$ and expectation of the distribution of $X$. $\endgroup$ – StubbornAtom Jun 2 at 10:27
  • $\begingroup$ Thank you for comment. And about the expectation of a function of $X$ and expectation of the distribution of $X$? @StubbornAtom $\endgroup$ – sci9 Jun 2 at 10:29
  • 1
    $\begingroup$ If $p$ is the pdf of $X$, then mean of $X$=$\int xp(x)\,dx$. And for a function $g$, mean of $g(X)$ is $\int g(x)p(x)\,dx$, which follows from this theorem. $\endgroup$ – StubbornAtom Jun 2 at 10:37
  • $\begingroup$ @StubbornAtom Thank you for the link. So we say that when the distribution of $g(X)$, i.e. $p_{g(X)}(x)$, is known, $E(g(X))$ is the same as mean of pdf $p_{g(X)}(x)$? $\endgroup$ – sci9 Jun 2 at 10:53
  • 1
    $\begingroup$ Yes, $E(g(X))=\int x p_{g(X)}(x)\,dx$. $\endgroup$ – StubbornAtom Jun 2 at 11:01
2
$\begingroup$

The expectation of a (Borel measurable) function $g$ relative to a probability density $f(x)$ is defined to be the integral of $g(x)f(x)\mathrm{d}x$ provided the integral of $|g(x)|f(x)\mathrm{d}x$ is finite. This is always equal to the expectation of $g(X)$ whenever $X$ is a random variable with $f$ for its density.

This is a fundamental result of probability theory and so is well worth learning and understanding. Although it has frequently been quoted on these pages, I don't believe it has been rigorously stated here, nor has any sketch of its proof been shown. For those details, read on.


Let's get clear about the definitions.

A random variable $X$ associates numerical values to outcomes in a probability space $(\Omega, \mathfrak F, \mathbb P).$ Its expectation is the mathematical expression of the average of $X$ as weighted by the probability, written

$$E[X] = \int_\Omega X(\omega)\,\mathrm{d}\mathbb{P}(\omega).$$

In this expression, which is a Lebesgue integral, "$\omega$" refers to elements of the sample space $\Omega$, $X(\omega)$ is the value associated by $X$ to $\omega,$ and $\mathrm{d}\mathbb{P}(\omega)$ can be understood as its proper weight in this average.

Similarly, when $g$ is a function of the possible values of $X$ (so it assigns numbers to numbers) and $g(X)$ also is a random variable, this formula shows $g(X)$ has an expectation

$$E[g(X)] = \int_\Omega g(X(\omega))\,\mathrm{d}\mathbb{P}(\omega).$$

The (probability) distribution $F_X$ of a random variable $X$ is a probability function defined on certain "nice" sets of numbers, the Borel sets. For any number $x,$ it is determined by the rule

$$F_X(x) = \mathbb{P}\left(\left\{\omega\in\Omega\mid X(\omega)\le x\right\}\right).$$

In words: the value of the distribution function $F_X$ at the number $x$ is the chance that $X$ will not exceed $x.$

When $F_X$ has a derivative $f_X,$ the Fundamental Theorem of Calculus says $F_X$ can be recovered by integrating $f_X:$

$$F_X(x) = \int_{-\infty}^x f_X(x)\mathrm{d}x.$$

In this case, we say $X$ has a probability density function (pdf) $f_X.$ Such a density function can be considered an assignment of a non-negative number $f_X(x)$--the "probability density at $x$"--to every number $x.$ This makes it a different kind of mathematical object than $X.$ Nevertheless, the two objects enjoy a fundamental relationship.

The Law of the Unconscious Statistician assserts the expectation of any sufficiently nice (i.e., measurable) function $g$ applied to $X,$ written above as an abstract integral over $\Omega,$ can always be computed as an integral over $f_X$ when $X$ has a pdf:

LOTUS (the Law of the Unconscious Statistician): $$E[|g(X)|] = \int_{-\infty}^\infty |g(x)| f(x)\mathrm{d}x$$ and, when this quantity is finite, $$E[g(X)] = \int_{-\infty}^\infty g(x) f(x)\mathrm{d}x.$$

This is not straightforward to prove. The standard demonstration echoes the definition of the Lebesgue integral: basically, you have to start from scratch by defining the two kinds of integrals over the simplest possible functions (those that take on only the values $0$ and $1$) and gradually generalizing them to more complicated functions, checking at each step that LOTUS holds. The stages of generality of $g$ are:

  1. Indicator functions (measurable functions with values in $0$ and $1$).

  2. Finite sums of indicator functions multiplied by positive constants ("simple functions").

  3. Non-negative (Borel) measurable functions. These can be approximated by simple functions.

  4. General measurable functions. These can be expressed as differences of non-negative measurable functions.

For details, see the reference.

Reference

Steven Shreve, Stochastic Calculus for Finance II: Continuous-Time Models (Springer 2000), section 1.5.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.