I am asked to find the expected value of a vector of two random variables when the joint density is given. Is the recipe for solving this problem:

  • Find the marginal distributions
  • Find the expected values of the marginal distributions

which will involve a relatively long integration process, since I have to find the two marginals and then the two expected values?

Or is there some shortcut?

  • 1
    $\begingroup$ Is this a question from a course or textbook? If so, please add the [self-study] tag & read its wiki. $\endgroup$ Dec 8, 2015 at 17:12
  • $\begingroup$ @Xi'an What are the double integrals I should solve? Something like $\int\int xyf(x,y)dxdy$? $\endgroup$
    – vec
    Dec 8, 2015 at 17:15
  • 1
    $\begingroup$ @gung yes; tag added $\endgroup$
    – vec
    Dec 8, 2015 at 17:16
  • 5
    $\begingroup$ The expected value of a vector $(X,Y)$ random variable is defined to be $E(X,Y])= (E[X], E[Y])$, that is, the vector of the individual expectations. So you don't really have a "shortcut" except that you can hide what you are doing in a double integral. $$E[X] = \int_{-\infty}^\infty x\cdot f_X(x)\,\mathrm dx = \int_{-\infty}^\infty x \left[\int_{-\infty}^\infty f_{X,Y}(x,y)\,\mathrm dy\right]\,\mathrm dx = \int_{-\infty}^\infty \int_{-\infty}^\infty x\cdot f_{X,Y}(x,y)\,\mathrm dy\,\mathrm dx,$$ and similarly for $Y$. Is solving the rightmost integral the same as first finding the marginal? $\endgroup$ Dec 8, 2015 at 18:08

2 Answers 2


As rightly pointed out by Dilip Sarwate, computing the expectation of one component as a two dimensional integral requires integrating out the other element of the vector: $$\mathbb{E}[X] = \iint x\,f_{X,Y}(x,y)\,\mathrm dy\,\mathrm dx,$$ (which is a special case of the so-called law of the uncounscious statistician). The only simplifications I can think of is

  1. when finding the conditional expectation of one component given the other is easier: $$\mathbb{E}[X] = \mathbb{E}[\mathbb{E}[X|Y]] = \int x\,f_{X|Y}(x|y)\,\mathrm dx\,f_Y(y)\,\mathrm dy,$$ by the double projection theorem, in the sense that this could require computing a single marginal instead of two;
  2. when computing the marginal cdf is easier, since $$\mathbb{E}[X] = \int_{-\infty}^0 F_X(x)\,\mathrm dx-\int_{+\infty}^0 (1-F_X(x))\,\mathrm dx$$ by an integration by parts.
  • $\begingroup$ In #1, about the conditional, are there supposed to be two $\mathrm dx$ in the expression? Forgive my ignorance, I'm not sure how to understand the expression as it is right now given there's only one integral symbol and two $\mathrm dx$. The link to the reference is broken, and Archive.org's Wayback Machine doesn't show the LaTeX equations. $\endgroup$ Aug 25, 2021 at 23:18
  • $\begingroup$ @AhmedFasih: typo corrected $\endgroup$
    – Xi'an
    Aug 26, 2021 at 5:22
  • $\begingroup$ FYI, 1. is called "The Law of Iterated Expectation" or "Adam's law". $\endgroup$ Aug 19, 2023 at 15:24

Using Monte Carlo estimate, the expected value for the marginal distribution will the $\frac{\displaystyle\sum_{i=1}^Nx_i}{N}$ where $x_i$ are x's samples from the (x,y) samples from joint distribution.

By Monte Carlo integration, $\int\int\phi(x,y)f(x,y)dxdy$ can be estimated by $\frac{1}{N}\displaystyle\sum_{i=1}^N\phi(x_i,y_i)$. Here, we have $\phi(x,y)= x$.

  • 2
    $\begingroup$ Please register &/or merge your accounts (you can find information on how to do this in the My Account section of our help center), then you will be able to edit & comment on your own question. $\endgroup$ Dec 5, 2017 at 0:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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